BR122022025392B1 - GENOMIC ANALYSIS PLATFORM TO PERFORM A SEQUENCE ANALYSIS SEGMENTATION - Google Patents

Abstract

A system, method and apparatus for performing a bioinformatics analysis on genetic sequence data is provided. In particular, a genomic analysis platform is provided for performing a sequence analysis segmentation. The genomic analysis platform includes one or more than one first integrated circuit, wherein each first integrated circuit forms a central processing unit (CPU) that is responsive to one or more software algorithms that are configured to instruct the CPU to perform a first set of genomic processing steps of the sequence analysis segmentation. Additionally, a second integrated circuit is also provided, wherein each second integrated circuit forms a field programmable gate array (FPGA), wherein the FPGA is configured by firmware to arrange a set of hardwired digital logic circuits that are interconnected by a plurality of physical interconnects for performing a second set of genomic processing steps of the sequence analysis segmentation, wherein the hardwired digital logic circuits of each FPGA are arranged as a set of processing engines for performing the second set of genomic processing steps. A shared memory is also provided.

Description

CROSS-REFERENCE TO RELATED REQUEST

[001] This application claims priority to U.S. Application No. 62/347,080, filed June 7, 2016, U.S. Application No. 62/399,582, filed September 26, 2016, U.S. Application No. 62/414,637, filed October 28, 2016, U.S. Application No. 15/404,146, filed January 11, 2017, U.S. Application No. 62/462,869, filed February 23, 2017, U.S. Application No. 62/469,442, filed March 9, 2017, and U.S. Application No. 15/497,149, filed April 25, 2017, whose disclosures of each order are incorporated herein by reference in their entirety.

FIELD OF REVELATION

[002] The subject matter described in this document relates to bioinformatics and, more particularly, to systems, apparatus and methods for implementing bioinformatics protocols, such as performing one or more functions for analyzing genomic data on an integrated circuit, such as on a hardware processing platform.

BACKGROUND TO THE REVELATION

[003] As described in detail herein, some key computational challenges for high-throughput DNA sequencing analysis are addressing the explosive growth in available genomic data, the need for increased accuracy and sensitivity when assembling such data, and the need for fast, efficient, and accurate computational tools when performing analysis on a wide range of sequencing datasets derived from such genomic data.

[004] The accompaniment of such increased sequencing throughput generated by Next Generation Sequencers has typically been manifest as multi-threaded software tools that have been run on ever-increasing numbers of faster processors in clusters of computers with expensive high-availability storage that requires substantial power and significant IT support costs. Importantly, future increases in sequencing throughput rates will translate into accelerating real dollar costs for these secondary processing solutions.

[005] The devices, systems, and methods of their use described herein are provided, at least in part, to address these and other such challenges.

SUMMARY OF THE REVELATION

[006] The present disclosure is directed to devices, systems, and methods for employing the same in performing one or more genomics and/or bioinformatics protocols on data generated through a primary processing procedure, such as on genetic sequence data. For example, in various aspects, the devices, systems, and methods provided herein are configured to perform secondary and/or tertiary analysis protocols on genetic data, such as data generated by RNA and/or DNA sequencing, e.g., by a Next Generation Sequencer ("NGS"). In particular embodiments, one or more secondary processing pipelines for processing genetic sequence data are provided. In other embodiments, one or more tertiary processing pipelines for processing genetic sequence data are provided, such as where the pipelines, and/or individual elements thereof, deliver superior sensitivity and improved accuracy over a broader range of sequence-derived data than is currently available in the art.

[007] For example, provided herein is a system, such as for performing one or more of a sequence and/or genomic analysis segments on genetic sequence data and/or other data derived therefrom. In various embodiments, the system may include one or more of an electronic data source that provides digital signals representing a plurality of genetic and/or genomic data reads, such as when each of the plurality of genomic data reads includes a nucleotide sequence. The system may further include, for example, a DRAM, or a cache, such as for storing one or more of the sequenced reads, one or a plurality of reference genetic sequences, and one or more indices of the one or more reference genetic sequences. The system may further include one or more integrated circuits, such as an FPGA, ASIC or sASIC, and/or a CPU and/or a GPU, wherein the integrated circuit, e.g., with respect to the FPGA, ASIC or sASIC may be formed by a set of hard-wired digital logic circuits that are interconnected by a plurality of physical electrical interconnects. The system may further include a quantum computing processing unit, for use in implementing one or more of the methods disclosed herein.

[008] In various embodiments, one or more of the plurality of electrical interconnects may include an input to the one or more integrated circuits that may be connected or connectable, e.g., directly, via a wired connection, or indirectly, such as via a wireless network connection (e.g., a cloud or hybrid cloud), to the electronic data source. Regardless of a connection to the sequencer, an integrated circuit of the disclosure may be configured to receive the plurality of genomic data reads, e.g., directly from the sequencer or from an associated memory. The reads may be digitally encoded in a standard FASTQ or BCL file format. Accordingly, the system may include an integrated circuit having one or more electrical interconnects that may be a physical interconnect that includes a memory interface so as to allow the integrated circuit to access memory.

[009] In particular, the hardwired digital logic circuitry of the integrated circuit may be arranged as a set of processing engines, such as wherein each processing engine may be formed by a subset of the hardwired digital logic circuitry for performing one or more steps in the sequence, genomic, and/or tertiary analysis segmentation, as described herein below, on the plurality of genetic data reads, as well as other data derived therefrom. For example, each subset of the hardwired digital logic circuitry may be in a hardwired configuration for performing one or more steps in the analysis segmentation. Additionally, when the integrated circuit is an FPGA, such steps in the sequence and/or further analysis process may involve partial reconfiguration of the FPGA during the analysis process.

[010] Particularly, the set of processing engines may include a mapping module, e.g., in a wired configuration for accessing, according to at least part of the nucleotide sequence in a read of the plurality of reads, the index of the one or more genetic reference sequences from memory via the memory interface in order to map the read to one or more segments of the one or more genetic reference sequences based on the index. Additionally, the set of processing engines may include an alignment module in the wired configuration for accessing the one or more genetic reference sequences from memory via the memory interface to align the read, e.g., the mapped read, to one or more positions in the one or more segments of the one or more genetic reference sequences, e.g., as received from the mapping module and/or stored in memory.

[011] Additionally, the set of processing engines may include a classification module for classifying each aligned read according to one or more positions in the one or more genetic reference sequences. Furthermore, the set of processing engines may include a variant calling module, such as for processing the mapped, aligned, and/or classified reads, such as against a reference genome, to thereby produce a read HMM and/or variant calling file for use with and/or detailing variations between the sequenced genetic data and the reference genomic data. In various instances, one or more of the plurality of physical electrical interconnects may include an output of the integrated circuit for communicating output data from the mapping module and/or the alignment and/or classification and/or variant calling modules.

[012] Particularly, with respect to the mapping module, in various embodiments, a system is provided for performing a mapping analysis segmentation on a plurality of genetic data reads using a reference genetic data index. In various cases, the genetic sequence, e.g., read, and/or the reference genetic data may be represented by a nucleotide sequence, which may be stored in a system memory. The mapping module may be included within the integrated circuit and may be formed by a set of preconfigured and/or wired digital logic circuits that are interconnected by a plurality of physical electrical interconnects, wherein the physical electrical interconnects may include a memory interface to allow the integrated circuit to access the memory. In more particular embodiments, the hardwired digital logic circuits may be arranged as a set of processing engines, such as wherein each processing engine is formed by a subset of the hardwired digital logic circuits to perform one or more steps in segmenting sequence analysis on the plurality of genomic data reads.

[013] For example, in one embodiment, the set of processing engines may include a mapping module in a wired configuration, wherein the mapping module and/or one or more processing engines thereof are configured to receive a read of genomic data, such as via one or more of a plurality of physical electrical interconnections, and to extract a portion of the read in such a manner as to generate a seed therefrom. In such a case, the read may be represented by a sequence of nucleotides, and a seed may represent a subset of nucleotides represented by the read. The mapping module may include or be connectable to a memory that includes one or more of the reads, one or more of the seeds of the reads, at least a portion of one or more of the reference genomes, and/or one or more indexes, such as an index constructed from the one or more reference genomes. In certain cases, a processing engine of the mapping module employs the seed and the index to calculate an address within the index based on the seed.

[014] Once an address has been calculated or derived and/or otherwise stored, such as in an integrated or non-integrated memory, the address may be accessed in the index in the memory so as to receive a record from the address, such as a record representing position information in the reference genetic sequence. The position information may then be used to determine one or more corresponding positions from the read to the reference genetic sequence based on the record. Then, at least one of the corresponding positions may be output to memory via the memory interface.

[015] In another embodiment, a set of the processing engines may include an alignment module, such as in a pre-configured and/or wired configuration. In this case, one or more of the processing engines may be configured to receive one or more of the mapped positions for the read data via one or more of the plurality of physical electrical interconnects. Then, memory (internal or external) may be accessed for each mapped position to retrieve a segment of the reference sequence/genome that corresponds to the mapped position. An alignment of the read to each retrieved reference segment may be calculated along with a score for the alignment. Once calculated, at least one best-scoring alignment of the read may be selected and output. In various cases, the alignment module may also implement a dynamic programming algorithm when calculating the alignment, such as one or more of a Smith-Waterman algorithm, e.g., with linear or affine gap scoring, a gapped alignment algorithm, and/or a gapless alignment algorithm. In particular cases, the alignment calculation may include first performing a gapless alignment for each reference segment and, based on the results of the gapless alignment, selecting reference segments with which to additionally perform gapped alignments.

[016] In various embodiments, a variant calling module may be provided to perform variant calling functions that, when implemented in one or both of the software and/or hardware configurations, yield superior processing speed, improved accuracy of processed output, and increased overall efficiency relative to methods, devices, and systems currently known in the art. Specifically, in one aspect, improved methods for performing variant calling operations in software and/or in hardware are provided, such as for performing one or more HMM operations on genetic sequence data. In another aspect, novel devices including an integrated circuit for performing such variant calling operations are provided, wherein at least a portion of the variant calling operation is implemented in hardware.

[017] Accordingly, in various instances, the methods disclosed herein include mapping, by a first subset of hardwired and/or quantum digital logic circuits, a plurality of reads to one or more segments of one or more genetic reference sequences. Additionally, the methods may include accessing, via the hardwired and/or quantum digital logic circuits, e.g., via one or more of the plurality of physical electrical interconnects from memory or a cache associated therewith, one or more of the mapped reads and/or one or more of the genetic reference sequences and aligning, via a second subset of the hardwired and/or quantum digital logic circuits, the plurality of mapped reads to the one or more segments of the one or more genetic reference sequences.

[018] In various embodiments, the method may further include accessing, via the integrated and/or quantum circuit, e.g., via one or more of the plurality of physical electrical interconnects from a memory or a cache associated therewith, the aligned plurality of reads. In such a case, the method may include sorting, via a third subset of the hardwired and/or quantum digital logic circuits, the aligned plurality of reads according to their positions in the one or more genetic reference sequences. In certain cases, the method may further include outputting, such as via one or more of the plurality of physical electrical interconnects of the integrated and/or quantum circuit, result data of the mapping and/or alignment and/or sorting, such as when the result data includes positions of the plurality of mapped and/or aligned and/or sorted reads.

[019] In some cases, the method may further include using the result data obtained, such as by a further subset of the hardwired and/or quantum digital logic circuits, for the purpose of determining how the mapped, aligned, and/or classified data derived from the sequenced genetic sample of the subject differs from a reference sequence, so as to produce a variant call file that delineates the genetic differences between the two samples. Accordingly, in various embodiments, the method may further include accessing, via the hardwired and/or quantum circuit, e.g., via one or more of the plurality of physical electrical interconnects from a memory or a cache associated therewith, the mapped and/or aligned and/or classified plurality of reads. In such a case, the method may include performing a variant calling function, e.g., an HMM or pairwise HMM operation, on the reads accessed by a third or fourth subset of the hardwired and/or quantum digital logic circuits so as to produce a variant calling file that details how the mapped, aligned, and/or classified reads vary relative to that of one or more reference sequences, e.g., haplotype.

[020] Accordingly, in accordance with particular aspects of the disclosure, there is disclosed herein a compact, e.g., chip-based, hardware or quantum accelerated platform for performing secondary and/or tertiary analysis on genetic and/or genomic sequencing data. In particular, there is provided a platform or array of wired and/or quantum digital logic circuits that has been specifically designed to perform secondary and/or tertiary analysis, such as on sequenced genetic data or genomic data derived therefrom. In particular, a set of digital hardwired and/or quantum logic circuits, which may be arranged as a set of processing engines, may be provided, such as where the processing engines may be present in a pre-configured configuration and/or hardwired and/or quantum configuration on a processing platform of the disclosure, and may be specifically designed to perform secondary mapping and/or alignment and/or variant calling operations related to genetic analysis on DNA and/or RNA data, and/or may be specifically designed to perform other tertiary processing on the result data.

[021] In particular cases, the present devices, systems and methods for employing the same in the performance of one or more genomics and/or bioinformatics secondary and/or tertiary processing protocols have been optimized so as to deliver an improvement in processing speed, i.e., orders of magnitude faster than standard secondary processing segmentations that are implemented in software. Additionally, the segmentations and/or components thereof as presented herein provide improved sensitivity and accuracy across a wide range of sequence-derived data sets for genomics and bioinformatics processing purposes. In various cases, one or more of these operations may be performed by an integrated circuit that is part of or configured as a general purpose central processing unit and/or a graphics processing unit and/or a quantum processing unit.

[022] For example, genomics and bioinformatics are fields that are concerned with the application of information technology and computer science to the field of genetics and/or molecular biology. In particular, bioinformatics techniques can be applied to process and analyze various genetic and/or genomic data, such as from an individual, in order to determine qualitative and quantitative information about that data that can then be used by various practitioners in the development of prophylactic, therapeutic, and/or diagnostic methods to prevent, treat, ameliorate, and/or at least identify disease states and/or their potential, and thereby improve the safety, quality, and effectiveness of health care on an individualized level. Therefore, because of their focus on advancing personalized health care, the fields of genomics and bioinformatics promote individualized health care that is proactive rather than reactive, and that gives the individual in need of treatment the opportunity to become more involved in his or her own well-being. An advantage of employing genetics, genomics and/or bioinformatics technologies disclosed herein is that qualitative and/or quantitative analyses of molecular biological, e.g. genetic, data can be performed on a broader range of sample sets at a much greater rate of speed and often more accurately, thereby expediting the emergence of a personalized healthcare system. Particularly, in various embodiments, the genomics and/or bioinformatics related tasks may form a genomics segmentation that includes one or more of a microarray analysis segmentation, a genome, e.g., whole genome analysis segmentation, genotyping analysis segmentation, exome analysis segmentation, epigenome analysis segmentation, metagenome analysis segmentation, microbiome analysis segmentation, genotyping analysis segmentation including joint genotyping, variant analysis segmentations including structural variants, somatic variants, and GATK, as well as RNA sequencing and other genetic analysis segmentations.

[023] Accordingly, to make use of these advantages, there are enhanced and more accurate software implementations to perform one or a series of such bioinformatics-based analytical techniques, such as for allocation by a general purpose CPU and/or GPU and/or may be implemented on one or more quantum circuits of a quantum processing platform. However, common characteristics of such methods and typically configured software-based bioinformatics systems are that they are labor intensive, time consuming to execute on such general purpose processors, and error prone. Therefore, bioinformatics systems, as implemented herein, that could execute these algorithms, as implemented in software by a quantum processing unit CPU and/or GPU in a less labor intensive manner and/or processing with a higher percentage accuracy could be implemented.

[024] Such implementations have been developed and are presented herein, such as when genomics and/or bioinformatics analyses are performed by optimized software running on a CPU and/or GPU and/or quantum computer in a system that makes use of the genetic sequence data derived by the processing units and/or integrated circuits of the developer. Furthermore, it should be realized that the cost of analyzing, storing, and sharing such raw digital data has far exceeded the cost of producing it. Accordingly, "just-in-time" storage and/or retrieval methods that optimize the storage of such data in a manner that trades off the speed of data regeneration in exchange for the cost of storing such data collectively are also presented herein. Thus, the "just-in-time" or "JIT" data generation, analysis, and storage methods presented herein address a key bottleneck that is a long-perceived but unaddressed obstacle that arises between the generation and storage of ever-increasing raw data and the actual medical insight sought therefrom.

[025] Therefore, disclosed herein are systems, apparatus, and methods for implementing genomics and/or bioinformatics protocols or portions thereof, such as for performing one or more functions for analyzing genomic data, e.g., in one or both of an integrated circuit, such as in a hardware processing platform, and a general purpose processor, such as for performing one or more bioanalytical operations in software and/or firmware. For example, as disclosed below herein, in various implementations, an integrated circuit and/or quantum circuit is provided to accelerate one or more processes in a primary, secondary, and/or tertiary processing platform. In various cases, the integrated circuit may be employed to perform genetics-related analytical tasks, such as mapping, alignment, variant calling, compression, decompression, and the like, in an accelerated manner, and as such the integrated circuit may include a hardware-accelerated configuration. Additionally, in various cases, an integrated and/or quantum circuit may be provided such as when the circuit is part of a processing unit that is configured to perform one or more genomics and/or bioinformatics protocols on the generated mapped and/or aligned and/or variant-called data.

[026] Particularly, in a first embodiment, a first integrated circuit may be formed by an FPGA, ASIC, and/or sASIC that is coupled to or otherwise secured to the motherboard and configured, or, in the case of an FPGA, may be firmware programmable to be configured, as a set of hard-wired digital logic circuits that are adapted to perform at least a first set of sequence analysis functions in a genomic analysis segment, such as when the integrated circuit is configured as described above herein to include one or more digital logic circuits that are arranged as a set of processing engines, which are adapted to perform one or more steps in a mapping, alignment, and/or variant calling operation on the genetic data so as to produce sequence analysis result data. The first integrated circuit may further include an output, e.g., formed by a plurality of physical electrical interconnections, such as for communicating the result data of the mapping and/or alignment and/or other procedures to memory.

[027] Additionally, a second integrated and/or quantum circuit may be included, coupled, or otherwise secured to the motherboard and in communication with the memory via a communications interface. The second integrated and/or quantum circuit may be formed as a central processing unit (CPU) or a graphics processing unit (GPU) or a quantum processing unit (QPU) that is configured to receive the mapped and/or aligned and/or variant-called sequence analysis result data and may be adapted to be responsive to one or more software algorithms that are configured to instruct the CPU or GPU to perform one or more genomic and/or bioinformatics functions of genomic analysis segmentation on the mapped, aligned, and/or variant-called sequence analysis result data. Specifically, the genomics and/or bioinformatics related tasks may form a genomics analysis segmentation that includes one or more of a microarray analysis, a genome segmentation, e.g., whole genome analysis segmentation, genotyping analysis segmentation, exome analysis segmentation, epigenome analysis segmentation, metagenome analysis segmentation, microbiome analysis segmentation, genotyping analysis segmentations including joint genotyping, variant analysis segmentations including structural variants, somatic variants, and GATK, as well as RNA sequencing analysis segmentation, and other genetic analysis segmentations.

[028] For example, in one embodiment, the CPU and/or GPU and/or QPU of the second integrated circuit may include software that is configured to arrange the genome analysis segmentation to perform a whole genome analysis segmentation, such as a whole genome analysis segmentation that includes one or more of whole genome variation analysis, whole exome DNA analysis, whole transcriptome RNA analysis, gene function analysis, protein function analysis, protein binding analysis, quantitative gene analysis, and/or a gene assembly analysis. In various instances, the whole genome analysis segmentation may be performed for the purposes of one or more of ancestry analysis, personal medical history analysis, disease diagnosis, drug discovery, and/or protein profiling. In a particular instance, the whole genome analysis segmentation is performed for the purposes of oncology analysis. In various instances, the results of this data may be available, for example, globally, throughout the system.

[029] In various cases, the CPU and/or GPU and/or a quantum processing unit (QPU) of the second integrated circuit and/or quantum may include software that is configured to arrange the genome analysis segmentation to perform a genotyping analysis, such as a genotyping analysis that includes joint genotyping. For example, the joint genotyping analysis may be performed using a Bayesian probability calculation, such as a Bayesian probability calculation that results in an absolute probability that a given genotype is a true genotype. In other cases, the software may be configured to perform a metagenome analysis so as to produce metagenome output data that may in turn be used in performing a microbiome analysis.

[030] In certain cases, the first and/or second integrated circuit and/or memory may be housed on an expansion card, such as a peripheral component interconnect (PCI) card. For example, in various embodiments, one or more of the integrated circuits may be one or more chips coupled to a PCIe card or otherwise associated with the motherboard. In various cases, the integrated and/or quantum circuit(s) and/or chip(s) may be a component within a sequencer or computer, or server, such as part of a server farm. In particular embodiments, the integrated and/or quantum circuit(s) and/or expansion card(s) and/or computer(s) and/or server(s) may be accessible via the internet, e.g., the cloud.

[031] Additionally, in some cases, the memory may be a volatile random access memory (RAM), e.g., a direct access memory (DRAM). Particularly, in various embodiments, the memory may include at least two memories, such as a first memory that is an HMEM, e.g., for storing reference haplotype sequence data, and a second memory that is an RMEM, e.g., for storing readout of genomic sequence data. In particular cases, each of the two memories may include a register port and/or a readout port, such as when the register port and the readout port each access a separate clock. Additionally, each of the two memories may include a flipflop configuration for storing a plurality of genetic sequence and/or processing result data.

[032] Accordingly, in another aspect, the system may be configured to share memory resources among its component parts, such as in relation to performing some computational tasks via software, such as executed by the CPU and/or GPU and/or quantum processing platform, and/or performing other computational tasks via firmware, such as via the hardware of an associated integrated circuit, e.g., FPGA, ASIC, and/or sASIC. This may be achieved in different ways, such as by a loose or tight direct coupling between the CPU/GPU/QPU and the FPGA, e.g., PCIe chip or card. Such configurations may be particularly useful when distributing operations related to the processing of large data structures associated with genomics and/or bioinformatics analyses to be used and accessed by both the CPU/GPU/QPU and the associated integrated circuit. Particularly, in various embodiments, when processing data through a genomics segmentation as described herein, such as to speed up the overall processing function, timing, and efficiency, a number of different operations may be performed on the data, where the operations may involve both software and hardware processing components.

[033] Accordingly, data may need to be shared and/or otherwise communicated between the software component(s) executing on the CPU and/or GPU and/or QPU and/or the on-chip hardware component, e.g., an FPGA. Accordingly, one or more of the steps in the genomics and/or bioinformatics processing pipeline, or a portion thereof, may be performed by one device, e.g., the CPU/GPU/QPU, and one or more of the multiple steps may be performed by a wired device, e.g., the FPGA. In such a case, the CPU/GPU/QPU and/or an FPGA may be communicatively coupled in such a manner as to permit the efficient transmission of such data, wherein the coupling may involve the shared use of memory resources. To achieve such distribution of tasks and sharing of information for the performance of such tasks, the multiple CPUs/GPUs/QPUs may be loosely or tightly coupled to each other and/or to hardware devices, e.g., FPGA, or other chipset, such as by a fast path interconnect.

[034] Particularly, in various embodiments, a genomics analysis platform is provided. For example, the platform may include a motherboard, a memory, and a plurality of integrated and/or quantum circuits, such as forming one or more of a CPU/GPU/QPU, a mapping module, an alignment module, a classification module, and/or a variant calling module. Specifically, in particular embodiments, the platform may include a first integrated and/or quantum circuit, such as an integrated circuit forming a central processing unit (CPU) or graphics processing unit (GPU) or a quantum circuit forming a quantum processor, that is responsive to one or more software or other algorithms that are configured to instruct the CPU/GPU/QPU to perform one or more sets of genomics analysis functions as described herein, such as when the CPU/GPU/QPU includes a first set of physical electronic interconnects for connection to the motherboard. In various cases, the memory may also be attached to the motherboard and may be further electronically connected to the CPU/GPU/QPU, such as through at least a portion of the first set of physical electronic interconnects. In such cases, the memory may be configured to store a plurality of genomic data reads and/or at least one or more reference gene sequences and/or an index of the one or more reference gene sequences.

[035] Additionally, the platform may include one or more other integrated circuits, such as when each of the other integrated circuits forms a field programmable gate array (FPGA) that has a second set of physical electronic interconnects for connection to the CPU/GPU/QPU and memory, such as via a point-to-point interconnect protocol. In such a case, such as when the integrated circuit is an FPGA, the FPGA may be firmware programmable to configure a set of hard-wired digital logic circuits that are interconnected by a plurality of physical interconnects to perform a second set of genomic analysis functions, e.g., mapping, alignment, variant calling, etc. In particular, the hardwired digital logic circuits of the FPGA may be arranged as a set of processing engines to perform one or more pre-configured steps in a sequence analysis segment of genomics analysis, such as when the set (or sets) of processing engines includes one or more of a mapping and/or alignment and/or variant calling module, wherein the modules may be formed by separate or equal subsets of processing engines.

[036] As indicated, the system may be configured to include one or more processing engines, and in various embodiments, an included processing engine itself may be configured to determine one or more transition probabilities for the nucleotide sequence of the genomic sequence read transitioning from one state to another, such as from a match state to an indel state, or a match state to a deletion state and/or back, such as from an insertion or deletion state back to a match state. Additionally, in various instances, the integrated circuit may have a segmented configuration and/or may include a second and/or third and/or fourth subset of hardwired digital logic circuitry, such as including a second set of processing engines, wherein the second set of processing engines includes a mapping module configured to map the genomic sequence read to the reference haplotype sequence to produce a mapped read. A third subset of hardwired digital logic circuitry may also be included, such as when the third set of processing engines includes an alignment module configured to align the mapped read to one or more positions in the reference haplotype sequence. A fourth subset of hardwired digital logic circuitry may additionally be included such as when the fourth set of processing engines includes a classification module to classify the mapped and/or aligned read to their relative positions on the chromosome. As above, in various such cases, the mapping module and/or the alignment module and/or the classification module, for example, together with the variant calling module, may be physically integrated into the expansion card. And, in certain embodiments, the expansion card may be physically integrated into a genetic sequencer, such as a next generation sequencer and the like.

[037] Accordingly, in one aspect, there is provided an apparatus for performing one or more steps of a sequence analysis segmentation, such as on genetic data, wherein the genetic data includes one or more of a reference genetic sequence (or sequences), such as a haplotype or hypothetical sequence, an index of the one or more reference genetic sequences, and/or a plurality of reads, such as of genetic and/or genomic data, wherein the data may be stored in one or more shared memory devices and/or processed by a distributed processing resource, such as a CPU/GPU/QPU and/or FPGA, that are coupled, for example, loosely or tightly. Therefore, in various cases, the apparatus may include an integrated circuit, wherein the integrated circuit may include one or more, for example, an array, of hard-wired digital logic circuits, wherein the array of hard-wired digital logic circuits may be interconnected, such as by means of one or a plurality of physical electrical interconnections.

[038] Accordingly, the system may be configured to include an integrated circuit formed by one or more digital logic circuits that are interconnected by a plurality of physical electrical interconnects, wherein one or more of the plurality of physical electrical interconnects have one or more of a memory and/or cache interface, for the integrated circuit to access memory and/or data stored therein and retrieve the same, such as in a cache-coherent manner between the CPU/GPU/QPU and the associated chip, e.g., FPGA. In various cases, the digital logic circuits may include at least a first subset of digital logic circuits, such as where the first subset of digital logic circuits may be arranged as a first set of processing engines, wherein the processing engine may be configured to access data stored in the cache and/or directly or indirectly coupled memory. For example, the first set of processing engines may be configured to perform one or more steps in a mapping and/or alignment and/or classification analysis, as described above, and/or an HMM analysis on reads of genomic sequence data and haplotype sequence data.

[039] More particularly, a first set of processing mechanisms may include an HMM module, such as in a first embodiment of the subset of digital logic circuitry, which is adapted to access in memory, e.g., via the memory interface, at least some of the nucleotide sequences in the genomic sequence data readout and the haplotype sequence data, and may also be configured to perform HMM analysis on at least some of the nucleotide sequences in the genomic sequence data readout and at least some of the nucleotide sequences in the haplotype sequence data so as to produce HMM result data. Additionally, the one or more of the plurality of physical electrical interconnects may include an output of the integrated circuit, such as for communicating the HMM result data from the HMM module, such as to a CPU/GPU/QPU of a server or cluster of servers.

[040] Accordingly, in one aspect, a method for performing a sequence analysis segmentation, such as on genetic sequence data, is provided. The genetic data may include one or more reference genetic or haplotype sequences, one or more indices of the one or more reference genetic and/or haplotype sequences, and/or a plurality of genomic data reads. The method may include one or more of receiving, accessing, mapping, aligning, classifying multiple iterations of the genetic sequence data, and/or employing the results thereof in a method for producing one or more variant call files. For example, in certain embodiments, the method may include receiving, at an input to an integrated circuit from an electronic data source, one or more of a plurality of genomic data reads, wherein each genomic data read may include a nucleotide sequence.

[041] In various cases, the integrated circuit may be formed by a set of hardwired digital logic circuits that may be arranged as one or more processing engines. In such a case, a processing engine may be formed by a subset of the hardwired digital logic circuits that may be in a hardwired configuration. In such a case, the processing engine may be configured to perform one or more preconfigured steps, such as to implement one or more of receiving, accessing, mapping, aligning, classifying multiple iterations of the genetic sequence data, and/or employing the results thereof in a method for producing one or more variant call files. In some embodiments, the provided digital logic circuits may be interconnected such as by a plurality of physical electrical interconnects, which may include an input.

[042] The method may further include accessing, by the integrated circuit at one or more of the plurality of physical electrical interconnections from a memory, data to perform one or more of the operations detailed herein. In various cases, the integrated circuit may be part of a chip set, such as embedded or otherwise contained as part of an FPGA, ASIC, or structured ASIC, and the memory may be directly or indirectly coupled to one or both of the chip and/or a CPU/GPU/QPU associated therewith. For example, the memory may be a plurality of memories, one each coupled to the chip and a CPU/GPU/QPU that is itself coupled to the chip, e.g., loosely.

[043] In other cases, the memory may be a single memory that may be coupled to a CPU/GPU/QPU that is itself tightly coupled to the FPGA, e.g., via a tight processing interconnect or fast path interconnect, e.g., QPI, and thus accessible to the FPGA, such as in a cache-coherent manner. Accordingly, the integrated circuit may be directly or indirectly coupled to the memory in order to access data relevant to performing the functions disclosed herein, such as to access one or more of a plurality of reads, one or more reference or theoretical reference gene sequences, and/or an index of the one or more reference gene sequences, e.g., in performing a mapping operation.

[044] Thus, in various instances, implementations of various aspects of the disclosure may include, but are not limited to: apparatuses, systems, and methods that include one or more features as described in detail herein, as well as articles comprising a tangibly embodied machine-readable medium operable to cause one or more machines (e.g., computers, etc.) to result in the operations described herein. Similarly, computer systems are also described, which may include one or more processors and/or one or more memories coupled to the one or more processors. Accordingly, computer-implemented methods that are consistent with one or more implementations of the present subject matter may be implemented by one or more data processors residing on a single computing system or on multiple computing systems containing multiple computers, such as on a computing bench or supercomputing array.

[045] Such multiple computing systems may be connected and may exchange data and/or commands or other instructions or the like through one or more connections, including but not limited to a connection over a network (e.g., the Internet, a wireless wide area network, a local area network, a wide area network, a wired network, a physical electrical interconnect, or the like), through a direct connection between one or more of the multiple computing systems, etc. A memory, which may include a computer-readable storage medium, may include, encode, store, or the like one or more programs that cause one or more processors to perform one or more of the operations associated with one or more of the algorithms described herein.

[046] Details of one or more variations of the subject matter described herein are set forth in the accompanying drawings and in the description below. Other features and advantages of the subject matter described herein will be apparent from the description and drawings, and from the claims. While certain features of the subject matter disclosed herein are described for illustrative purposes with respect to an enterprise resource software system or other business software solution or architecture, it should be readily understood that such features are not intended to be limiting. The claims that follow this disclosure are intended to define the scope of the protected subject matter.

BRIEF DESCRIPTION OF THE FIGURES

[047] The accompanying drawings, which are incorporated into and constitute a part of this descriptive report, show certain aspects of the subject matter disclosed herein and, in conjunction with the description, help to explain some of the principles associated with the disclosed implementations.

[048] Figure 1A represents a sequencing platform with a plurality of genetic samples therein, a plurality of exemplary tiles are also represented, as well as a three-dimensional representation of the sequenced reads.

[049] Figure 1B shows a representation of a flow cell with the various lanes represented.

[050] Figure 1C depicts a lower corner of the flow cell platform of Figure 1B, which shows a constellation of sequenced reads.

[051] Figure 1D represents a virtual arrangement of the sequencing results performed on the reads in Figures 1 and 2, where the reads are presented in an output column in column order.

[052] Figure 1E depicts the method by which transposition of result readings from column-by-column order to row-by-row reading order can be implemented.

[053] Figure 1F represents the transposition of the result readings from column-by-column order to row-by-row reading order.

[054] Figure 1G represents the system components to perform the transposition.

[055] Figure 1H represents the transposition order.

[056] Figure 1I represents the architecture for electronically transposing sequenced data.

[057] Figure 2 represents a model based on 3 HMM states that illustrates the transition probabilities of going from one state to the other.

[058] Figure 3A is a high-level view of an integrated circuit of the disclosure, including an HMM interface structure.

[059] Figure 3B represents the integrated circuit of Figure 3A, which shows an HMM clustering feature in more detail.

[060] Figure 4 represents an overview of HMM-related data flow throughout the system, including both software and hardware interactions.

[061] Figure 5 depicts exemplary HMM cluster collar connections.

[062] Figure 6 depicts a high-level view of the main functional blocks within an exemplary HMM hardware accelerator.

[063] Figure 7 depicts an exemplary HMM matrix structure and hardware processing flow.

[064] Figure 8 is an enlarged view of a portion of Figure 2 showing the data flow and dependencies between nearby cells in the HMM M, I, and D state computations within the matrix.

[065] Figure 9 depicts exemplary computations useful for M, I, D state updates.

[066] Figure 10 depicts M, I, and D state update circuits, including the effects of simplifying assumptions from Figure 9 related to transition probabilities and the effect of sharing some adder features of M, I, D with the final sum operations.

[067] Figure 11 represents logarithmic domain M, I, D state calculation details.

[068] Figure 12 represents an HMM state transition diagram that shows the relationship between transition probabilities, GOP and GCP.

[069] Figure 13 represents an HMM Transprobs and Priors generation circuit to support the general state transition diagram of Figure 12.

[070] Figure 14 represents a simplified HMM state transition diagram that shows the relationship between transition probabilities, GOP and GCP.

[071] Figure 15 represents a Transprobs and HMM Priors generation circuit to support simplified state transition.

[072] Figure 16 represents an exemplary theoretical HMM matrix and illustrates how much an HMM matrix can be traversed.

[073] Figure 17A presents a method for performing a multi-region joint detection preprocessing procedure.

[074] Figure 17B presents an exemplary method for computing a connection matrix as in the preprocessing procedure of Figure 17A.

[075] Figure 18A represents an exemplary event between two homologous sequenced regions in a read pool.

[076] Figure 18B represents the constructed reads from Figure 18A, demarcating the nucleotide difference between the two sequences.

[077] Figure 18C depicts several bubbles from a De Brujin graph that can be used to perform an accelerated variant calling operation.

[078] Figure 18D presents a representation of a pruning tree function as described herein.

[079] Figure 18E represents one of the bubbles in Figure 18C.

[080] Figure 19 is a graphical representation of the exemplary accumulation in accordance with the connection matrix of Figure 17.

[081] Figure 20 is a processing matrix for performing the preprocessing procedure of Figures 17A and B.

[082] Figure 21 is an example of a bubble formation on a De Brujin chart in accordance with the methods of Figure 20.

[083] Figure 22 is an example of a variant path through a De Brujin graph.

[084] Figure 23 is a graphical representation of an exemplary classification function.

[085] Figure 24 is another example of a processing matrix for joint detection processing of multiple pruned regions.

[086] Figure 25 illustrates a joint accumulation of paired reads for two regions.

[087] Figure 26 presents a probability table in accordance with what is disclosed in this document.

[088] Figure 27 is another example of a processing matrix for joint detection processing of multiple regions.

[089] Figure 28 represents a selection of candidate solutions for the joint accumulation of Figure 25.

[090] Figure 29 represents another selection of candidate solutions for the accumulation of Figure 28, after a pruning function has been performed.

[091] Figure 30 represents the final candidates from Figure 28, and their associated probabilities, after performing an MRJD function.

[092] Figure 31 illustrates the ROC curves for MRJD and a conventional detector.

[093] Figure 32 illustrates the same results as Figure 31 displayed as a function of the sequence similarity of the references.

[094] Figure 33A depicts an exemplary architecture that illustrates loose coupling between a CPU and an FPGA of the disclosure.

[095] Figure 33B depicts an exemplary architecture that illustrates tight coupling between a CPU and an FPGA of the disclosure.

[096] Figure 34A represents a direct coupling of a CPU and an FPGA of the disclosure.

[097] Figure 34B represents an alternative embodiment of the direct coupling of a CPU and an FPGA of Figure 34A.

[098] Figure 35 depicts an embodiment of a combined CPU and FPGA package, wherein the two devices share a common memory and/or cache.

[099] Figure 36 illustrates a core of CPUs that share one or more memories and/or caches, where the CPUs are configured to communicate with one or more FPGAs that may also include common or shared memory or caches.

[0100] Figure 37 illustrates an exemplary method of transferring data throughout the system.

[0101] Figure 38 represents the embodiment of Figure 36 in more detail.

[0102] Figure 39 depicts an exemplary method for processing one or more jobs of a development system.

[0103] Figure 40A depicts a block diagram for a genomic infrastructure for on-premises and/or cloud-based genomic processing and analysis.

[0104] Figure 40B depicts a block diagram of a cloud-based genomics processing platform for performing the BioIT analysis disclosed herein.

[0105] Figure 40C depicts a block diagram for an exemplary genome processing and segmentation analysis.

[0106] Figure 40D depicts a block diagram for an exemplary genome processing and segmentation analysis.

[0107] Figure 41A depicts a block diagram of a local and/or cloud-based computing function of Figure 40A for a genomic infrastructure for on-premises and/or cloud-based genomic processing and analysis.

[0108] Figure 41B depicts the block diagram of Figure 41A which illustrates greater detail regarding the compute function for a genomic infrastructure for on-premises and/or cloud-based genomic processing and analysis.

[0109] Figure 41C depicts the block diagram of Figure 40 that illustrates greater detail regarding third-party analytics functionality for a genomic infrastructure for on-premises and/or cloud-based genomic processing and analysis.

[0110] Figure 42A depicts a block diagram illustrating a hybrid cloud configuration.

[0111] Figure 42B depicts the block diagram of Figure 42A in greater detail, illustrating a hybrid cloud configuration.

[0112] Figure 42C depicts the block diagram of Figure 42A in greater detail, illustrating a hybrid cloud configuration.

[0113] Figure 43A depicts a block diagram illustrating a primary, secondary, and/or tertiary analysis segmentation as presented herein.

[0114] Figure 43B provides an exemplary tertiary processing epigenetic analysis for execution by the methods and apparatus of the system herein.

[0115] Figure 43C provides an exemplary tertiary processing methylation analysis for execution by the methods and apparatus of the system herein.

[0116] Figure 43D provides an analysis of exemplary tertiary processing structural variants for execution by the methods and devices of the system herein.

[0117] Figure 43E provides an exemplary tertiary cohort processing analysis for execution by the methods and apparatus of the system herein.

[0118] Figure 43F provides an exemplary joint genotyping tertiary processing analysis for execution by the methods and apparatus of the system herein.

[0119] Figure 44 depicts a flowchart for a disclosure analysis segmentation.

[0120] Figure 45 is a block diagram of a hardware processor architecture in accordance with an implementation of the disclosure.

[0121] Figure 46 is a block diagram of a hardware processor architecture according to another implementation.

[0122] Figure 47 is a block diagram of a hardware processor architecture in accordance with yet another implementation.

[0123] Figure 48 illustrates a genetic sequence analysis segmentation.

[0124] Figure 49 illustrates processing steps using a genetic sequence analysis hardware platform.

[0125] Figure 50A illustrates an apparatus in accordance with an implementation of the disclosure.

[0126] Figure 50B illustrates another apparatus in accordance with an alternative implementation of the disclosure.

[0127] Figure 51 illustrates a genomics processing system in accordance with an implementation.

DETAILED DESCRIPTION OF THE REVELATION

[0128] As summarized above, the present disclosure is directed to devices, systems, and methods for employing the same in performing one or more genomics and/or bioinformatics protocols, such as a mapping, alignment, classification, and/or variant calling protocol on data generated through a primary processing procedure, such as on genetic sequence data. For example, in various aspects, the devices, systems, and methods provided herein are configured to perform secondary analysis protocols on genetic data, such as data generated by RNA and/or DNA sequencing, e.g., by a Next Generation Sequencer ("NGS"). In particular embodiments, one or more secondary processing suites for processing genetic sequence data are provided, such as where the suites, and/or individual elements thereof, may be implemented in software, hardware, or a combination thereof in a distributed and/or optimized manner so as to deliver superior sensitivity and improved accuracy across a broader range of sequence-derived data than is currently available in the art. Additionally, as summarized above, the present disclosure is directed to devices, systems, and methods for employing the same in performing one or more tertiary genomics and/or bioinformatics protocols, such as a microarray analysis protocol, a genome, e.g., whole genome analysis protocol, genotyping analysis protocol, exome analysis protocol, epigenome analysis protocol, metagenome analysis protocol, microbiome analysis protocol, genotyping analysis protocol, including pooled genotyping, variant analysis protocols, including structural variants, somatic variants, and GATK, as well as RNA sequencing protocols and other genetic analysis protocols, such as mapped, aligned, and/or other genetic sequence data, such as employing one or more variant call files.

[0129] Accordingly, provided herein are software and/or hardware accelerated platform analysis technologies, e.g., chip-based, for performing secondary and/or tertiary analysis of DNA/RNA sequencing data. More particularly, a platform, or stack, of processing engines, such as in a software-deployed and/or wired configuration, that has been specifically designed to perform secondary genetic analysis, e.g., mapping, alignment, classification, and/or variant calling; and/or may be specifically designed to perform tertiary genetic analysis, such as microarray analysis, a genome, e.g., whole genome analysis, genotyping analysis, exome analysis, epigenome analysis, metagenome analysis, microbiome analysis, genotyping analysis including joint genotyping analysis, variant analysis including structural variant analysis, somatic variant analysis, and GATK analysis, as well as RNA sequencing analysis and other genetic analysis, such as in relation to genetic-based sequencing data, which may have been generated in an optimized format that delivers an improvement in processing speed that is magnitudes faster than standard segmentations that are implemented in known software alone. Additionally, the segmentations presented herein provide improved sensitivity and accuracy across a wide range of sequence-derived data sets, such as nucleic acid or protein-derived sequences.

[0130] As indicated above, in many cases, it is an objective of bioinformatics to determine genomes and/or protein sequences of individual individuals, where the determinations can be used in testing protocols, such as for prophylaxis and/or therapeutic regimens, to better improve the health of each individual individual and of the human race as a whole. Additionally, knowledge of an individual's genome and/or protein composition can be used, such as in drug testing and/or FDA testing, to better predict which, if any, drugs are most likely to work in an individual and/or which are most likely to have harmful side effects, such as by analyzing the individual's genome and/or a protein profile derived therefrom and comparing the same to a predicted biological response to the administration of such a drug.

[0131] Such bioinformatics processing typically involves three well-defined but typically separate phases of information processing. The first phase, called primary processing, involves DNA/RNA sequencing, in which a subject's DNA and/or RNA is obtained and subjected to various processes by which the subject's genetic code is converted into a machine-readable digital code, e.g., a FASTQ file. The second phase, called secondary processing, involves using the generated digital genetic code of the subject to determine the genetic makeup of the individual, e.g., to determine the individual's genomic nucleotide sequence. And the third phase, called tertiary processing, involves performing one or more analyses on the subject's genetic heritage in order to determine therapeutically useful information therefrom.

[0132] Accordingly, once the subject's genetic code is sequenced, such as by a next generation sequencer, so as to produce a machine-readable digital representation of the subject's genetic code, e.g., in a FASTQ and/or BCL file format, it may be useful to process the digitally encoded genetic sequence data obtained from the sequencer and/or sequencing protocol, such as by subjecting the digitally represented data to secondary processing. Such secondary processing, for example, may be used to map and/or align and/or otherwise assemble an entire genomic and/or protein profile of an individual, such as when the individual's entire genetic makeup is determined, e.g., when each nucleotide of each chromosome is determined in sequential order such that the composition of the individual's entire genome has been identified. In such processing, the individual's genome may be assembled as such by comparison with a reference genome, such as a reference standard, for example one or more genomes obtained from the Human Genome Project or similar, in order to determine how the individual's genetic heritage differs from that of the referent(s). This process is commonly known as variant calling. Since the difference between the DNA of any one person and another is 1 in 1,000 base pairs, such a variant calling process can be very laborious and time-consuming, requiring many steps that may need to be performed one after the other and/or simultaneously, as in segmentation, in order to analyze the subject's genomic data and determine how that genetic sequence differs from a given reference.

[0133] In performing a secondary analysis segmentation, such as to generate a variant call file for a given query sequence from an individual subject; a genetic sample, e.g., DNA, RNA, protein sample or the like, may be obtained from the subject. The subject's DNA/RNA may then be sequenced, e.g., by a Next Generation Sequencer (NGS) and/or a sequencer-on-a-chip technology, e.g., in a primary processing step, so as to produce a plurality of read sequence segments ("reads") that cover all or a portion of the subject's genome, such as in an oversampled manner. The final product generated by the sequencing device may be a collection of short sequences, e.g., reads, that represent small segments of the subject's genome, e.g., short genetic sequences that represent the entire genome of the subject. As indicated, typically the information represented by these readings may be in an image file or in a digital format such as a FASTQ, BCL, or similar file format.

[0134] In particular, in a typical secondary processing protocol, a subject's genetic heritage is assembled by comparison with a reference genome. This comparison involves reconstructing the individual's genome from millions upon millions of short read sequences and/or comparing the individual's entire DNA to an exemplary DNA sequence template. In a typical secondary processing protocol, an image, FASTQ, and/or BCL file is received from the sequencer that contains the raw sequenced read data. In order to compare the subject's genome to that of the standard reference genome, it is necessary to determine where each of these reads maps to the reference genome, such as when each is aligned relative to each other, and/or how each read can also be sorted by chromosomal order in order to determine at what position and to which chromosome each read belongs. One or more of these functions may occur prior to performing a variant calling function on the full length sequence, e.g., once assembled. Specifically, once it is determined where in the genome each read belongs, the full-length genetic sequence can be determined and then the differences between the subject's genetic code and that of the referent can be assessed.

[0135] For example, reference-based assembly in a typical secondary processing assembly protocol involves comparing sequenced genomic DNA/RNA from an individual to that of one or more standards, e.g., known reference sequences. Various mapping, alignment, classification, and/or variant calling algorithms have been developed to help accelerate these processes. These algorithms may therefore include some variation of one or more of: mapping, aligning, and/or classifying the millions of reads received from the image, FASTQ, and/or BCL file communicated by the sequencer, to determine where on each chromosome each particular read is located. It is noted that these processes may be implemented in software or hardware, such as by the methods and/or devices described in U.S. Patents 9,014,989 and 9,235,680, both assigned to Edico Genome Corporation and incorporated by reference herein in their entirety. Typically, a common feature behind the functioning of these various algorithms and/or hardware implementations is their use of an index and/or an array to expedite their processing function.

[0136] For example, with respect to mapping, a large number, e.g., all, of the sequenced reads may be processed to determine the possible locations in the reference genome to which these reads may possibly align. One methodology that may be used for this purpose is to make a direct comparison of the read with the reference genome in order to find all matching positions. Another methodology is to employ a prefix or suffix array, or to construct a prefix or suffix tree, for the purpose of mapping the reads to various positions in the reference genome. A typical algorithm useful in performing such a function is a Burrows-Wheeler transform, which is used to map a selection of reads to a reference using a compression formula that compresses repeated sequences of data.

[0137] An additional methodology is to employ a hash table, such as where a selected subset of the reads, a k-mer of a selected length “k”, e.g., a seed, is placed in a hash table as keys and the reference sequence is broken into equivalent k-mer length chunks and these chunks and their location are inserted by an algorithm into the hash table at those locations in the table to which they map, according to a hash function. A typical algorithm for performing this function is “BLAST”, a Basic Local Alignment Search Tool. Such hash table-based programs compare query nucleotide or protein sequences with one or more standard reference sequence databases and calculate the statistical significance of matches. In such ways as these, it can be determined where any given read is likely to be located relative to a reference genome. These algorithms are useful because they require less memory, fewer searches, LUTs, and therefore require less processing resources and time in performing their functions than would otherwise be the case, such as if the subject's genome were being assembled by direct comparison, such as without the use of these algorithms.

[0138] Additionally, an alignment function may be performed to determine, among all possible locations that a given read may map to in a genome, such as in those cases where a read may map to multiple positions in the genome, which is in fact the location from which it is actually derived, such as by being sequenced from it by the original sequencing protocol. This function may be performed on several of the reads, e.g., mapped reads, of the genome and an ordered nucleotide base stream representing a portion or all of the subject's DNA/RNA genetic sequence may be obtained. In conjunction with the ordered genetic sequence, a score may be given for each nucleotide at a given position representing the probability that, for any given nucleotide position, the nucleotide, e.g., "A", "C", "G", "T" (or "U"), predicted to be at that position is in fact the nucleotide that belongs to that assigned position. Typical algorithms for performing alignment functions include the Needleman-Wunsch and Smith-Waterman algorithms. In either case, these algorithms perform sequence alignments between a stream of the subject's research genome sequence and a stream of the reference genome sequence whereby, rather than comparing all genome sequences with each other, segments of a selection of possible lengths are compared.

[0139] Once reads have been assigned to a position, such as relative to the reference genome, which may include identifying which chromosome the read belongs to and/or its offset from the beginning of that chromosome, the reads can be sorted by position. This may allow downstream analyses to take advantage of the oversampling procedures described herein. All reads that overlap a given position in the genome will be adjacent to each other after sorting and they can be recognized in a cluster and readily examined to determine whether the majority of them agree with the reference value or not. If they do not, a variant can be flagged.

[0140] For example, in various embodiments, the methods of the disclosure may include generating a variant call file (VCF) that identifies one or more, e.g., all genetic variants in the subject whose DNA/RNA was sequenced, e.g., relevant to one or more reference genomes. For example, once the actual sample genome is known and compared to the reference genome, variations between the two may be determined, and a list of all variations/deviations between the reference genome (or genomes) and the sample genome may be called, e.g., a variant call file may be produced. Particularly, in one aspect, a variant call file containing all variations from the subject's genetic sequence to the reference sequence (or sequences) may be generated.

[0141] As indicated above, such variations between two genetic sequences can be due to a variety of reasons. Therefore, in order to generate such a file, the subject's genome needs to be sequenced and reconstructed before its variants can be determined. However, there are several problems that can occur when attempting to generate such an assembly. For example, there may be problems with chemistry, sequencing machine and/or human error that occurs in the sequencing process. Furthermore, there may be genetic artefacts that make such reconstructions problematic. For example, a typical problem with performing such assemblies is that there are sometimes large portions of the genome that repeat, such as long sections of the genome that include the same nucleotide chains. Therefore, because any genetic sequence is not unique everywhere, it may be difficult to determine where in the genome an identified read actually maps and aligns. Additionally, there may be a single nucleotide polymorphism (SNP), such as where one base in the subject's genetic sequence has been replaced by another; there may be more extensive substitutions of a plurality of nucleotides; there may be an insertion or a deletion, such as when one or a plurality of bases have been added to or deleted from the subject's genetic sequence; and/or there may be a structural variant, for example, such as caused by the confluence of legs of two chromosomes; and/or there may simply be a deviation that causes a change in the sequence.

[0142] Accordingly, there are two main possibilities for variation. For one, there is actual variation at the particular site in question, e.g., when the person's genome is actually different at a particular site than that of the reference, e.g., there is natural variation due to a SNP (a base substitution), an Insertion or Deletion (of one or more nucleotides in length), and/or there is a structural variant, such as when DNA material from one chromosome is crossed over into a different chromosome or leg or when a particular region is copied twice in the DNA. Alternatively, a variation may be caused by a problem in the read data, through a chemical or machine error, sequencer or aligner error, or other human error. The methods disclosed herein may be employed in a manner to compensate for these types of errors and, more particularly, to distinguish errors in variation due to chemical, machine or human variations from actual variations in the sequenced genome. More specifically, the methods, apparatus and systems for employing the same, as described herein, have been developed in order to clearly distinguish between these two different types of variations and therefore better ensure the accuracy of any call files generated in order to correctly identify true variants.

[0143] Therefore, in particular embodiments, a platform of technologies for performing genetic analyses is provided, wherein the platform may include performing one or more of: mapping, alignment, classification, local realignment, duplicate flagging, base quality score recalibration, variant calling, compression, and/or decompression functions. For example, in various aspects, segmentation may be provided, wherein segmentation includes performing one or more analytical functions as described herein on a genomic sequence from one or more individuals, such as data obtained in an image file and/or a digital file format, e.g., FASTQ or BCL, from an automated sequencer. A typical segmentation to be performed may include one or more of sequencing genetic material, such as a portion or a complete genome, from one or more individual subjects, wherein the genetic material may include DNA, ssDNA, RNA, rRNA, tRNA, and the like, and/or in some cases the genetic material may represent coding or non-coding regions, such as exomes and/or episomes of DNA. The segmentation may include one or more of performing an image processing procedure, a base calling operation, and/or error correction, such as on digitized genetic data, and/or may include one or more of performing a mapping, an alignment, and/or a classification function on the genetic data. In certain cases, the segmentation may include performing one or more of a realignment, a deduplication, a quality recalibration or base scoring, a reduction and/or compression, and/or a decompression on the digitized genetic data. In certain cases, segmentation may include performing a variant calling operation, such as a Hidden Markov Model, on the genetic data.

[0144] Accordingly, in certain cases, the implementation of one or more of these platform functions is for the purpose of performing one or more of determining and/or reconstructing a consensus genomic sequence of a subject, comparing a genomic sequence of the subject to a referent sequence, e.g., a reference or template genetic sequence, determining the way in which the subject's genomic DNA or RNA is different from a referent, e.g., called a variant, and/or performing a tertiary analysis on the subject's genomic sequence, such as for genome-wide variation analysis, gene function analysis, protein function analysis, e.g., protein binding analysis, quantitative and/or assembly analysis of genomes and/or transcriptomes, as well as for various diagnostic and/or prophylactic and/or therapeutic evaluation analyses.

[0145] As indicated above, in one aspect, one or more of these platform functions, e.g., mapping, alignment, classification, realignment, duplicate marking, base quality score recalibration, variant calling, compression, and/or decompression functions, are configured for implementation in software. In some aspects, one or more of these platform functions, e.g., mapping, alignment, classification, local realignment, duplicate marking, base quality score recalibration, decompression, variant calling, compression, and/or decompression functions, are configured for implementation in hardware, e.g., firmware. In certain aspects, these genetic analysis technologies may employ improved algorithms that can be implemented through software that operates in a less processing-intensive and/or less time-consuming manner and/or with higher percentage accuracy, and, e.g., the hardware-implemented functionality is faster, less processing-intensive, and more accurate.

[0146] For example, in certain embodiments, improved algorithms are provided for performing such primary, secondary, and/or tertiary processing as disclosed herein. The improved algorithms are directed to more efficiently and/or more accurately performing one or more of the functions of mapping, alignment, classification, and/or variant calling, such as on an image file and/or a digital representation of DNA/RNA sequence data obtained from a sequencing platform, such as in a FASTQ or BCL file format obtained from an automated sequencer, such as one of those set forth above. In certain embodiments, the improved algorithms may be directed to more efficiently and/or more accurately performing one or more of the functions of local realignment, duplicate flagging, base quality score recalibration, variant calling, compression, and/or decompression. Additionally, as described in greater detail herein below, in certain embodiments, these genetic analysis technologies may employ one or more algorithms, such as enhanced algorithms, that may be implemented by one or more software and/or hardware that operates in a less processing intensive and/or less time consuming manner and/or with higher percentage accuracy than various traditional software implementations for doing the same. In various cases, enhanced algorithms for implementation on a quantum processing platform are provided.

[0147] Thus, in various aspects, systems, apparatus, and methods for implementing bioinformatics protocols are disclosed herein, such as for performing one or more functions for analyzing genetic data, such as genomic data, e.g., by one or more optimized algorithms and/or in one or more optimized integrated and/or quantum circuits, such as in one or more hardware processing platforms. In one instance, systems and methods are provided for implementing one or more algorithms, e.g., in software and/or in firmware and/or by a quantum processing circuit, for performing one or more steps for analyzing genomic data in a bioinformatics protocol, such as where the steps may include performing one or more of: mapping, alignment, classification, local realignment, duplicate flagging, base quality score recalibration, variant calling, compression, and/or decompression, and may further include one or more steps in a tertiary processing platform. Accordingly, in certain instances, methods, including software, firmware, hardware, and/or quantum processing algorithms for performing the methods are disclosed herein, wherein the methods involve performing an algorithm, such as an algorithm for implementing one or more genetic analysis functions, such as mapping, alignment, classification, realignment, duplicate flagging, base quality score recalibration, variant calling, compression, decompression, and/or one or more tertiary processing protocols, wherein the algorithm, e.g., including firmware, has been optimized according to the form in which it is to be implemented.

[0148] In particular, when the algorithm is to be implemented in a software solution, the algorithm and/or its correlated processes have been optimized to be performed faster and/or with better accuracy for execution by those media. Likewise, when the functions of the algorithm are to be implemented in a hardware solution, e.g., as firmware, the hardware has been designed to perform those functions and/or their correlated processes in a manner optimized to be performed faster and/or with better accuracy for execution by those media. Additionally, when the algorithm is to be implemented in a quantum processing solution, the algorithm and/or its correlated processes have been optimized to be performed faster and/or with better accuracy for execution by those media. Such methods, for example, may be employed as in an iterative procedure of mapping, alignment, classification, variant calling, and/or tertiary processing. In another case, systems and methods are provided for implementing the functions of one or more algorithms for performing one or more steps for analyzing genomic data in a bioinformatics protocol as set forth herein, wherein the functions are implemented in a hardware and/or quantum accelerator, which may or may not be coupled to one or more general purpose processors and/or supercomputers and/or quantum computers.

[0149] More specifically, in some cases, methods and/or mechanisms for implementing those methods for performing secondary analyses on data pertaining to the genetic makeup of a subject are provided. In one case, the analyses to be performed may involve reference-based reconstruction of the subject's genome. For example, reference-based mapping involves the use of a reference genome, which may be generated from genome sequencing of a single or multiple individuals, or it may be an amalgam of DNA/RNA from multiple individuals that have been combined in such a way as to produce a prototypical standard reference genome to which any genetic material of the individual, e.g., DNA/RNA, may be compared, e.g., in order to determine and reconstruct the individual's genetic sequence and/or to determine the difference between his or her genetic heritage and that of the standard reference, e.g., called a variant.

[0150] In particular, one reason to perform a secondary analysis on a subject's sequenced DNA/RNA is to determine how the subject's DNA/RNA varies from that of the reference, such as to determine one, a plurality, or all of the differences in the subject's nucleotide sequence from that of the reference. For example, the differences between the genetic sequences of any two random people is 1 in about 1,000 base pairs, which, when taken in view of the complete genome of over 3 billion base pairs, equates to a variation of up to 3,000,000 divergent base pairs per person. Determining such differences may be useful, such as in a tertiary analysis protocol, for example, in order to predict the potential for a disease state to occur, such as due to a genetic abnormality, and/or the likelihood of success of a prophylactic or therapeutic modality, such as based on how a prophylactic or therapeutic modality is expected to interact with the subject's DNA or proteins generated therefrom. In many cases, it may be useful to perform both a de novo and a reference-based reconstruction of the subject's genome in order to confirm the results of one against the other and to, where desired, improve the accuracy of a variant calling protocol.

[0151] Accordingly, in one aspect, in various embodiments, once the subject's genome has been reconstructed and/or a VCF has been generated, such data may then be subjected to tertiary processing in order to interpret the same, such as to determine what the data means in relation to identifying what diseases such a person may or may have the potential to suffer from and/or to determine what treatments or lifestyle changes such a subject may wish to employ in order to ameliorate and/or prevent a disease state. For example, the subject's genetic sequence and/or its variant call file may be analyzed to determine clinically relevant genetic markers that indicate the existence or potential for a disease state and/or the efficacy that a proposed therapeutic or prophylactic regimen may have in the subject. Such data may then be used to provide the subject with one or more therapeutic or prophylactic regimens in order to improve the subject's quality of life, such as to treat and/or prevent a disease state.

[0152] In particular, once one or more of an individual's genetic variations are determined, such variant call file information can be used to develop medically useful information, which in turn can be used to determine, e.g., using various known statistical analysis models, health-related data and/or medical information useful, e.g., for diagnostic purposes, e.g., diagnosing a disease or a potential for the same, clinical interpretation (e.g., searching for markers representing a disease variant), whether the subject should be included or excluded in various clinical trials, and other such purposes. More particularly, in various instances, the processed genomics and/or bioinformatics results data generated may be employed in the performance of one or more tertiary genomics and/or bioinformatics protocols, such as a microarray analysis protocol, a genome, e.g., whole genome analysis protocol, a genotyping analysis protocol, an exome analysis protocol, an epigenome analysis protocol, a metagenome analysis protocol, a microbiome analysis protocol, a genotyping analysis protocol including pooled genotyping, variant analysis protocols including structural variants, somatic variants, and GATK, as well as RNA sequencing protocols, and other genetic analysis protocols.

[0153] Since there are a finite number of disease states that are caused by genetic malformations, in tertiary processing, variants of a given type, e.g., those known to be related to the emergence of disease states, may be screened for, such as by determining whether one or more genetically based disease markers are included in the subject's variant call file. Accordingly, in various instances, the methods disclosed herein may involve analyzing, e.g., scanning, the VCF and/or the generated sequence against a known disease sequence variant, such as in a genomic marker database, in order to identify the presence of the genetic marker in the VCF and/or the generated sequence and, if present, make a call on the presence of or potential for a genetically induced disease state. Since there are a large number of known genetic variations and a large number of individuals suffering from diseases caused by such variations, in some embodiments, the methods disclosed herein may entail generating one or more databases that link sequenced data for a complete genome and/or a variant call file pertaining thereto, e.g., such as from an individual or a plurality of individuals, and a disease state and/or searching the generated databases to determine whether a particular subject has a genetic makeup that would predispose him or her to having such a disease state. Such a search may involve comparing a complete genome to one or more others or a fragment of a genome, such as a fragment containing only the variations, to one or more fragments of one or more other genomes, such as in a database of reference genomes or fragments thereof.

[0154] Therefore, in various instances, a segmentation of the disclosure may include one or more modules, wherein the modules are configured to perform one or more functions, such as an image processing or a base calling operation and/or error correction and/or a mapping and/or an alignment, e.g., a gapped or gapless alignment, and/or a classification function on genetic data, e.g., genetic sequenced data. Furthermore, in various instances, the segmentation may include one or more modules, wherein the modules are configured to perform one or more of a local realignment, a deduplication, a base quality score recalibration, a variant calling, e.g., HMM, a reduction and/or compression, and/or a decompression on the genetic data. Additionally, the segmentation may include one or more modules, wherein the modules are configured to perform a tertiary analysis protocol, such as microarray protocols, genome, e.g., whole genome protocols, genotyping protocols, exome protocols, epigenome protocols, metagenome protocols, microbiome protocols, genotyping protocols including pooled genotyping protocols, variant analysis protocols including structural variant protocols, somatic variant protocols, and GATK protocols, as well as RNA sequencing protocols and other genetic analysis protocols.

[0155] Many of these modules may be performed by software or hardware, locally or remotely, e.g., via software or hardware, such as in the cloud, e.g., on a remote server and/or server farm, such as a quantum computing cluster. Additionally, many of these modules and/or many of these segmentation steps are optional and/or may be arranged in any logical order and/or omitted entirely. For example, the software and/or hardware disclosed herein may or may not include an image processing and/or a base calling or sequence correction algorithm, such as where there may be a concern that such functions may result in statistical bias. Accordingly, the system may or may not include a base calling and/or sequence correction function, respectively, depending on the level of accuracy and/or efficiency desired. Furthermore, as noted above, one or more of the segmentation functions may be employed in generating a genomic sequence of a subject, such as through a reference-based genomic reconstruction. Additionally, as indicated above, in certain cases, the output of secondary processing segmentation may be a variant call file (VCF, gVCF) that indicates a portion or all of the variants in a genome or a portion thereof.

[0156] In particular, once reads are assigned a position relative to the reference genome, which may include identifying which chromosome the read belongs to and its offset from the beginning of that chromosome, they may be deduplicated and/or classified, such as by position. This allows downstream analyses to take advantage of the various supersampling protocols described herein. All reads that overlap a given position in the genome may be positioned adjacent to each other after classification, and they may be accumulated, for example to form a backlog, and promptly examined to determine whether the majority of them agree with the reference value or not. If they do not agree, as indicated above, a variant may be flagged.

[0157] Accordingly, as indicated above with respect to mapping, the image file, BCL file, and/or FASTQ file obtained from the sequencer is comprised of a plurality, e.g., millions to a billion or more, of reads consisting of short streams of nucleotide sequence data representing a portion or the entire genome of an individual. For example, a first step in the secondary analysis segmentations disclosed herein is the receipt of genomic and/or bioinformatics data, such as from a genomic data generating device such as a sequencer. Typically, the data produced by a sequencer, e.g., a Next Generation Sequencer, may be in a BCL file format, which in some cases may be converted to a FASTQ file format prior to or subsequent to transmission, such as in a secondary processing platform described herein. In particular, when sequencing a human genome, a subject's DNA and/or RNA needs to be identified, in a base-by-base fashion, where the results of such sequencing are a BCL file. A BCL file is a binary file that includes the base calls and quality scores produced for each base of each sequence in the collection of sequences that make up at least part or all of a subject's genome.

[0158] Traditionally, the sequencer-generated BCL file is converted to a FASTQ file, which can then be transmitted to a secondary processing platform, as disclosed herein, for further processing, such as to determine genomic variance thereof. A FASTQ file is a text-based file format for transmitting both a biological sequence (e.g., nucleotide sequence) and its corresponding quality scores, wherein both the sequence letter, e.g., A, C, G, T, and/or U, and the quality score may each be encoded with a single ASCII character for brevity. Consequently, within this and other systems, it is the FASTQ file that is used for further processing purposes. While employing a FASTQ file for genomic processing is useful, converting the generated BCL file to a FASTQ file, as implemented in the sequencer apparatus, is time-consuming and inefficient. Thus, in one aspect, devices and methods are provided for directly converting a BCL file to a FASTQ file and/or for directly inserting such data into the present platform segmentations, as described herein.

[0159] For example, in various embodiments, a Next Generation sequencer, or a sequencer on a chip technology, may be configured to perform a sequencing operation on received genetic data. For example, as can be seen with respect to Figure 1A, genetic data 6a may be coupled to a sequencing platform 6 for input into a Next Generation sequencer to be sequenced in an iterative manner, such that each sequence will be grown by the gradual addition of one nucleotide after another. Specifically, the sequencing platform 6 may include a plurality of template nucleotide sequences 6a from the subject that are arranged in a grid-like manner to form tiles 6b on the platform 6, in which the template sequences 6a are to be sequenced. The platform 6 may be added to a flow cell 6c of the sequencer that is adapted to perform the sequencing reactions.

[0160] As the sequencing reactions occur, at each step a nucleotide having a fluorescent label 6d is added to the platform 6 of the flow cell 6c. If a hybridization reaction occurs, fluorescence is observed, an image is taken, the image is then processed, and an appropriate base call is performed. This is repeated base by base until all template sequences, e.g. the entire genome, have been sequenced and converted to reads, thus producing the read data for the system. Then, once sequenced, the generated data, e.g. reads, need to be transferred from the sequencing platform to the secondary processing system. For example, typically, this image data is converted to a BCL and/or FASTQ file that can then be transported to the system.

[0161] However, in many cases, this conversion and/or transfer process can be accomplished more efficiently. Specifically, methods and architectures for converting output BCL into files that can be rapidly processed within the secondary processing system are disclosed herein. More specifically, in particular cases, rather than transmitting raw BCL or FASTQ files, the produced images representing each tile from the sequencing operation can be transferred directly to the system and prepared for mapping and alignment, and the like. For example, the tiles can be transmitted over a suitably configured PCIe and to the ASIC, FPGA, or QPU, where read data can be extracted therefrom directly, and the reads advanced to mapping and alignment and/or other processing mechanisms.

[0162] Particularly, with respect to transferring data from the tiles obtained by the sequencer to the FPGA/CPU/GPU/QPU, as can be seen with respect to Figure 1A, the sequencing platform 6 can be imaged as a 3-D cube 6c, within which the growing sequences 6a are generated. Essentially, as can be seen with respect to Figure 1B, the sequencing platform 6 can be composed of 16 lanes, 8 in the front and 8 in the back, which can be configured to form about 96 tiles 6b. Within each tile 6b are several template sequences 6a to be sequenced, thus forming reads, where each read represents the nucleotide sequence for a given region of a subject's genome, where each column represents a file, and as digitally encoded represents 1 byte for each file, with 8 bits per file, such as where 2 bits represents a base call, and the remaining 6 bits represent the quality score.

[0163] More particularly, with respect to Next Generation Sequencing, sequencing is typically performed on glass plates 6 that form flow cells 6c that are inserted into the automated sequencer for sequencing. As can be seen with respect to Figure 1B, a flow cell 6c is a platform 6 composed of 8 vertical columns and 8 horizontal rows (front and back), which together form 16 lanes, each lane being sufficient for sequencing an entire genome. DNA and/or RNA 6a from a subject to be sequenced is associated with designated positions between fluidly isolated intersections of the columns and rows of the platform 6 so as to form tiles 6b, each tile comprising template genetic material 6a to be sequenced. The sequencing platform 6 therefore includes a plurality of template nucleotide sequences from the subject, the sequences being arranged in a manner similar to a grid of tiles on the platform. (See Figure 1B.) The genetic data 6 are sequenced in an iterative manner, where each sequence is grown by stepwise introducing one nucleotide after another into the flow cell, with each iterative grown step representing a sequencing cycle.

[0164] As indicated, an image is captured after each step, and the growing sequence, e.g., of images, forms the basis from which the BCL file is generated. As can be seen with respect to Figure 1C, the reads from the sequencing procedure can form clusters, and it is these clusters that form the theoretical 3-D cube 6c. Accordingly, within this theoretical 3-D cube, each base of each growing nucleotide strand that is sequenced will have an x-dimension and a y-dimension. The image data, or tiles 6b, of this 3-D cube 6c can be extracted and compiled into a two-dimensional map, from which a matrix, as seen in Figure 1AD, can be formed. The matrix is formed by the sequencing cycles, which represent the horizontal axis, and the read identities, which represent the vertical axis. Accordingly, as can be seen with reference to Figure 1C, the sequenced reads form clusters in the flow cell 6c, where the clusters can be defined by a vertical and horizontal axis, cycle by cycle, and the base by base data of each cycle for each read can be inserted into the matrix of Figure 1D, such as in a continuous transmission and/or segmented manner.

[0165] Specifically, each cycle represents the potential growth of each read within the flow cell by the addition of one nucleotide, which when sequencing one or more human genomes can represent the growth of about 1 billion or more reads per lane. Growth of each read, for example by the addition of one nucleotide base, is identified by iteratively capturing images of tiles 6b of flow cell 6c between growth steps. From these images, base calls are produced and quality scores are determined, the virtual matrix of Figure 1D is formed. Accordingly, both a base call and a quality score will be inserted into the matrix, with each tile of each cycle representing a separate file. It should be appreciated that when sequencing is performed on an integrated circuit, the detected electronic data may be replaced by the image data.

[0166] For example, as can be seen with respect to Figure 1D, the matrix itself will grow iteratively as images are captured and processed, bases are called, and quality scores are determined for each read, cycle by cycle. This is repeated for each base in the read, for each tile of the flow cell. For example, the 1C cluster of reads can be numbered and entered into the matrix as the vertical axis. Similarly, the number of cycles can be entered as the horizontal axis, and the base call and quality score can then be entered so as to fill the matrix column by column, row by row. Consequently, each read will be represented by a number of bases, e.g., about 100 or 150 to 1,000 or more bases per read depending on the sequencer, and there can be up to 10 million or more reads per tile. So if there are about 100 tiles, each having 10 million readings, the array could contain about 1 billion readings, which need to be organized and transmitted to the secondary processing apparatus.

[0167] Accordingly, such organization is critical to quickly and efficiently processing the data. Accordingly, in one aspect, methods are disclosed herein for transposing the data represented by the virtual sequencing array in a manner that the data can be more directly and efficiently transmitted to the segments of the system disclosed herein. For example, the generation of the sequencing data, as represented by the star cluster of Figure 1C, is largely disorganized, which is problematic from a data processing standpoint. In particular, as the data is generated by the sequencing operation, it is organized as one file per cycle, meaning that by the end of the sequencing operation, there are millions upon millions of files generated, where the files are represented in Figure 1E by the data in the columns, demarcated by the solid lines. However, for purposes of secondary and/or tertiary processing, as disclosed herein, the file data needs to be reorganized into read data, demarcated by the dashed lines of Figure 1E.

[0168] More particularly, in order to more efficiently transmit the sequencer-generated data to the secondary processing data, the data represented by the virtual matrix must be transposed, such as by reorganizing the file data from a column-by-column tile-per-cycle phase to a row-by-row basis, identifying the bases of each of the reads. Specifically, the data structure of the generated files forming the matrix, as produced by the sequencer, is organized on a cycle-by-cycle, column-by-column basis. By the processes disclosed herein, such data may be transposed, for example, substantially simultaneously, so as to be represented, as seen within the virtual matrix, on a read-by-read, row-by-row basis, where each row represents an individual read, and each read is represented by a sequential number of base calls and quality scores, thereby identifying both the sequence for each read and its confidence. Thus, in a transposition operation as described herein, data within memory may be rearranged, for example within the virtual matrix, from a column-by-column basis, which represents the input data order, to a row-by-row basis, which represents the output data order, thereby transposing the data order from a vertical to a horizontal organization. Furthermore, while the process can be implemented efficiently in software, it can be accomplished even more efficiently and quickly by being implemented in hardware and/or by a quantum processor.

[0169] For example, in many cases, this transposition process can be accelerated by being implemented in hardware. For example, in one implementation, in a first step, the host software, e.g., the sequencer, can write input data to the memory associated with the FPGA on a column-by-column basis, e.g., in the order of input. Specifically, as data is generated and stored in an associated memory, the data can be organized into files, cycle by cycle, where the data is saved as separate individual files. Such data can be represented by the 3D cube of Figure 1A. This generated column-organized data may then be queued and/or transmitted, e.g., in-flight, to hardware, where dedicated processing engines will queue the column-organized data and transpose this data from a column-by-column cycle order configuration to a row-by-row read order configuration, in a manner as described above herein, such as by converting the 3-D tiled data into a 2-D matrix, so that the column data may be reorganized into row data, e.g., on a read-by-read basis. This transposed data may then be stored in memory in a more strategic order.

[0170] For example, the host software may be configured to write input data into memory associated with the chip, e.g., FPGA, such as in a column-by-column order, and similarly, the hardware may be configured to queue the data so that it is read into memory in a strategic manner, such as shown in Figure 1F. Specifically, the hardware may include a register array 8a in which cycle files may be scattered and reorganized into individual read data, such as by writing a column-by-column basis into registers that are arranged in rows. More specifically, as can be seen with respect to Figure 1G, the hardware device 1 including the transposition processing engine 8 may include a DRAM port 8a that can queue data to be transposed, wherein the port is operatively coupled to a memory interface 8b that is associated with a plurality of registers and/or an external memory 8c, and is configured to handle an increased number of transactions per cycle, wherein the queued data is transmitted in bursts.

[0171] In particular, this transposition may occur one data segment at a time, such as when memory accesses are queued in such a manner as to take maximum advantage of the DDR throughput. For example, with respect to DRAM, the minimum DDR shot length may be, e.g., 64 bytes. Accordingly, column-arranged data stored in host memory may be accessed in a manner such that with each memory access, a corresponding column value of bytes, e.g., 64, is obtained. Thus, with one memory access, a portion of a tile, e.g., representing "64" corresponding cycles or files, may be accessed, on a column-by-column basis.

[0172] However, as can be seen with respect to Figure 1F, although data in host memory is accessed as column data, when transmitted to hardware, it may be uploaded into associated smaller memories, e.g., registers, in a different order, so that the data may be converted into bytes, e.g., 64 bytes, of read data row by row, such as in compliance with the DDR minimum firing rate, so as to generate "64" corresponding memory units or blocks per access. This is exemplified by the virtual array of Figure 1D, where multiple reads, e.g., 64 reads, are accessed in blocks, and the memory read in segments, as depicted in Figure 1E, such as where each register, or flip-flop, represents a particular read, e.g., 64 cycles x 64 reads x 8 bits per read = 32K flip-flops. Specifically, this can be accomplished in several different ways in hardware, such as when the input write is arranged to match column ordering, and the output write is arranged to match row ordering. Thus, in this configuration, the hardware can be adapted to both read and/or write "64" different addresses per cycle.

[0173] More particularly, the hardware may be associated with a register arrangement such that each base of a read is directed and written to a single register (or multiple registers in a row), so that when each block is completed, the newly ordered row data may be transmitted to memory as an output, e.g., FASTQ data, in a row-by-row arrangement. The FASTQ data may then be accessed by one or more processing engines of the secondary processing system for further processing, such as by a mapping, alignment, and/or variant calling engine as described herein. It should be appreciated, as described herein, that the transposition is performed in small blocks, however, the system may be adapted to process larger blocks as well, as the case may be.

[0174] As indicated, once a BCL file has been converted to a FASTQ file as described above, and/or a BCL or FASTQ file has otherwise been received by the secondary processing platform, a mapping operation may be performed on the received data. Mapping generally involves plotting the reads to all locations in the reference genome at which there is a match. For example, depending on the length of the read, there may be one or a plurality of locations at which the read substantially matches a corresponding sequence in the reference genome. Thus, the mapping and/or other functions disclosed herein may be configured to determine where among all possible locations that one or more reads may match in the reference genome is actually the actual location to which they map.

[0175] For example, in various cases, an index of a reference genome may be generated or otherwise provided such that reads or portions of reads may be searched, e.g., within a Lookup Table (LUT) by reference to the index, thereby retrieving indications of locations in the reference in order to map the reads to the reference. Such an index of the reference may be constructed in a variety of ways and searched in a variety of ways. In some methods, the index may include a prefix and/or suffix tree. In particular methods, the index may be derived from a Burrows/Wheeler transform of the reference. Thus, alternatively, or in addition to employing a prefix tree or a suffix tree, a Burrows/Wheeler transform may be performed on the data. For example, a Burrows/Wheeler transform can be used to store a tree-like data structure that is abstractly equivalent to a prefix and/or suffix tree, in a compact format, such as in the space allocated to store the reference genome. In many cases, the stored data is not in a tree-like structure, but instead the reference sequence data is in a linear list that may have been shuffled in a different order to transform it into a very particular form, so that the accompanying algorithm allows the reference to be searched against the sample reads in order to efficiently traverse the “tree”.

[0176] Additionally, in various cases, the index may include one or more hash tables, and the methods disclosed herein may include a hash function that may be performed on one or more portions of the reads in an effort to map the reads to the reference, e.g., to the reference index. For example, alternatively, or in addition to utilizing one or both a prefix/suffix tree and a Burrows/Wheeler transform on the reference genome and the subject sequence data in order to find where one maps against the other, another such method involves producing a hash table index and/or performing a hash function. The hash table index may be a large reference structure that is constructed from sequences of the reference genome that may then be compared to one or more portions of the read to determine where one portion may match another. Similarly, the hash table index can be constructed from portions of the read that can then be compared to one or more reference genome sequences and thus used to determine where one may match the other.

[0177] Implementing a hash table is a fast method for performing pattern matching. Each search takes an approximately constant amount of time to perform. Such a method can be contrasted with the Burrows-Wheeler method, which may require many probes (the number may vary depending on how many bits are required to find a unique pattern) per search to find a match, or a binary search method which has log2(N) probes, where N is the number of seed patterns in the table. Additionally, although the hash function can break the reference genome into seed segments of any given length, e.g. 28 base pairs, and can then convert the seeds into a digital, e.g. binary, 56-bit representation, not all 56 bits need to be accessed completely at the same time or in the same way. For example, the hash function may be employed such that the address for each seed is designated by a number smaller than 56 bits, such as about 18 to about 44 or 46 bits, such as about 20 to about 40 bits, such as about 24 to about 36 bits, including about 28 to about 32 or about 30 bits may be used as a key or starting address in order to access the hash table. For example, in certain cases, about 26 to about 29 bits may be used as a primary access key for the hash table, leaving about 27 to about 30 bits remaining, which may be employed as a means to double-check the first key, e.g., if both the first and second keys arrive at the same cell in the hash table, then it is relatively clear that said location is where they belong.

[0178] For example, a first portion of the digitally represented seed, e.g., about 26 to about 32, such as about 29 bits, may form a primary access key and be hashed and searchable in a first pass. Further, in a second pass, the remaining about 27 to about 30 bits, e.g., a secondary access key, may be inserted into the hash table, such as in a hash chain, as a means to confirm the first pass. Accordingly, for any seed, its original address bits may be hashed in a first pass and the secondary address bits may be used in a second confirmation step. In such a case, the first portion of the seeds may be inserted into a primary record location and the second portion may be embedded in the table in a secondary record chain location. Further, as noted above, in various cases, these two different record locations may be positionally separated, such as by a record in chain format.

[0179] In particular cases, a brute force linear scan may be employed to compare the reference to the read, or portions thereof. However, using a brute force linear search to scan the reference genome for locations at which a seed matches, over 3 billion locations may have to be checked. Such searching may be performed, in accordance with the methods disclosed herein, in software or hardware. However, using a hashing approach as disclosed herein, each seed lookup may occur in approximately a constant amount of time. Typically, the location may be determined in a few accesses, e.g., a single access. However, in cases where multiple seeds map to the same location in the table, e.g., are not unique enough, a few additional accesses may be made to find the seed currently being searched. Thus, although there are 30M or more possible locations for a given 100-nucleotide-long read to match against a reference genome, the hash table and hash function can quickly determine where that read will appear in the reference genome. Using a hash table index, therefore, it is not necessary to brute-force search the entire reference genome to determine where the read maps and aligns.

[0180] In view of the foregoing, any suitable hash function may be employed for these purposes, however, in many cases, the hash function used to determine the table address for each seed may be a cyclic redundancy check (CRC) that may be based on a 2k-bit primitive polynomial, as indicated above. Alternatively, a trivial hash function mapper may be employed, such as simply dropping some of the 2k bits. However, in many cases, the CRC may be a stronger hash function that can better separate similar seeds, while at the same time avoiding table congestion. This may be especially beneficial when there is no speed penalty when calculating CRCs, such as with the dedicated hardware described herein. In such cases, the hash record populated for each seed may include the reference position at which the seed occurred, and the flag indicating whether it was reverse complemented prior to hashing.

[0181] The output returned from performing a mapping function may be a list of possibilities as to where one or more, e.g., each, read maps to one or more reference genomes. For example, the output for each mapped read may be a list of possible locations where the read may map to a corresponding sequence in the reference genome. In various embodiments, an exact match to the reference for at least a portion, e.g., a seed of the read, if not all, of the read may be sought. Accordingly, in many cases, it is not necessary for all portions of all reads to exactly match all portions of the reference genome.

[0182] As described herein, all of these operations may be performed via software or may be hard-wired, such as on an integrated circuit, such as on a chip, for example as part of a circuit board. For example, the operation of one or more of these algorithms may be embedded on a chip, such as an FPGA (field programmable gate array) chip or an ASIC (application specific integrated circuit) chip, and may be optimized for more efficient performance due to its implementation in such hardware. Additionally, one or more, for example two or all three, of these mapping functions may form a module, such as a mapping module, which may form part of a system, for example a segmentation, that is used in a process for determining an entire actual genomic sequence, or a portion thereof, of an individual.

[0183] One advantage of implementing the hash module in hardware is that the processes can be accelerated and therefore performed much more quickly. For example, where software may include multiple instructions to perform one or more of these multiple functions, implementation of such instructions typically requires that data and instructions be stored and/or fetched and/or read and/or interpreted, such as prior to execution. As noted above, however, and described in greater detail herein, a chip may be wired to perform these functions without having to fetch, interpret, and/or perform one or more of a sequence of instructions. Instead, the chip may be wired to perform such functions directly. Accordingly, in many aspects, the disclosure is directed to a custom wired machine that may be configured such that portions or all of the mapping, e.g., hashing, module described above may be implemented by one or more network circuits, such as wired integrated circuits on a chip, such as an FPGA or an ASIC.

[0184] For example, in various cases, the hash table index may be constructed and the hash function may be performed on a chip, and in other cases, the hash table index may be generated from the chip, such as by software operated by a host CPU, but once generated, it is loaded or otherwise made accessible to hardware and employed by the chip, such as in the operation of the hash module. In particular, in various cases, the chip, such as an FPGA, may be configured to be tightly coupled to the host CPU, such as by a low-latency interconnect such as a QPI interconnect. More particularly, the chip and the CPU may be configured to be tightly coupled together in such a manner that they share one or more memory resources, e.g., a DRAM, in a cache-coherent configuration, as described in more detail below. In such a case, the host memory may construct and/or include the reference index, e.g., the hash table, which may be stored in the host memory but be made readily accessible to the FPGA such as for its use in performing a hash or other mapping function. In particular embodiments, one or both of the CPU and the FPGA may include one or more caches or registers that may be coupled together to be in a coherent configuration, such that data stored in one cache may be substantially mirrored by the other.

[0185] Accordingly, in view of the foregoing, at runtime, one or more previously constructed hash tables, e.g., containing an index of a reference genome, or a constructed or to-be-constructed hash table, may be loaded into embedded memory or may at least be made accessible by its host application, as described in more detail below herein. In such a case, the reads, e.g., stored in FASTQ file format, may be sent by the host application to embedded processing mechanisms, e.g., a memory or cache or other registry associated therewith, such as for use by a mapping and/or alignment and/or classification engine, such as when the results thereof may be sent and used to perform a variant calling function. In this regard, as indicated above, in various cases, an accumulation of overlapping seeds may be generated, for example, by means of a seed generation function, and extracted from the sequenced reads, or pairs of reads, and, once generated, the seeds may be hashed, such as against an index, and looked up in the hash table so as to determine candidate read mapping positions in the reference.

[0186] More particularly, in various instances, a mapping module may be provided, such as when the mapping module is configured to perform one or more mapping functions, such as in a wired configuration. Specifically, the wired mapping module may be configured to perform one or more functions typically performed by one or more algorithms executed on a CPU, such as functions that would typically be implemented in a software-based algorithm that produces a prefix and/or suffix tree, a Burrows-Wheeler Transform, and/or performs a hash function, e.g., a hash function that makes use of, or otherwise relies on, a hash table indexing, such as of a reference, e.g., a reference genome sequence. In such cases, the hash function may be structured so as to implement a strategy, such as an optimized mapping strategy that may be configured to minimize the number of memory accesses, e.g., large random memory accesses, that is performed so as to maximize the utility of embedded or otherwise associated memory bandwidth, which may be fundamentally constrained such as by space within the chip architecture.

[0187] Furthermore, in certain cases, in order to make the system more efficient, the host CPU/GPU/QPU may be tightly coupled to associated hardware, e.g., FPGA, such as by a low-latency interface, e.g., Fast Path Interconnect ("QPI"), so as to allow the processing engines of the integrated circuit to have ready access to the host memory. In particular cases, the interaction between the host CPU and the coupled chip and their respective associated memories, e.g., one or more DRAMs, may be configured to be cache-coherent. Thus, in various embodiments, an integrated circuit may be provided wherein the integrated circuit has been preconfigured, e.g., prewired, in such a manner as to include one or more digital logic circuits that may be in a wired configuration, which may be interconnected, e.g., by one or a plurality of physical electrical interconnections, and in various embodiments, the wired digital logic circuits may be arranged in one or more processing engines to form one or more modules, such as a mapping module.

[0188] Accordingly, in various instances, a mapping module may be provided, such as in a first pre-configured wired-connected configuration, e.g., wired, wherein the mapping module is configured to perform various mapping functions. For example, the mapping module may be configured to access at least some of a nucleotide sequence in a read from a plurality of reads derived from a sequenced genetic sample of the subject, and/or a reference genetic sequence, and/or an index of one or more reference genetic sequences, from a memory or cache associated therewith, e.g., via a memory interface, such as a process interconnect, e.g., a Fast Path Interconnect, and the like. The mapping module may be further configured to map the read to one or more segments of the one or more reference genetic sequences, such as based on the index. For example, in various particular embodiments, the algorithm and/or mapping module presented herein may be employed to establish, or otherwise construct, a hash table such that the read, or a portion thereof, of the subject's sequenced genetic material may be compared to one or more segments of a reference genome to produce mapped reads. In such a case, once mapping has been performed, an alignment may be performed.

[0189] For example, after it has been determined where all the possible matches are for the seeds against the reference genome, it is necessary to determine which of all the possible locations a given read may correspond to is in fact the correct position to which it aligns. Thus, after mapping, there may be a plurality of positions that one or more reads appear to match in the reference genome. Consequently, there may be a plurality of seeds that appear to be indicating exactly the same thing, for example, they may all correspond to exactly the same position in the reference, if the position of the seed in the read is taken into account. The actual alignment therefore needs to be determined for each read. This determination can take a number of different forms.

[0190] In one case, all reads may be evaluated to determine their correct alignment against the reference genome based on the positions indicated by each seed from the read that returned position information during the mapping search process, e.g., hashing. However, in many cases, prior to performing an alignment, a seed strand filtering function may be performed on one or more of the seeds. For example, in certain cases, seeds associated with a given read that appear to map to the same general location as against the reference genome may be aggregated into a single strand that references the same general region. All seeds associated with a read may be grouped into one or more seed strands such that each seed is a member of only one strand. If such strand(s) do, then the read is aligned to each indicated position in the reference genome.

[0191] Specifically, in many cases, all of the seeds that have the same supporting evidence indicating that they belong to the same general locus (or loci) in the reference may be collected to form one or more chains. Seeds that therefore cluster together, or at least appear to be close to each other in the reference genome, e.g., within a certain band, will be grouped into one seed chain, and those that fall outside of that band will be placed into a different seed chain. Once these diverse seeds have been aggregated into one or more diverse seed chains, it can be determined which of the chains actually represents the correct chain to be aligned. This can be done, at least in part, by the use of a filtering algorithm that is an arithmetic heuristic designed to eliminate weak seed chains that have a high probability of not being the correct chain.

[0192] The output of performing one or more of these mapping, filtering and/or editing functions is a list of reads that includes, for each read, a list of all possible locations at which the read may match the reference genome. Thus, a mapping function may be performed in order to quickly determine where the reads from the image file, BCL file and/or FASTQ file obtained from the sequencer map to the reference genome, e.g., where in the complete genome the various reads map. However, if there is an error in any of the reads or a genetic variation, it may not be possible to achieve an exact match to the reference and/or there may be multiple locations at which one or more reads appear to match. Therefore, it needs to be determined where the various reads actually align relative to the genome as a whole.

[0193] Accordingly, after mapping and/or filtering and/or editing, the locus positions for a large number of reads have been determined, wherein for some of the individual reads a multiplicity of locus positions have been determined, and it is now necessary to determine which of all the possible loci is in fact the true or most likely locus to which the multiple reads align. Such alignment may be performed by one or more algorithms, such as a dynamic programming algorithm that matches the mapped reads to the reference genome and operates an alignment function thereon. An exemplary alignment function compares one or more, e.g., all, reads to the reference, such as by placing them in a graphical relationship to each other, e.g., such as in a table, e.g., an array or a virtual matrix, in which the sequence of one of the reference genome or mapped reads is placed on one dimension or geometric axis, e.g., the horizontal geometric axis, and the other is placed on the opposite dimensions or geometric axis, such as the vertical geometric axis. A conceptual scoring wavefront is then passed through the array to determine the alignment of the reads to the reference genome, such as by computing alignment scores for each cell in the matrix.

[0194] The scoring wavefront represents one or more, e.g., all, cells of a matrix or a portion of such cells, which may be scored independently and/or simultaneously according to the dynamic programming rules applicable in the alignment algorithm, such as Smith-Waterman, and/or Needleman-Wunsch, and/or related algorithms. The alignment scores may be computed sequentially or in other orders, such as computing all scores in the top row from left to right, followed by all scores in the next row from left to right, etc. Thus, the diagonally sweeping diagonal wavefront represents an ideal sequence of batches of scores computed simultaneously or in parallel in a series of wavefront steps.

[0195] For example, in one embodiment, a window of the reference genome containing the segment to which a read has been mapped can be placed on the horizontal axis and the read can be positioned on the vertical axis. In such a manner, an array or matrix is generated, e.g., a virtual matrix, by which the nucleotide at each position in the read can be compared to the nucleotide at each position in the reference window. As the wavefront passes through the array, all potential ways of aligning the read to the reference window are considered, including the possibility that changes to a sequence may be required to make the read match the reference sequence, such as changing one or more nucleotides of the read to other nucleotides, or inserting one or more new nucleotides into a sequence, or deleting one or more nucleotides from a sequence.

[0196] An alignment score, representing the extent of changes that would be required to be made to achieve an exact alignment, is generated, where this score and/or other associated data may be stored in the given cells of the array. Each cell of the array corresponds to the likelihood that the nucleotide at its position on the read axis aligns with the nucleotide at its position on the reference axis, and the score generated for each cell represents the partial alignment that ends with the cell's positions in the read and in the reference window. The highest score generated in any cell represents the best overall alignment of the read to the reference window. In many cases, the alignment may be global, where the entire read must be aligned to some portion of the reference window, such as using a Needleman-Wunsch or similar algorithm; or, in other cases, the alignment may be local, where only a portion of the read can be aligned to a portion of the reference window, such as using a Smith-Waterman or similar algorithm.

[0197] Accordingly, in various cases, an alignment function may be performed, such as on data obtained from the mapping module. Thus, in various cases, an alignment function may form a module, such as an alignment module, which may form part of a system, e.g., a segmentation system, that is used, such as in addition to a mapping module, in a process for determining the entire actual genomic sequence, or a portion thereof, of an individual. For example, the output returned from the performance of the mapping function, such as from a mapping module, e.g., the list of possibilities as to where one or more or all of the reads map to one or more positions in one or more reference genomes, may be used by the alignment function to determine the actual sequence alignment of the sequenced DNA of the subject.

[0198] Such an alignment function may sometimes be useful due to the fact that, as described above, sometimes, for a variety of different reasons, the sequenced reads do not always exactly match the reference genome. For example, there may be a SNP (single nucleotide polymorphism) in one or more of the reads, e.g., a substitution of one nucleotide for another at a single position; there may be an “indel,” insertion or deletion, of one or more bases along one or more of the read sequences, where the insertion or deletion is not present in the reference genome; and/or there may be a sequencing error (e.g., errors in sample preparation and/or sequencer readout and/or sequencer output, etc.) that makes one or more of these variations evident. Consequently, when a read varies from the reference, such as by a SNP or Indel, this may be due to the fact that the reference is different from the true DNA sequence sampled, or due to the fact that the read is different from the true DNA sequence sampled. The problem is figuring out how to correctly align the reads to the reference genome given the fact that, in all likelihood, the two sequences will vary from each other in a multitude of different ways.

[0199] In many cases, the input to an alignment function, such as from a mapping function such as a prefix/suffix tree, or a Burrows/Wheeler transform, or a hash table and/or a hash function, may be a list of possibilities as to where one or more reads may match one or more positions of one or more reference sequences. For example, for any given read, it may match any number of positions in the reference genome, such as 1 location, or 16, or 32, or 64, or 100, or 500, or 1,000 or more locations to which a given read maps in the genome. However, any individual read was derived, e.g., sequenced, from only a specific portion of the genome. Thus, in order to find the true location from which a given particular read was derived, an alignment function can be performed, e.g., a Smith-Waterman gapped or gapless alignment, a Needleman-Wunsch alignment, etc., in order to determine where in the genome one or more of the reads were actually derived, such as by comparing all possible locations at which a match occurs and determining which of all the possibilities is the most likely location in the genome from which the read was sequenced, based on which location's alignment score is the highest.

[0200] As indicated, typically, an algorithm is used to perform such an alignment function. For example, a Smith-Waterman and/or Needleman-Wunsch alignment algorithm may be employed to align two or more sequences against each other. In this case, they may be employed in a manner to determine the probabilities that, for any given position where the read maps to the reference genome, the mapping is in fact the position from which the read originated. Typically, these algorithms are configured to be performed by software, however, in many cases, such as in the present document, one or more of these algorithms may be configured to be performed in hardware, as described in greater detail in the present document below.

[0201] In particular, the alignment function operates, at least in part, to align one or more, e.g., all, reads to the reference genome regardless of the presence of one or more portions of divergences, e.g., SNPs, insertions, deletions, structural artifacts, etc. in order to determine where the reads are likely to fit the genome correctly. For example, the one or more reads are compared against the reference genome and the best possible fit for the read against the genome is determined, while considering substitutions and/or Indels and/or structural variants. However, to best determine which of the modified versions of the read fits best against the reference genome, the proposed changes need to be considered and as such a scoring function may also be performed.

[0202] For example, a scoring function may be implemented, for example as part of a general alignment function, whereby, as the alignment module performs its function and introduces one or more changes to a sequence that is compared with another, for example in order to achieve a better fit or the best fit between the two, for each change that is made in order to achieve the better alignment, a number is subtracted from an initial score, for example a perfect score or an initial score of zero, in a manner such that the alignment is performed and the score for the alignment is also determined, such that where matches are detected, the score is increased, and for each change introduced a penalty is incurred, and in this way the best fit for the possible alignments can be determined, for example by finding which of all possible modified reads fits the genome with the highest score. Consequently, in many cases, the alignment function can be configured to determine the best combination of changes that need to be made to the read (or reads) to achieve the highest scoring alignment, where the alignment can then be determined to be the correct or most likely alignment.

[0203] In view of the above, there are therefore at least two goals that can be achieved by performing an alignment function. One is a report of the best alignment, which includes the position in the reference genome and a description of what changes are needed to make the read match the reference segment at that position, and the other is an alignment quality score. For example, in many cases, the output of the alignment module may be a Compact Idiosyncratic Incomplete Alignment Report, e.g., a CIGAR stream, where the CIGAR stream output is a report that details all the changes that were made to the reads in order to achieve their best matching alignment, e.g., detailed alignment instructions that indicate how the query actually aligns to the reference. Such a CIGAR current display may be useful in further processing stages in order to better determine that, for a given subject's genomic nucleotide sequence, predicted variations in comparison against a reference genome are, in fact, true variations and not simply due to machine, software or human error.

[0204] As set forth above, in various embodiments, the alignment is typically performed in a sequential manner, wherein the algorithm and/or firmware receives read sequence data, such as from a mapping module, pertaining to a read and one or more possible locations to which the read may potentially map to the one or more reference genomes, and further receives genomic sequence data, such as from one or more memories, such as associated DRAMs, pertaining to one or more positions in the one or more reference genomes to which the read may map. In particular, in various embodiments, the mapping module processes the reads, such as from a FASTQ file, and maps each of them to one or more positions in the reference genome to which they may possibly align. The aligner then takes these predicted positions and uses them to align the reads to the reference genome, such as by constructing a virtual array by which the reads may be compared to the reference genome.

[0205] In performing this function, the aligner evaluates each mapped position for each individual read and particularly evaluates those reads that map to multiple possible locations in the reference genome and scores the likelihood that each position is the correct position. It then compares the best scores, e.g. the two best scores, and makes a decision about where the particular read actually aligns. For example, in comparing the first and second best alignment scores, the aligner looks at the difference between the scores and if the difference between them is large then the confidence score that the one with the higher score is correct will be high. However, when the difference between them is small e.g. zero then the confidence score about being able to tell from which of the two positions the read was actually derived is low and further processing may be useful to be able to clearly determine the true location in the reference genome from which the read is derived.

[0206] Thus, the aligner is looking, in part, for the largest difference between the first and second best confidence scores about making its call that a given read maps to a given location in the reference genome. Ideally, the score for the best possible alignment choice is significantly higher than the score for the second best alignment for that sequence. There are many different ways that an alignment scoring methodology can be implemented, for example, every cell in the array can be scored or a subportion of cells can be scored, such as according to the methods disclosed herein. In many cases, the scoring parameters for nucleotide matches, nucleotide mismatches, insertions, and deletions can have any positive or negative value, or zero. In many cases, these scoring parameters can be modified based on available information. For example, accurate alignments can be achieved by making scoring parameters, including any or all of the nucleotide divergence scores, nucleotide divergence scores, gap penalties (insertion and/or deletion), gap opening penalties, and/or extended gap penalties, vary based on a base quality score associated with the current nucleotide or read position. For example, scoring bonuses and/or penalties can be made smaller when a base quality score indicates a high probability that a sequencing error or other error is present. Base quality-sensitive scoring can be implemented, for example, using a fixed or configurable lookup table, accessed using a base quality score, which returns corresponding scoring parameters.

[0207] In a hardware implementation on an integrated circuit, such as an FPGA or an ASIC, a scoring wavefront may be implemented as a linear array of scoring cells, such as 16 cells, or 32 cells, or 64 cells, or 128 cells, or the like. Each of the scoring cells may be constructed of digital logic elements in a wired configuration to compute alignment scores. Thus, for each step of the wavefront, e.g., each clock cycle, or some other fixed or variable unit of time, each of the scoring cells, or a portion of the cells, computes the score or scores required for a new cell in the virtual alignment matrix. Notionally, the various scoring cells are considered to be at various positions in the alignment matrix, which correspond to a scoring wavefront, as discussed herein, e.g., along a straight line extending from lower left to upper right in the matrix. As will be well understood in the field of digital logic design, the physical scoring cells and their underlying digital logic need not be physically arranged similarly on the integrated circuit.

[0208] Accordingly, as the wavefront takes steps to sweep through the virtual alignment matrix, the notional positions of the scoring cells update accordingly, each cell, e.g., notionally “moving” one step to the right or, e.g., one step down in the alignment matrix. All scoring cells make the same relative notional movement, keeping the diagonal wavefront arrangement intact. Each time the wavefront moves to a new position, e.g., with a vertical step down or a horizontal step to the right in the matrix, the scoring cells arrive at new notional positions and compute alignment scores for the virtual alignment matrix cells they entered. In such an implementation, neighboring scoring cells in the linear array are coupled to communicate nucleotide lookup (readout), reference nucleotides, and previously computed alignment scores. The reference window nucleotides may be fed sequentially into one end of the wavefront, e.g. the upper right scoring cell in the linear array, and may switch from there sequentially down the length of the wavefront, such that at any given time a segment of reference nucleotides having a length equal to the number of scoring cells is present within the cells, one successive nucleotide in each successive scoring cell.

[0209] For example, each time the wavefront steps horizontally, another reference nucleotide is fed into the upper right cell and other reference nucleotides commute to the lower left across the wavefront. This commutation of reference nucleotides may be the underlying reality of the notional movement of the wavefront of scoring cells to the right across the alignment matrix. Thus, readout nucleotides may be fed sequentially into the opposite end of the wavefront, e.g., the lower left scoring cell in the linear array, and commute from there sequentially up the length of the wavefront, such that at any given time, a segment of query nucleotides having a length equal to the number of scoring cells is present within the cells, one successive nucleotide in each successive scoring cell. Similarly, each time the wavefront steps are vertical, another query nucleotide is fed into the lower left cell and other query nucleotides switch to the upper right across the wavefront. This switching of query nucleotides is the underlying effect of the notional movement of the scoring cell wavefront down through the alignment matrix. Consequently, by commanding a reference nucleotide switch, the wavefront can be moved one step horizontally, and by commanding a query nucleotide switch, the wavefront can be moved one step vertically. Thus, to produce generally diagonal wavefront movement, such as to follow a typical alignment of query and reference sequences without insertions or deletions, wavefront steps can be commanded in alternating vertical and horizontal directions.

[0210] Accordingly, neighboring scoring cells in the linear array may be coupled to communicate previously calculated alignment scores. In various alignment scoring algorithms, such as a Smith-Waterman or Needleman-Wunsch, or such variant, the alignment score (or scores) in each cell of the virtual alignment matrix may be calculated using scores previously calculated in other cells of the matrix, such as the three cells positioned immediately to the left of the current cell, above the current cell, and diagonally to the upper left of the current cell. When a scoring cell calculates new score (or scores) for another matrix position that it has entered, it needs to retrieve such previously calculated scores that correspond to such other matrix positions. Such previously calculated scores may be obtained from the storage of previously calculated scores within the same cell and/or from the storage of previously calculated scores in the one or two neighboring scoring cells in the linear array. This is due to the fact that the three contributing scoring positions in the virtual alignment matrix (immediately left, above, and diagonally to the upper left) would have been scored by the current scoring cell or by one of its neighboring scoring cells in the linear array.

[0211] For example, the cell immediately to the left in the array would have been scored by the current scoring cell if the most recent wavefront step is horizontal (to the right), or would have been scored by the neighboring lower left cell in the linear array if the most recent wavefront step is vertical (down). Similarly, the cell immediately above in the array would have been scored by the current scoring cell if the most recent wavefront step was vertical (down), or would have been scored by the neighboring upper right cell in the linear array if the most recent wavefront step was horizontal (to the right). In particular, the diagonally upper left cell in the array would have been scored by the current scoring cell if the two most recent wavefront steps were in different directions, e.g. down then left, or right then down, or would have been scored by the neighboring upper right cell in the linear array if the two most recent wavefront steps were both horizontal (to the right), or would have been scored by the neighboring lower left cell in the linear array if the two most recent wavefront steps were both vertical (to the bottom).

[0212] Accordingly, given information in the last one or two wavefront step directions, a scoring cell may select appropriate previously calculated scores by accessing them within itself, and/or within neighboring scoring cells, using coupling between neighboring cells. In one variation, scoring cells at the two ends of the wavefront may have outwardly wired scoring inputs for invalid, or zero, or minimum value scores, such that they do not affect new score calculations in those extreme cells. A wavefront that is thus deployed in a linear array of scoring cells, with such coupling to switch the reference and query nucleotides through the array in opposite directions in order to notionally move the wavefront in vertical and horizontal, e.g., diagonal, steps, and coupling to access previously computed scores via neighboring cells in order to compute alignment score (or scores) at new virtual array cell positions inserted by the wavefront, it is consequently possible to score a band of cells in the virtual array, the width of the wavefront, such as by commanding successive steps of the wavefront to sweep across the array.

[0213] For a new read and reference window to be aligned, therefore, the wavefront may begin positioned within the scoring array or, advantageously, may gradually enter the scoring array from outside, starting, for example, to the left of, or above, or diagonally left and above the upper left corner of the array. For example, the wavefront may begin with its upper left scoring cell positioned just to the left of the upper left cell of the virtual array, and the wavefront may then sweep right across the array by a series of horizontal steps, scoring a horizontal band of cells in the upper left region of the array. When the wavefront achieves a predicted alignment relationship between the reference and the query or when a match is detected from increasing alignment scores, the wavefront may begin to sweep diagonally from the lower right, by alternating vertical and horizontal steps, scoring a diagonal band of cells across the middle of the array. When the lower left wavefront scoring cell reaches the bottom of the alignment array, the wavefront may begin to sweep to the right again by successive horizontal steps, until some or all of the wavefront cells sweep outside the boundaries of the alignment array, scoring a horizontal band of cells in the lower right region of the array.

[0214] One or more such alignment procedures may be performed by any suitable alignment algorithm, such as a Needleman-Wunsch alignment algorithm and/or a Smith-Waterman alignment algorithm, which may have been modified to accommodate the functionality described herein. In general, both of these algorithms and those similar thereto perform in some cases in a substantially similar manner. For example, as set forth above, these alignment algorithms typically construct the virtual array in a similar manner such that, in many cases, the upper horizontal boundary may be configured to represent the reference genomic sequence, which may be arranged across the top row of the array according to its base pair composition. Likewise, the vertical boundary may be configured to represent the sequenced and mapped query sequences that have been positioned in order, down the first column, such that their nucleotide sequence order generally matches the nucleotide sequence of the reference to which they map. The intervention cells can then be populated with scores on the probability that the relevant search base at a given position is positioned at that location relative to the reference. In performing this function, a range can be moved diagonally across the matrix that populates scores within the intervention cells and the probability for each search base to be at the indicated position can be determined.

[0215] With respect to a Needleman-Wunsch alignment function, which generates optimal global (or semi-global) alignments, the alignment of the entire read sequence to some segment of the reference genome, the wavefront steering may be configured such that it typically scans all the way from the top edge of the alignment matrix to the bottom edge. When the wavefront scan is complete, the maximum score at the bottom edge of the alignment matrix (which corresponds to the end of the read) is selected and the alignment is traced back to a cell at the top edge of the matrix (which corresponds to the start of the read). In many of the cases disclosed herein, the reads may be of any length, may be of any size, and need not have extensive parameters as to how the alignment is performed, e.g., in many cases, the read may be as large as a chromosome. In such a case, however, memory size and chromosome length may be a limiting factor.

[0216] With respect to a Smith-Waterman algorithm, which generates optimal site alignments, the alignment of the entire read sequence or part of the read sequence to some segment of the reference genome, this algorithm can be configured to find the best possible score based on a complete or partial alignment of the read. Thus, in many cases, the wavefront scoring band may not extend to the upper and/or lower edges of the alignment matrix, such as if a very long read had only seeds at its average mapping to the reference genome, but commonly the wavefront may still score from the top to the bottom of the matrix. Site alignment is typically achieved through two adjustments. First, alignment scores are never allowed to fall below zero (or some other floor), and if an otherwise calculated cell score were negative, a zero score is substituted, representing the start of a new alignment. Second, the maximum alignment score produced in any cell in the matrix, not necessarily along the lower edge, is used as the endpoint of the alignment. The alignment is traced back from this maximum score up and left through the matrix to a score of zero, which is used as the starting position of the location alignment, even if it is not in the top row of the matrix.

[0217] In view of the foregoing, there are several different possible trajectories through the virtual array. In various embodiments, the wavefront starts from the upper left corner of the virtual array and moves downward toward identifiers of the maximum score. For example, the results of all possible alignments may be collected, processed, correlated, and scored to determine the maximum score. When the endpoint of a threshold or the endpoint of the array is reached and/or a computation that leads to the highest score for all processed cells is determined (e.g., the highest overall score identified), then a backward trace may be performed in order to find the trajectory that was taken to achieve that highest score. For example, a trajectory that leads to a predicted maximum score may be identified, and once identified, a retracement may be performed in order to determine how that maximum score was derived, for example, by moving backward following the best-scoring alignment arrows that retrace the trajectory that led to achieving the identified maximum score, as computed by the scoring wavefront cells.

[0218] This backward reconstruction or backtracking involves starting from a given maximum score and working backward through the previous cells, navigating the path of cells that have the scores that led to reaching the maximum score all the way up the table and back to a starting boundary, such as the beginning of the array or a score of zero in the case of local alignment. During a backtracking, having reached a particular cell in the alignment matrix, the next backtracking step is to the neighboring cell, immediately to the left or above, or diagonally from top to left, that contributed the best score that was selected to build the score in the current cell. In this way, the evolution of the maximum score can be determined, thereby discovering how the maximum score was reached. The backtracking may end at a corner, or an edge, or a boundary, or it may end at a score of zero, such as the top left corner of the array. Accordingly, it is such backtracking that identifies the appropriate alignment and thereby produces the CIGAR strand display that represents how the genomic sample sequence derived from the individual, or a portion thereof, matches, or otherwise aligns with, the genomic sequence of the reference DNA.

[0219] Once it has been determined where each read maps to and further determined where each read aligns, for example, each relevant read has been given a position and a quality score that reflects the probability that the position is the correct alignment, such that the nucleotide sequence for the subject's DNA is known, then the order of the various reads and/or the subject's genomic nucleic acid sequence can be verified, such as by performing a backtracking function that moves backward through the array to determine the identity of each nucleic acid in its proper order in the genomic sample sequence. Accordingly, in some aspects, the present disclosure is directed to a backtracking function, such as is part of an alignment module that performs both an alignment and a backtracking function, such as a module that may be part of a segmentation of modules, such as a segmentation that is directed to taking raw read sequence data, such as from a genomic sample of an individual, and mapping and/or aligning that data, wherein the data may then be classified.

[0220] To facilitate the backtracking operation, it is useful to store a score vector for each scored cell in the alignment matrix, encoding the score selection decision. For implementations of classical Smith-Waterman and/or Needleman-Wunsch scoring with linear gap penalties, the score vector may encode four possibilities, which may optionally be stored as a 2-bit integer from 0 to 3, for example: 0 = new alignment (zero score selected); 1 = vertical alignment (score from cell above selected, modified by gap penalty); 2 = horizontal alignment (score from cell left selected, modified by gap penalty); 3 = diagonal alignment (score from cell top selected, modified by match score or nucleotide divergence). Optionally, the score (or scores) computed for each scored matrix cell can also be stored (in addition to the maximum alignment score achieved which is stored by default), but this is not generally necessary for backtracking and can consume large amounts of memory. Performing backtracking then becomes a matter of following the score vectors; when backtracking has reached a given cell in the matrix, the next step in backtracking is determined by the score vector stored for that cell, e.g.: 0 = end backtracking; 1 = backtracking upwards; 2 = backtracking to the left; 3 = backtracking diagonally from top to left.

[0221] Such scoring vectors may be stored in a two-dimensional table arranged according to the dimensions of the alignment matrix, in which only entries that correspond to cells scored by the wavefront are populated. Alternatively, to conserve memory, more easily record scoring vectors as they are generated, and more easily accommodate alignment matrices of varying sizes, the scoring vectors may be stored in a table with each row sized to store scoring vectors from a single wavefront of scoring cells, e.g., 128 bits to store 64 2-bit scoring vectors from a 64-cell wavefront, and a number of rows equal to the maximum number of wavefront steps in an alignment operation. Additionally, for this option, a record may be kept of the directions of the various wavefront steps, for example by storing an extra bit, e.g. 129°, in each row of the table, coding, for example, 0 for the vertical wavefront step preceding that wavefront position and 1 for the horizontal wavefront step preceding that wavefront position. This extra bit may be used during backtracking to keep track of which virtual score matrix positions the score vectors in each row of the table correspond to, so that the appropriate score vector can be retrieved after each successive backtracking step. When a backtracking step is either vertical or horizontal, the next score vector must be retrieved from the previous table row, but when a backtracking step is diagonal, the next score vector must be retrieved from the previous two rows, due to the fact that the wavefront had to make two steps to move from the score of any one cell to the score of the cell diagonally down from it.

[0222] In the case of affine gap scoring, the scoring vector information may be extended, for example, to 4 bits per scored cell. In addition to, for example, the 2-bit score choice direction indicator, two 1 flags may be added, a vertical extension flag and a horizontal extension flag. According to the methods of affine gap scoring extensions to Smith-Waterman or Needleman-Wunsch or similar alignment algorithms, for each cell, in addition to the primary alignment score representing the best-scoring alignment ending in that cell, a “vertical score” must be generated which corresponds to the maximum alignment score reaching that cell with a final vertical step, and a “horizontal score” must be generated which corresponds to the maximum alignment score reaching that cell with a final horizontal step; and when computing any of the three scores, a vertical step in the cell may be computed using the primary score from the cell above minus a gap opening penalty or using the vertical score from the cell above minus a gap extension penalty, whichever is greater; and a horizontal step in the cell may be computed using the primary score from the cell to the left minus a gap opening penalty or using the horizontal score from the cell to the left minus a gap extension penalty, whichever is greater. In cases where the vertical score minus a gap extension penalty is selected, the vertical extension flag in the score vector shall be set, e.g., "1", and otherwise it shall be unset, e.g., "0".

[0223] In cases where horizontal scoring minus a gap extension penalty is selected, the horizontal extension flag in the scoring vector must be set, e.g., “1”, and otherwise must be undefined, e.g., “0”. During backtracking for affine gap scoring, any time the backtracking takes a vertical step upward from a given cell, if that vertical extension flag of the cell's scoring vector is set, the next backtracking step must also be vertical, regardless of the scoring vector for the cell above. Likewise, each time the backtracking takes a horizontal step leftward from a given cell, if that horizontal extension flag of the cell's scoring vector is set, the next backtracking step must also be horizontal, regardless of the scoring vector for the cell to the left. Consequently, such a scoring vector table, e.g., 129 bits per row for 64 cells using linear gap scoring or 257 bits per row for 64 cells using affine gap scoring, with some NR number of rows, is suitable to support backtracking after completing alignment scoring where wavefront scoring took NR steps or less.

[0224] For example, when aligning reads of 300 nucleotides, the number of wavefront steps required may always be less than 1,024, so that the table may be 257 x 1,024 bits or approximately 32 kilobytes, which in many cases may be a reasonable local memory within the integrated circuit. However, if many long reads are to be aligned, e.g. 100,000 nucleotides, the memory requirements for scoring vectors may be quite large, e.g. 8 megabytes, which may be too expensive to include as a local memory within the integrated circuit. For each holder, the scoring vector information may be recorded in mass memory outside the integrated circuit, e.g. DRAM, but then the bandwidth requirements, e.g. 257 bits per clock cycle per aligner module, may be excessive, which may bottleneck and dramatically reduce aligner performance. Consequently, it is desirable to have a method for disposing of scoring vectors before the alignment is complete, so that their storage requirements can be kept together, e.g., to perform incremental backtracking, generating incremental partial CIGAR streams, e.g., from antecedent portions of an alignment scoring vector history, so that such antecedent portions of the scoring vectors can then be discarded. The challenge is that the backtracking must start at the alignment terminal's maximum scoring cell, which is unknown until the alignment scoring is complete, so that any backtracking that started before the alignment was complete may start from the wrong cell, not along the eventual final optimal alignment path.

[0225] Thus, a method is provided for performing incremental backtracking from partial alignment information, e.g., comprising partial scoring vector information for the alignment cell array scored thus far. Starting from a currently completed alignment boundary, e.g., a particular scored wavefront position, backtracking is initiated from all cell positions on the boundary. Such backtracking from all boundary cells may be performed sequentially, or advantageously, especially in a hardware implementation, all backtracks may be performed together. It is not necessary to extract alignment notations, e.g., CIGAR streams, from these multiple backtracks; only to determine which alignment position array positions they pass through during backtracking. In a simultaneous backtracking implementation from a score boundary, multiple 1-bit registers can be used, corresponding to the number of alignment cells initialized, e.g., all to “1”s, representing whether any of the backtracks passed through a matching position. For each simultaneous backtracking step, score vectors that correspond to all current “1”s in those registers, e.g., from a row of the score vector table, can be examined to determine a next backtracking step that corresponds to each “1” in the registers, leading to a next position for each “1” in the registers, for the next simultaneous backtracking step.

[0226] Importantly, it is easily possible for multiple “1”s in the records to merge into common positions, which correspond to multiples of the simultaneous return traces merging into common return trace paths. Once two or more of the simultaneous return traces merge, they remain fused indefinitely, due to the fact that they will henceforth use scoring vector information from the same cell. It has been observed, both empirically and on theoretical grounds, that with high probability all of the simultaneous return traces merge into a singular return trace path within a relatively small number of return trace steps, which, for example, may be a small multiple, e.g. 8, times the number of scoring cells in the wavefront. For example, with a 64-cell wavefront, with high probability all of the return traces from a given wavefront boundary merge into a single return trace path within 512 return trace steps. Alternatively, it is also possible, and not uncommon, for all backtracks to end within the number, e.g. 512, of backtrack steps.

[0227] Accordingly, multiple simultaneous return traces may be performed from a scoring boundary, e.g., a scored wavefront position, far enough away that they all terminate or merge into a single return trace path, e.g., in 512 return trace steps or less. If they all merge into a singular return trace path, then from the location in the scoring matrix where they merge, or any farther distance along the singular return trace path, an incremental return trace from partial alignment information is possible. An additional return trace from the merge point, or any farther distance, is started by normal singular return trace methods, including recording the corresponding alignment notation, e.g., a partial CIGAR stream. This incremental backtrack, and e.g. this partial CIGAR stream, must be part of any possible final backtrack, and e.g. a complete CIGAR stream, which would result after alignment is complete, unless such final backtrack ends before reaching the score threshold at which the simultaneous backtrack begins, because if it reaches the score threshold, it must follow one of the simultaneous backtrack paths and merge into the now incrementally extracted singular backtrack path.

[0228] Therefore, all score vectors for matrix regions that correspond to the incrementally extracted return trace, e.g., in all table rows for wavefront positions preceding the start of the extracted singular return trace, may be safely discarded. When the final return trace is performed from a top-scoring cell, if it terminates before reaching the score threshold (or, alternatively, if it terminates before reaching the start of the extracted singular return trace), the incremental alignment notation, e.g., partial CIGAR chain, may be discarded. If the final return trace continues until the start of the extracted singular return trace, its alignment notation, e.g., CIGAR chain, may then be grafted onto the incremental alignment notation, e.g., partial CIGAR chain. Additionally, in a very long alignment, the process of performing a simultaneous backtrack from a scoring boundary, e.g., scored wavefront position, until all backtracks terminate or merge, followed by a singular backtrack with alignment notation extraction, may be repeated multiple times from several successive scoring boundaries. The incremental alignment notation, e.g., partial CIGAR current, from each successive incremental backtrack may then be grafted onto the previous accumulated alignment notations, unless the new simultaneous backtrack or singular backtrack terminates early, in which case the previous accumulated alignment notations may be discarded. The eventual final backtrack likewise grafts its alignment notation onto the most recent accumulated alignment notations for a complete backtrack description, e.g., CIGAR current.

[0229] Accordingly, in this way, the memory for storing scoring vectors may be kept unitary, assuming that concurrent backtracks always merge at a unitary number of steps, e.g., 512 steps. In the rare cases where concurrent backtracks fail to merge or terminate at the unitary number of steps, several exceptional actions may be taken, including failing the current alignment or repeating it with a larger union or no union at all, perhaps by a different or traditional method, such as storing all scoring vectors for the full alignment, such as in external DRAM. In a variation, it may be reasonable to fail such an alignment, because it is extremely rare, and it is even rarer that such a failed alignment would have been a better-scoring alignment to be used in the alignment report.

[0230] In an optional variation, the score vector storage may be divided, physically or logically, into several distinct blocks, e.g., 512 rows each, and the final row in each block may be used as a score boundary to begin a concurrent backtrack. Optionally, a concurrent backtrack may be required to end or merge within a single block, e.g., 512 steps. Optionally, if the concurrent backtracks merge in fewer steps, the fused backtrack may still be continued through the entire block before beginning an extraction of a singular backtrack in the previous block. Accordingly, after the score vectors are completely written in block N and begin to be written in block N+1, a concurrent backtrack may begin in block N, followed by a singular backtrack and an alignment notation extraction in block N-1. If the speeds of simultaneous backtracking, singular backtracking, and alignment scoring are all similar or identical, and they can be performed simultaneously, for example in parallel hardware on an integrated circuit, then singular backtracking in block N-1 can be concurrent with the scoring vectors that fill block N+2, and when block N+3 needs to be filled, block N-1 can be released and recycled.

[0231] Thus, in such an implementation, a minimum of 4 scoring vector blocks may be employed, and may be used cyclically. Thus, the total scoring vector storage for an aligner module may be 4 blocks of 257 x 512 bits each, for example, or approximately 64 kilobytes. In a variation, if the current maximum alignment score corresponds to a block further back than the current wavefront position, that block and the previous block may be preserved instead of being recycled, so that a final backtrack may start from that position if it remains at the maximum score; having 2 extra blocks to keep preserved thus brings the minimum, for example, to 6 blocks.

[0232] In another variation, to support overlapping alignments, the scoring wavefront gradually crossing from one alignment matrix to the next, as described above, additional blocks, e.g., 1 or 2 additional blocks, may be used, e.g., 8 total blocks, e.g., approximately 128 kilobytes. Accordingly, if such a limited number of blocks, e.g., 4 blocks or 8 blocks, are used cyclically, alignment and backtracking of arbitrarily long reads is possible, e.g., 100,000 nucleotides or an entire chromosome, without the use of external memory for scoring vectors. It should be understood, as with reference to the foregoing, that although a mapping function may in some cases have been described, such as with reference to a mapper, and/or an alignment function may in some cases have been described, such as with reference to an aligner, these different functions may be performed sequentially by the same architecture, which has been commonly referred to in the art as an aligner. Accordingly, in many cases, both the mapping function and the alignment function as described herein may be performed by a common architecture that may be understood to be an aligner, especially in those cases where an alignment function is performed, where a mapping function needs to be performed first.

[0233] In various instances, the devices, systems, and methods of use thereof of the present disclosure may be configured to perform one or more of a gapless and/or full-read gapped alignment which may then be scored to determine the proper alignment for the reads in the data set. For example, in various instances, a gapless alignment procedure may be performed on data to be processed, wherein the gapless alignment procedure may then be followed by one or more of a gapped alignment and/or a selective Smith-Waterman alignment procedure. For example, in a first step, a gapless alignment string may be generated. As described herein, such gapless alignment functions may be performed rapidly, such as without the need to represent gaps, which, after a first step of performing a gapless alignment, may then be followed by performing a gapped alignment.

[0234] For example, an alignment function may be performed to determine how any given nucleotide sequence, e.g., read, aligns to a reference sequence without the need to insert gaps in one or more of the reads and/or the reference. An important part of performing such an alignment function is to determine when and how there are divergences in the sequence in question versus the reference genome sequence. However, due to the large amount of technology within the human genome, in theory, any given nucleotide sequence will broadly match a representative reference sequence. When there are divergences, they will likely be due to a single nucleotide polymorphism, which is relatively easy to detect, or they will be due to an insertion or deletion in the sequences in question, which are much more difficult to detect.

[0235] Consequently, to perform an alignment function, most of the time, the sequence in question will match the reference sequence, and when there is a divergence due to a SNP, this is easily determined. Therefore, a relatively large amount of processing power is not required to perform such an analysis. Difficulties arise, however, when there are insertions or deletions in the sequence in question relative to the reference sequence, due to the fact that such insertions and deletions add up to gaps in the alignment. Such gaps present a more expensive and complicated processing platform in order to determine the correct alignment. However, because there will be only a small percentage of indels, only a relatively smaller percentage of gapped alignment protocols need to be performed compared to the millions of gapless alignments performed. Therefore, only a small percentage of all gapless alignment functions result in a need for additional processing due to the presence of an indel in the sequence and will therefore require a gapped alignment.

[0236] When an indel is indicated in a gapless alignment procedure, only those sequences are passed through an alignment engine for further processing, such as an alignment engine configured to perform an advanced alignment function, such as a Smith Waterman alignment (SWA). Thus, because either a gapless or gapped alignment must be performed, the devices and systems disclosed herein are a much more efficient use of resources. More particularly, in certain embodiments, both a gapless and gapped alignment can be performed on a given selection of sequences, e.g., one right after the other, then the results are compared for each sequence, and the best result is chosen. Such an arrangement can be implemented, for example, when an increase in accuracy is desired, and an increased amount of time and resources to perform the necessary processing is acceptable.

[0237] In particular, in many cases, a first alignment step can be performed without involving a processor-intensive Smith Waterman function. Thus, a plurality of gapless alignments can be performed in a less resource-intensive, less time-consuming manner, and because fewer resources are required, less space needs to be dedicated to such processing on the chip. Thus, more processing needs to be performed, using fewer processing elements, requiring less time, so more alignments can be performed, and better accuracy can be achieved. More particularly, fewer on-chip resource allocations need to be dedicated to performing Smith Waterman alignments using less chip area, since it does not require as much chip area for the processing elements required to perform gapless alignments than it does to perform gapped alignments. As chip resource requirements decrease, more processing can be performed in a shorter period of time, and with more processing able to be performed, better accuracy can be achieved.

[0238] Accordingly, in such cases, a gapless alignment protocol, e.g., to be performed by suitably configured gapless alignment resources, may be employed. For example, as disclosed herein, in various embodiments, an alignment processing engine is provided such as wherein the processing engine is configured to receive digital signals, e.g., representing one or more reads of genomic data, such as digital data denoting one or more nucleotide sequences, from an electronic data source, and map and/or align such data to a reference sequence, such as by first performing a gapless alignment function on such data, wherein the gapless alignment function may then be followed, if necessary, by a gapped alignment function, such as by performing a Smith Waterman alignment protocol.

[0239] Accordingly, in many cases, a gapless alignment function is performed on a contiguous portion of the read, e.g., by employing a gapless aligner, and if the gapless alignment occurs end-to-end, e.g., the read is complete, a gapped alignment is not performed. However, if the results of the gapless alignment are indicative of an indel being present, e.g., the read is trimmed or otherwise incomplete, then a gapped alignment may be performed. Thus, the gapless alignment results may be used to determine whether a gapped alignment is necessary, e.g., when the gapless alignment extends into a gapped region but does not extend the entire length of the read, such as when the read can be recalled, e.g., by soft trimming to some degree, and when trimmed, then a gapped alignment may be performed.

[0240] Thus, in various embodiments, based on alignment completeness and scores, only if the gapless alignment ends up being clipped, e.g., does not run end-to-end, a gapped alignment is performed. More particularly, in various embodiments, the best gapless and/or identifiable gapped alignment score can be estimated and used as a cutoff line to decide whether the score is good enough to warrant further analysis, such as performing a gapped alignment. Thus, alignment completeness and its score can be employed such that a high score is indicative of the alignment being complete and therefore gapless, and a lower score is indicative of the alignment not being complete and a gapped alignment needs to be performed. Thus, when a high score is obtained, a gapped alignment is not performed, but only when the score is low enough, a gapped alignment is performed. Obviously, in many cases, a brute force alignment approach can be employed such that a number of gapped and/or non-gaped aligners are deployed in the chip architecture so as to allow a larger number of alignments to be performed and thus a larger amount of data can be searched.

[0241] More particularly, in various embodiments, each mapping and/or alignment engine may include one or more, e.g., two, Smith-Waterman aligner modules. In certain cases, these modules may be configured to support global gapless (end-to-end) alignment and/or local gapped (clipped) alignment, perform affine gap scoring, and may be configured to generate non-clipped scoring bonuses at each end. Base quality-sensitive matching or divergence scoring may also be supported. When two alignment modules are included, e.g., as part of the integrated circuit, for example, each Smith-Waterman aligner may be constructed as an antidiagonal wavefront of scoring cells, wherein the wavefront “moves” across a virtual alignment rectangle, scoring cells that it sweeps.

[0242] However, for longer reads, the SmithWaterman wavefront can also be configured to support auto-steering to track the best alignment through accumulated indels, such as to ensure that the alignment wavefront and the cells that are scored do not stray from the scoring band. In the background, logic mechanisms can be configured to examine current wavefront scores, find maxima, flag subsets of cells by a threshold distance beyond the maxima, and target the midpoint between two extreme flags. In such a case, auto-steering can be configured to operate diagonally when the target is at the center of the wavefront, but can be configured to operate straight horizontally or vertically as needed to re-center the target if it drifts, such as due to the presence of indels.

[0243] The output of the alignment module is a SAM (Text) file or a BAM (e.g., binary version of a SAM) together with a mapping quality score (MAP), where the quality score reflects the confidence that the predicted and aligned location of the read to the reference is actually where the read is derived. Accordingly, once it has been determined where each read maps and further determined where each read aligns, e.g., each relevant read has been given a position and a quality score that reflect the probability that the position is the correct alignment, such that the nucleotide sequence for the subject's DNA is known, as well as how the subject's DNA differs from that of the reference (e.g., the CIGAR chain has been determined), then the various reads representing the subject's genomic nucleic acid sequence can be sorted by chromosomal location, such that the exact location of the read on the chromosomes can be determined. Accordingly, in some aspects, the present disclosure is directed to a classification function, as may be performed by a classification module, wherein the classification module may be part of a segmentation of modules, such as a segmentation that is directed to taking raw read sequence data, such as from a genomic sample of an individual, and mapping and/or aligning that data, wherein the data may then be classified.

[0244] More particularly, once the reads have been assigned to a position, such as relative to the reference genome, which may include identifying which chromosome the read belongs to and/or its offset from the beginning of that chromosome, the reads may be classified by position. Classification may be useful, such as in downstream analyses, whereby all reads that overlap a given position in the genome may be formed into a cluster so as to be adjacent to each other, such as after being processed through the classification module, whereby it may be readily determined whether the majority of the reads agree with the reference value or not. Thus, when the majority of reads do not agree with the reference value, a variant call may be flagged. Sorting, therefore, may involve one or more of the sorting of reads that align relatively to the same position, such as the same chromosome position, in order to produce a cluster such that all reads covering the same location are physically grouped together, and may further involve analyzing the reads in the cluster to determine where the reads may indicate an actual variant in the genome, as compared to the reference genome, where the variant may be distinguishable, such as by the consensus of the cluster, from an error, such as a machine read error or an error in the sequencing methods that may be exhibited by a small minority of the reads.

[0245] Once the data has been obtained, there are one or more other modules that can be operated in order to clean the data. For example, a module that may be included, for example, in a sequence analysis segment, such as to determine the genomic sequence of an individual, may be a local realignment module. For example, it is typically difficult to determine insertions and deletions that occur at the end of the read. This is because the Smith-Waterman alignment process or an equivalent does not have enough context beyond the indel to allow the score to detect its presence. Consequently, the actual indel may be reported as one or more SNPs. In such a case, the accuracy of the predicted location for any given read may be improved by performing a local realignment on the mapped and/or aligned and/or classified read data.

[0246] In such cases, stacks can be used to help clarify the proper alignment, such as when a position in question is at the end of any given read, it is likely that that same position is in the middle of some other read in the stack. Consequently, in performing a local realignment, the various reads in a stack can be analyzed to determine whether any of the reads in the stack indicate that there was an insertion or deletion at a given position when another read does not include the indel, or instead includes a substitution, at that position; then the indel can be inserted, as in the reference, when it is not present, and the reads in the local stack that overlap that region can be realigned to see if, collectively, a better score is achieved then when the insertion and/or deletion was not there. If there is an improvement, the entire set of reads in the stack can be reviewed, and if the score of the overall set has improved, then it is clear to make the call that there was indeed an indel at that position. In such a way, the fact that there is not enough context to more accurately align a read at the end of a chromosome for any given read can be compensated for. Thus, when a local realignment is performed, one or more clusters where one or more indels might be positioned are examined and it is determined whether adding an indel at any given position can improve the overall alignment score.

[0247] Another module that may be included, for example, in a sequence analysis pipeline, such as for determining the genomic sequence of an individual, may be a duplicate marking module. For example, a duplicate marking function may be performed in order to compensate for chemistry errors that may occur during the sequencing phase. For example, as described above, during some sequencing procedures, nucleic acid sequences are fixed to beads and constructed from these using identified nucleotide bases. Ideally, there will be only one read per bead. However, sometimes multiple reads become fixed to a single bead and this results in an excessive number of copies of the fixed read. This phenomenon is known as read duplication.

[0248] After an alignment is performed and the results obtained, and/or a classification function, local realignment and/or deduplication are performed, a variant calling function may be employed on the resulting data. For example, a typical variant calling function or portions thereof may be configured to be implemented in a software and/or wired configuration, such as on an integrated circuit. In particular, variant calling is a process that involves placing all of the reads that align to a given location in the reference into clusters, such that all of the overlapping regions of all of the various aligned reads form a "cumulation". Then, the accumulation of reads covering a given region of the genome is analyzed to determine what the most likely current content of the sampled individual's DNA/RNA is within that region. This is then repeated, in steps, for each region of the genome. The determined content generates a list of differences called "variations" or "variants" from the reference genome, each with an associated confidence level along with other metadata.

[0249] The most common variants are single nucleotide polymorphisms (SNPs), in which a single base differs from the reference. SNPs occur at about 1 in 1,000 positions in a human genome. The next most common are insertions (in reference) and deletions (from reference), or "indels" collectively. They are most common at shorter lengths, but can be of any length. Additional complications arise, however, because the collection of sequenced segments ("reads") is random, some regions will have deeper coverage than others. There are also more complex variants that include multiple base substitutions and combinations of indels and substitutions that can be across substitutions that alter the length. Standard software-based variant callers have difficulty identifying all of them and with varying limits on variant lengths. More specialized variant callers in both software and hardware are needed to identify longer-term variants, and many exotic "structural variants" involving large chromosome changes.

[0250] However, variant calling is a difficult procedure to implement in software, and much more difficult to implement in hardware. In order to represent and/or detect these types of errors, typical variant callers may perform one or more of the following tasks. For example, they may propose a set of hypothetical genotypes (contents of the one or two chromosomes at a locus), use Bayesian calculations to estimate the posterior probability that each genotype is the true one given the observed evidence, and report the most likely genotype along with its confidence level. As such, variant callers can be simple to complex. The simplest variant callers search only the column of bases in the accumulation of reads aligned to the exact position of a call being made. The most advanced variant callers are "haplotype-based callers", which can be configured to consider the context, such as a window, surrounding the call being made.

[0251] A "haplotype" is particular DNA content (nucleotide sequence, list of variants, etc.) on a single common "strand", e.g., one of two diploid strands in a region, and a haplotype-based caller considers the Bayesian implications of which differences are linked by appearing in the same read. Accordingly, a variant calling protocol as proposed herein may implement one or more enhanced functions, such as those performed in a Genome Analysis Toolkit (GATK) haplotype caller and/or using a Hidden Markov Model (HMM) tool and/or a De Bruijn Graph function, such as when one or more of those functions typically employed by a GATK haplotype caller and/or an HMM tool and/or a De Bruijn Graph function may be implemented in software and/or hardware.

[0252] More particularly, as implemented herein, a number of different variant calling operations may be configured to be performed in software or hardware, and may include one or more of the following steps. For example, the variant calling function may include hot region identification, such as to identify locations where multiple reads diverge from the reference, and to generate a window around the identified hot region such that only those regions may be selected for further processing. Additionally, localized haplotype assembly may occur, such as where, for each hot region, all overlapping reads may be assembled into a "De Bruijn graph" (DBG) matrix. From that DBG, a number of paths through the matrix may be extracted, where each path constitutes a candidate haplotype, e.g., hypotheses, for what may be the true DNA sequence on at least one strand. Furthermore, haplotype alignment can occur, such as when each extracted haplotype candidate can be aligned, e.g., Smith-Waterman aligned, back to the reference genome, in order to determine what variation (or variations) from the reference it implies. Furthermore, a read probability calculation can be performed, such as when each read can be tested against each haplotype, or hypothesis, to estimate a probability of observing that the read assuming the haplotype was the true original DNA sampled.

[0253] With respect to such processes, the calculation of read probability will typically be the most resource-intensive and time-consuming operation to perform, often requiring pairwise HMM evaluation. Additionally, the construction of De Bruijn graphs for each accumulation of reads, with associated operations to identify locally and globally unique K-mers, as described below, may also be resource-intensive and/or time-consuming. Accordingly, in various embodiments, one or more of the various calculations involved in performing one or more of these steps may be configured to be implemented in optimized software or hardware form, such as to be performed in an accelerated manner by an integrated circuit, as described herein.

[0254] As indicated above, in various embodiments, a Haplotype caller of the disclosure, implemented in software and/or hardware or a combination thereof, may be configured to include one or more of the following operations: Active Region Identification, Localized Haplotype Assembly, Haplotype Alignment, Read Probability Calculation, and/or Genotyping. For example, the devices, systems, and/or methods of the disclosure may be configured to perform one or more of a mapping, alignment, and/or classification operation on data obtained from the subject's sequenced DNA/RNA to generate mapped, aligned, and/or classified result data. Such result data may then be cleaned, such as by performing a deduplication operation thereon, and/or such data may be communicated to one or more dedicated haplotype caller processing engines to perform a variant calling operation, including one or more of the aforementioned steps, on such result data so as to generate a variant call file with respect thereto. Therefore, all reads that have been sequenced and/or mapped and/or aligned to particular positions in the reference genome can be subjected to further processing in order to determine how the given sequence differs from a reference sequence at any given point in the reference genome.

[0255] Accordingly, in various embodiments, a device, a system, and/or a method of use thereof as disclosed herein may include a variant or haplotype calling system that is implemented in a software and/or wired configuration to perform a hot region identification operation on the obtained result data. Hot region identification involves identifying and determining locations where multiple reads, e.g., in a backlog of reads, diverge from a reference and further involves generating one or more windows around the disagreements (“hot regions”) such that the region within the window may be selected for further processing. For example, during a mapping and/or alignment step, identified reads are mapped and/or aligned to regions in the reference genome where they are expected to have originated in the subject's genetic sequence.

[0256] However, as sequencing is performed in such a manner as to create an oversampling of sequenced reads for any given region of the genome, at any given position in the reference sequence there may be seen an accumulation of any and/or all of the sequenced reads that queue up and align with that region. All of these reads that align and/or overlap at a given cluster region or position may be input into the variant calling system. Thus, for any given read that is analyzed, the read may be compared to the reference at its suspected overlap region, and that read may be compared to the reference to determine whether it shows any differences in its sequence from the known sequence of the reference. If the read aligns to the reference, if any insertions or deletions and all bases are the same, then the alignment is determined to be good.

[0257] Thus, for any given mapped and/or aligned read, the read may have bases that are different from the reference, e.g., the read may include one or more SNPs, creating a position where a base is mismatched; and/or the read may have one or more of an insertion and/or deletion, e.g., creating a gap in the alignment. Consequently, in any of these cases, there will be one or more divergences that need to be considered by further processing. However, to save time and increase efficiency, such further processing should be limited to those cases where a perceived divergence is non-trivial, e.g., a non-noise difference. To determine the significance of a divergence, locations where multiple reads in a pool diverge from the reference can be identified as a hot region, a window around the hot region can then be used to select a locus of disagreement that can then be subjected to further processing. The disagreement, however, must be non-trivial. This can be determined in many ways, for example, the probability of non-reference can be calculated for each locus in question, such as by looking at match quality scores vs baseline divergence, such as above a given threshold considered to be a sufficiently significant amount of indication of those reads that diverge from the reference in a significant way.

[0258] For example, if 30 of all mapped and/or aligned reads align and/or overlap so as to form a cluster at a given position in the reference, e.g., a hot region, and only 1 or 2 of the 30 reads diverge from the reference, then the minimum threshold for further processing may be considered not to have been met, and the non-compliant read (or reads) may be disregarded in view of the 28 or 29 reads that do agree. However, if 3 or 4, or 5, or 10, or more of the reads in the cluster diverge, then the disagreement may be statistically significant enough to warrant further processing, and an active region around the identified region (or regions) of difference should be determined. In such a case, an active region window that determines the bases surrounding this mismatch can be taken to generate enhanced context for the region surrounding the mismatch, and additional processing steps, such as performing a Gaussian distribution and summing non-reference probabilities distributed across neighboring positions, can be performed to investigate and further process this region to see if an active region should be declared and, if so, what, if any, variances from the reference are actually present within this region. Thus, determining an active region identifies those regions where extra processing may be necessary to clearly determine whether a true variance or a misread has occurred.

[0259] In particular, because in many cases it is not desirable to subject every region in a sequence backlog to further processing, an active region may be identified such that it is just that region where extra processing may be needed to clearly determine whether a true variance or a read error has occurred that can be determined to need further processing. And, as indicated above, it may be the size of the variance supported that determines the size of the active region window. For example, in various cases, the boundaries of the active window may range from 1 or 2 or about 10 or 20 or even about 25 or about 50 to about 200 or about 300, or about 500 or about 1,000 bases in length or more, where it is just within the boundaries of the active window that additional processing is occurring. Of course, the size of the active window may be any suitable length so long as it provides the context for determining the statistical significance of a difference.

[0260] Thus, if there are only one or two isolated differences, then the active window may only need to cover one or more to a few dozen bases in the active region in order to have enough context to produce a statistical call that a real variant is present. However, if there is a cluster or many differences, or if there are indels present for which more context is desired, then the window may be set to be longer. In any case, it may be desirable to analyze any and all differences that may occur in clusters, so as to analyze all of them in one or more active regions, because doing so can provide information about each individual difference and will save processing time by decreasing the number of windows involved. In many cases, the boundaries of the active region can be determined by active probabilities passing a given threshold, such as about 0.00001 or about 0.00001 or about 0.0001 or less to about 0.002 or about 0.02 or about 0.2 or more. And if the active region is longer than a given threshold, for example, about 300 to 500 bases or 1,000 bases or more, then the region can be broken into subregions, such as by subregions defined by the locus with the lowest active probability score.

[0261] In various cases, after an active region is identified, a localized haplotype assembly procedure may be performed. For example, within each active region, all accumulated and/or overlapping reads may be assembled into a "De Bruijn Graph" (DBG). A DBG may be a directed graph based on all reads that overlapped the selected active region, wherein the active region may be about 200 or about 300 to about 400 or about 500 bases in length or more, within which active region the presence and/or identity of variants is to be determined. In various cases, as noted above, the active region may be extended, for example, by including another about 100 or about 200 or more bases in each direction from the locus in question so as to generate an extended active region, such as when additional context surrounding a difference may be desired. Consequently, it is from the active region window, extended or not, that all reads that have portions that overlap the active region are accumulated, i.e. to produce an accumulation, the overlapping portions are identified, and the read sequences are input into the haplotype caller system and are thus assembled together in the form of a De Bruin graph, much like the pieces of a jigsaw puzzle.

[0262] Accordingly, for any given active window, there will be reads that form a cluster such that, in bulk, the cluster will include a sequence trajectory through which the overlapping regions of the multiple overlapping reads in the cluster cover the entire sequence within the active window. Thus, at any given locus in the active region, there will be a plurality of reads that overlap that locus, although any given read may not span the entire active region. The result of this is that multiple regions of multiple reads within the cluster are employed by the DBG in determining whether or not a variant is actually present for any given locus in the sequence within the active region. Since it is within the active window in which this determination is being made, it is those portions of any given read within the borders of the active window that are considered, and those portions that fall outside the active window may be discarded.

[0263] As indicated, it is those sections of the readings that overlap the reference within the active region that are fed into the DBG system. The DBG system then assembles the readings like a jigsaw puzzle on a graph, and then for each position in the sequence, it is determined based on the collection of overlapping readings for that position whether there is a match or a mismatch for any given position, and if there is a mismatch, what the probability of that mismatch is. For example, where there are discrete locations where segments of the readings in the accumulation overlap each other, they can be aligned to each other based on their areas of match, and from chaining or stitching the matching readings together as determined by their points of match, it can be established, for each position within that segment, whether and to what extent the readings at any given position match or mismatch each other. Thus, if two or more readings that are compiled align and match each other identically for a period, a graph that has a single chain will result; however, when the two or more readings come into a point of difference, a branch in the graph will be formed, and two or more divergent chains will result, until correspondence between the two or more readings is resumed.

[0264] Thus, trajectories through the graph are often not a straight line. For example, when the k-mers of a read vary from the k-mers of the reference and/or the k-mers of one or more overlapping reads, e.g. in the stack, a "bubble" will form in the graph at the point of difference, resulting in two divergent chains that will continue along two different pathlines until the two sequences are matched again. Each vertex can be assigned a weighted score that identifies how many times the respective k-mers overlap across all reads in the stack. In particular, each trajectory that extends through the generated graph from one end to the other can be assigned a count. And when the same k-mers are generated from a plurality of reads, e.g. where each k-mer has the same sequence pattern, they can be represented in the graph by increasing the count for that trajectory where the k-mer overlaps an already existing k-mer trajectory. Hence, when the same k-mer is generated from a multitude of overlapping reads that have the same sequence, the trajectory pattern across the graph will be repeated again and the count for crossing that trajectory across the graph will be incrementally increased accordingly. In such a case, the pattern is only recorded for the first instance of the k-mer, and the count is incrementally increased for each k-mer that repeats that pattern. In this way, the multiple reads in the accumulation can be collected to determine which variations occur where.

[0265] In such a manner, a graph matrix may be formed by taking all N possible base k-mers, e.g., 10 base k-mers, which may be generated from each read by sequentially traversing the length of the read in ten base segments, where the start of each new base segment is offset by one base relative to the last 10 base segment generated. This procedure may then be repeated by doing the same for each read in the accumulation within the active window. The generated k-mers may then be aligned with each other such that areas of identical correspondence between the generated k-mers correspond to areas where they overlap, so as to construct a data structure, e.g., graph, which may then be scanned and the percentage of correspondence and divergence determined. In particular, the reference and any previously processed k-mers aligned to it can be scanned against the next generated k-mer to determine whether the present generated k-mer matches and/or overlaps with any portion of a previously generated k-mer, and where it shows a match to the present generated k-mer it can then be inserted into the graph at the appropriate position.

[0266] Once constructed, the graph can be scanned and it can be determined based on this correspondence whether any given SNPs and/or indels in the reads relative to the reference are likely to be true variation in the subject's genetic code or the result of a processing or other error. For example, if all or a significant portion of the k-mers from all or a significant portion of all reads in a given region include the same SNP and/or indel divergence but differ from the reference in the same manner, then it can be determined that there is indeed SNP and/or indel variation in the subject's genome compared to the reference genome. However, if only a limited number of k-mers from a limited number of reads evidence the artifact, it is likely to be caused by machine and/or processing and/or other error and not indicative of true variation at the position in question.

[0267] As indicated, when there is a suspicious variance, a bubble will form within the graph. Specifically, when all k-mers within a given region of reads match the reference, they will align in such a way as to form a linear graph. However, when there is a difference between the bases at a given locus, it is at that locus of difference that this graph will branch. This branch can be at any position within the k-mer, and consequently, at that point of difference the 10-base k-mer including that difference will diverge from the rest of the k-mers in the graph. In such a case, a new node, which forms a different trajectory through the graph, will be formed.

[0268] Thus, when everything may have been in agreement, for example, the sequence in the given new k-mer that is graphed is matching the sequence to which it aligns in the graph, up to the point of difference at which the trajectory for that k-mer will match the trajectory for the graph overall and will be linear, but after the point of difference, a new trajectory through the graph will emerge to accommodate the difference represented in the sequence of the newly graphed k-mer. This divergence is represented by a new node within the graph. In such a case, any new k-mers added to the graph that match the newly diverged trajectory will increase the count at that node. Thus, for each read that supports the arc, the count will be incrementally increased.

[0269] In many such cases, the k-mer and/or the read it represents will start matching again, for example after the point of divergence, so that there is now a point of convergence where the k-mers start matching the main trajectory through the graph represented by the k-mers of the reference sequence. For example, naturally after a period of time, the read (or reads) supporting the branched node should rejoin the graph over time. Thus, over time, the k-mers for that read will rejoin the main trajectory again. More particularly, for a SNP at a given locus within a read, the k-mer starting at that SNP will diverge from the main graph and remain separated by about 10 nodes, due to the fact that there are 10 bases per k-mer that overlap that locus of divergence between the read and the reference. Thus, for a SNP at position 11, the k-mers covering that locus within the read will be rejoined to the main trajectory as the exact match is resumed. Consequently, it will take ten shifts for the k-mers from a read that has a SNP at a given locus to be rejoined to the main graph represented by the reference sequence.

[0270] As indicated above, there is typically a path or line or parent strand that is the reference path, and when there is a divergence, a bubble is formed at a node where there is a difference between a read and the parent strand graph. Thus, there are some reads that diverge from the parent strand and form a bubble, where the divergence may be indicative of the presence of a variant. As the graph is processed, bubbles within bubbles within bubbles may be formed along the reference parent strand, such that they are stacked and a plurality of trajectories through the graph may be created. In such a case, there may be a parent pathway represented by the reference parent strand, a pathway from a first divergence, and another pathway from a second divergence within the first divergence, all within a given window, each trajectory through the graph may represent a real variation or may be an artifact such as caused by sequencing error, and/or PCR error, and/or a processing error, and the like.

[0271] Once such a graph has been produced, it needs to be determined which trajectories through the graph represent actual variation present within the sample genome and which are mere artifacts. While it is expected that reads containing handling or machine errors will not be supported by the majority of reads in the sample backlog, however, this is not always the case. For example, errors in PCR processing may typically be the result of a cloning mistake that occurs when preparing the DNA sample; such mistakes tend to result in an insertion and/or a deletion being added to the cloned sequence. Such indel errors may be more consistent between reads, and may result in the generation of multiple reads that have the same error from that PCR cloning mistake. Consequently, a higher line count for such a divergence point may result due to such errors.

[0272] Thus, once a graph matrix has been formed, with many paths through the graph, the next stage must traverse and thus extract all paths through the graph, e.g. left to right. One path will be the main reference chain, but there will be other paths that follow various bubbles along the path. All paths need to be traversed and their count tabulated. For example, if the graph includes a trajectory with a two-level bubble at one point and a three-level bubble at another point, there will be (2 x 3)<6> paths through that graph. Thus, each of the paths will need to be individually extracted, where the extracted paths are referred to as candidate haplotypes. Such candidate haplotypes represent theories that could be truly representative of the actual DNA of the subject that was sequenced, and the following processing steps, including one or more of haplotype alignment, read probability calculation, and/or genotyping, can be employed to test these theories in order to find the probabilities that any and/or each of these theories is correct. Implementing a De Bruijn graph reconstruction therefore represents a way to reliably extract a good set of hypotheses for testing.

[0273] For example, in performing a variant calling function as disclosed herein, an active region identification operation may be implemented, such as to identify locations where multiple reads in a cluster within a given active region diverge from the reference, and to generate a window around the identified active region such that only those regions may be selected for further processing. Additionally, localized haplotype assembly may occur, such as where, for each active region, all overlapping reads in the cluster may be assembled into a "De Bruijn graph" (DBG) matrix. From this DBG, multiple paths through the matrix may be extracted, where each path constitutes a candidate haplotype, e.g., hypotheses, for what may be the true DNA sequence on at least one strand.

[0274] Further, haplotype alignment may occur, such as where each extracted haplotype candidate may be aligned, e.g., Smith-Waterman aligned, back to the reference genome, so as to determine what variation (or variations) from the reference it entails. Further, a read probability calculation may be performed, such as where each read may be tested against each haplotype to estimate a probability of observing that the read assuming the haplotype was the true original DNA sampled. Finally, a genotyping operation may be implemented, and a variant call file produced. As indicated above, any or all of these operations may be configured to be implemented in an optimized manner in software and/or in hardware, and in various cases, due to the resource-intensive and time-consuming nature of constructing a DBG matrix and extracting candidate haplotypes therefrom and/or due to the resource-intensive and time-consuming nature of performing a haplotype alignment and/or a read probability calculation, which may include the involvement of a Hidden Markov Model (HMM) evaluation, these operations (e.g., localized haplotype assembly and/or haplotype alignment and/or read probability calculation) or a portion thereof may be configured to have one or more functions of their operation implemented in a hard-wired fashion, such as to be performed in an accelerated manner by an integrated circuit as described herein. In various cases, these tasks may be configured to be implemented by one or more quantum circuits, such as in a quantum computing device.

[0275] Accordingly, in various instances, the devices, systems, and methods for performing the same may be configured to perform a haplotype alignment and/or a read probability calculation. For example, as indicated, each extracted haplotype may be aligned, such as by Smith-Waterman alignment, back to the reference genome, so as to determine what variation (or variations) from the reference it entails. In various exemplary instances, scoring may occur, such as in accordance with the following exemplary scoring parameters: a match = 20.0; a mismatch = -15.0; a gap opening -26.0; and a gap extension = -1.1, other scoring parameters may be used. Accordingly, in this manner, a CIGAR strand may be generated and associated with the haplotype to produce an assembled haplotype, wherein the assembled haplotype may eventually be used to identify variants. Consequently, in a manner such as this, the probability of a given read being associated with a given haplotype can be calculated for all read/haplotype combinations. In such cases, the probability can be calculated using a Hidden Markov Model (HMM).

[0276] For example, the various assembled haplotypes may be aligned according to a dynamic programming model similar to a SW alignment. In such a case, a virtual matrix may be generated such as when the candidate haplotype, e.g. generated by DBG, may be positioned on one geometric axis of a virtual array, and the read may be positioned on the other geometric axis. The matrix may then be populated with the scores generated by traversing the extracted paths through the graph and calculating the probabilities that any given path is the true path. Thus, in such a case, a difference in this alignment protocol from a typical SW alignment protocol is that, in order to find the most likely path through the array, a maximum likelihood calculation is used, such as a calculation performed by an HMM model that is configured to provide the total probability for alignment of the reads to the haplotype. Thus, an actual CIGAR strand alignment in this case does not need to be produced. Instead, all possible alignments are considered and their probabilities are summed. Pairwise HMM evaluation is resource and time intensive, and thus deploying its operations within a wired configuration within an integrated circuit or via quantum circuits on a quantum computing platform is very advantageous.

[0277] For example, each read may be tested against each candidate haplotype in order to estimate a probability of observing that the read assuming the haplotype is the true representative of the original sampled DNA. In various cases, this calculation may be performed by evaluating a "pairwise hidden Markov model" (HMM), which may be configured to model the various possible ways that the haplotype candidate may have been modified, such as by PCR or sequencing errors, and the like, and variation introduced into the observed read. In such cases, the HMM evaluation may employ a dynamic programming method to calculate the total probability of any series of Markov state transitions arriving at the observed read in light of the possibility that any divergence in the read may be the result of a model error. Accordingly, such HMM calculations may be configured to analyze all possible SNPs and Indels that could have been introduced into one or more of the reads, such as by amplification and/or sequencing artifacts.

[0278] In particular, the pairwise HMM considers in a virtual matrix all possible alignments of the read to the reference candidate haplotypes together with a probability associated with each of them, where all the probabilities are added. The sum of all the probabilities of all the variants along a given path is added to obtain a comprehensive probability for each read. This process is then performed for each pair, for each haplotype, of reads. For example, if there is a cluster of six clusters overlapping a given region, e.g. a region of six haplotype candidates, and if the cluster includes about a hundred reads, then 600 HMM operations will need to be performed. More particularly, if there are 6 haplotypes, then there will be 6 branches through the path and the probability that each is the correct trajectory that corresponds to the subject's actual genetic code for that region needs to be calculated. Consequently, each trajectory for all reads needs to be considered, and the probability for each read that could arrive at this given haplotype must be calculated.

[0279] The Pairwise Hidden Markov Model is an approximate model for how a true haplotype in the sampled DNA can transform into a different possible detected read. These types of transformations have been observed to be a combination of SNPs and Indels that have been introduced into the defined genetic sample by the PCR process, by one or more of the sample preparation steps, and/or by an error caused by the sequencing process and the like. As can be seen from Figure 2, to account for these types of errors, an underlying 3-state model can be employed, such as where: (M = alignment match, I = insertion, D = deletion), additionally when any transition is possible except I <-> D.

[0280] As can be seen from Figure 2, the 3-state base model transitions are not in time sequence, but instead are in a progression sequence through the candidate haplotype and read sequences, starting at position 0 in each sequence, where the first base is position 1. A transition to M implies position +1 in both sequences; a transition to I implies position +1 in the read sequence only; and a transition to D implies position +1 in the haplotype sequence only. The same 3-state model can be configured to base Smith-Waterman and/or Needleman-Wunsch alignments as described herein as well. Consequently, such a 3-state model as presented in this paper can be employed in a SW and/or NW process, thus allowing for affine gap (indel) scoring, where the gap opening (entering state I or D) is assumed to be less likely than the gap extension (remaining in state I or D). Hence, in this case, the pairwise HMM can be viewed as an alignment, and a CIGAR string can be produced to encode a sequence of the various state transitions.

[0281] In many cases, the 3-state base model can be complicated by allowing the transition probabilities to vary by position. For example, the probabilities of all transitions to M can be multiplied by the prior probabilities of observing the next read base given its base quality score and the corresponding next haplotype base. In such a case, the base quality scores can translate into a probability of a sequencing SNP error. When the two bases match, the prior probability is taken as one minus this error probability, and when they disagree, it is taken as the error probability divided by 3, since there are 3 possible SNP outcomes.

[0282] The above discussion is with respect to an abstract "Markovish" model. In many cases, the maximum likelihood transition sequence may also be determined, which is referred to herein as an alignment, and may be performed using a Needleman-Wunsch or other dynamic programming algorithm. But in many cases, in performing a variant calling function as disclosed herein, maximum likelihood alignment, or any particular alignment, need not be a primary concern. Instead, the total probability may be computed, for example, by computing the total probability of observing the given read haplotype, which is the sum of the probabilities of all possible transition paths through the graph from the zero read position at any haplotype position to the final read position at any haplotype position, where each component path probability is simply the product of the various constituent transition probabilities.

[0283] Finding the sum of trajectory probabilities may also be accomplished by employing a virtual array and using a dynamic programming algorithm as described above, such that in each cell of a (0...N) x (0...M) matrix, there are three calculated probability values corresponding to transition states M, D, and I. (Or, equivalently, there are 3 matrices.) The top row (zero read position) of the matrix may be initialized at probability 1.0 in states D, and 0.0 in states I and M; and the remainder of the left column (zero haplotype position) may be initialized all at zeros. (In software, the initial D probabilities may be set close to the maximum double-precision value, e.g., 2A1020, to avoid underflow, but this factor may be normalized later.)

[0284] This 3-to-1 computation dependence restricts the order in which cells can be computed. They can be computed from left to right in each row, processing through the rows from top to bottom, or from top to bottom in each column, processing to the right. Additionally, they can be computed in antidiagonal wavefronts, where the next step must compute all (n,m) cells where n+m is equal to the number of incremental steps. This wavefront order has the advantage that all cells on the antidiagonal can be computed independently of each other. The bottom row of the matrix, then, at the final read position, can be configured to represent the completed alignments. In such a case, the Haplotype Caller will work by summing the probabilities I and M of all cells in the bottom row. In various embodiments, the system may be configured such that no D transitions are allowed within the bottom row, or a D transition probability of 0.0 may be used here, so as to avoid double counting.

[0285] As described herein, in various cases, each HMM evaluation may operate on a pair of sequences, such as a candidate haplotype and a pair of reads. For example, within a given active region, each of a set of haplotypes may be evaluated by HMM vs. each of a set of reads. In such a case, software and/or hardware input bandwidth may be reduced and/or minimized by transferring the set of reads and the set of haplotypes once and letting the software and/or hardware generate the NxM pairwise operations. In certain cases, a Smith-Waterman evaluator may be configured to queue individual HMM operations, each with its own copy of read and haplotype data. A Smith-Waterman (SW) alignment module may be configured to perform a pairwise HMM calculation in linear space or may operate in log-likelihood space. This is useful for maintaining accuracy across a huge range of probability values with fixed point values. However, in other cases, oscillating point operations can be used.

[0286] There are three parallel multiplications (e.g., additions in logarithmic space), then two serial dilutions (approximately 5 to 6 stage approximation segmentations), then an additional multiplication. In such a case, the complete segmentation may be about L = 12 to 16 cycles long. The I and D computations may be about half as long. The segmentation may be fed a multiplicity of probabilities, such as 2 or 3 or 5 or 7 or more input probabilities at each cycle, such as from one or more already computed neighboring cells (M and/or D from left, M and/or I from top, and/or M and/or I and/or D from top left). This may also include one or more haplotype bases and/or one or more read bases, such as with associated parameters, e.g., preprocessed parameters, at each cycle. It outputs the result of M and I and D defined for a cell every cycle, after drop latency.

[0287] As indicated above, in performing a variant calling function as disclosed herein, a De Bruijn Graph can be formulated where when all reads in a pool are identical, the DBG will be linear. However, when there are differences, the graph will form "bubbles" that are indicative of regions of differences that result in multiple paths that diverge from matching the reference alignment and then come together in matching alignment. From this DBG, multiple paths can be extracted that form candidate haplotypes, i.e., hypotheses for which true DNA sequence may be on at least one strand, where the hypotheses can be tested by performing an HMM, or modified HMM, operation on the data. Additionally, a genotyping function can be employed such that possible diploid combinations of the candidate haplotypes can be formed, and for each of these, a conditional probability of observing the entire pool of reads can be calculated. These results can then be fed into a Bayesian formula module to calculate an absolute probability that each genotype is the true one, given the entire observed accumulation of reads.

[0288] Thus, in accordance with the devices, systems, and methods of use described herein, in various instances, a genotyping operation may be performed, wherein the genotyping operation may be configured to be implemented in an optimized manner in software and/or hardware and/or by a quantum processing unit. For example, possible diploid combinations of candidate haplotypes may be formed, and for each combination, a conditional probability of observing the entire read stack may be calculated, such as using the constituent probabilities of observing each read given each haplotype from pairwise HMM evaluation. The results of these calculations are fed into a Bayesian formula to calculate an absolute probability that each genotype is the true genotype given the entire read stack observed.

[0289] Accordingly, in various aspects, the present disclosure is directed to a system for performing a haplotype or variant calling operation on generated and/or provided data so as to produce a variant calling file with respect thereto. Specifically, as described above herein, in particular instances, a variant calling file may be a digital file or other such file that encodes the difference between one sequence and another, such as the difference between a sample sequence and a reference sequence. Specifically, in various instances, the variant calling file may be a text file that presents or otherwise details the genetic and/or structural variations in a person's genetic makeup as compared to one or more reference genomes.

[0290] For example, a haplotype is a set of genetic variations, e.g. DNA and/or RNA, such as polymorphisms, that reside on a person's chromosomes and as such can be passed on to offspring and thus inherited together. In particular, a haplotype may refer to a combination of alleles, e.g. one of a plurality of alternative forms of a gene as may arise by mutation, where the allelic variations are typically found at the same location on a chromosome. Therefore, in determining a person's genome identity, it is important to know which form of several different possible alleles a person's genetic sequence encodes. In particular cases, a haplotype may refer to one or more, e.g. a set, of nucleotide polymorphisms (e.g. SNPs) that may be found at the same position on the same chromosome.

[0291] Typically, in various embodiments, in order to determine the genotype, e.g., allelic haplotypes, for a subject as described herein and above, a software-based algorithm may be involved, such as an algorithm that employs a haplotype calling program, e.g., GATK, to simultaneously determine SNPs and/or insertions and/or deletions, i.e., indels, in the genetic sequence of an individual. In particular, the algorithm may involve one or more haplotype assembly protocols, such as for local de novo assembly of a haplotype in one or more active regions of the genetic sequence being processed. Such processing typically involves the implementation of a processing function called a Hidden Markov Model (HMM), which is a stochastic and/or statistical model used to model randomly varying systems, such as when it is assumed that future states within the system depend only on the present state and not on the sequence of events preceding it.

[0292] In such cases, the system that is modeled has features or is otherwise assumed to be a Markov process with unobserved (hidden) states. In particular cases, the model may involve a simple dynamic Bayesian network. In particular, with respect to determining genetic variation, in its simplest form, there is one of four possibilities for the identity of any given base in a sequence that is processed, such as when comparing a segment of a reference sequence, e.g., a hypothetical haplotype, and that of a subject's DNA or RNA, e.g., a read derived from a sequencer. However, in order to determine such variation, in the first instance, a subject's DNA/RNA needs to be sequenced, e.g., by means of a Next Generation Sequencer ("NGS"), to produce a read or "reads" that identify the subject's genetic code. Next, once the subject's genome has been sequenced to produce one or more reads, the multiple reads, representative of the subject's DNA and/or RNA, need to be mapped and/or aligned, as described in more detail above in this document. The next step in the process, then, is to determine how the subject's genes that have been determined, for example, to have been mapped and/or aligned, vary from that of a prototypic reference sequence. In performing such an analysis, therefore, it is assumed that the read that potentially represents a given gene from a subject is a representation of the prototypic haplotype, albeit with multiple SNPs and/or indels that are currently to be determined.

[0293] Specifically, in particular aspects, devices, systems, and/or methods for practicing the same are provided, such as for performing a haplotype and/or variant calling function, such as by implementing an HMM function, e.g., in an accelerated haplotype caller. In various instances, in order to overcome these and other such various problems in the art, the HMM accelerator disclosed herein may be configured to be operated in a manner such as to be software-implemented, hardware-implemented, or a combination of being implemented and/or otherwise controlled in part by software and/or in part by hardware, and/or may include quantum computing implementations. For example, in a particular aspect, the disclosure is directed to a method by which data regarding the DNA and/or RNA sequence identity of a subject and/or how the subject's genetic information may differ from that of a reference genome may be determined.

[0294] In such a case, the method may be performed by implementing a haplotype or variant calling function, such as by employing an HMM protocol. In particular, the HMM function may be performed in hardware, software, or via one or more quantum circuits, such as in an accelerator device, in accordance with a method described herein. In such a case, the HMM accelerator may be configured to receive and process the sequenced, mapped, and/or aligned data, to process the same, for example, to produce a variant calling file, as well as to transmit the processed data back throughout the system. Accordingly, the method may include employing a system in which data may be sent from a processor, such as a CPU or GPU or even a software-controlled QPU, to a haplotype caller that deploys an accelerated HMM, wherein the haplotype caller may be deployed on a microprocessor chip, such as an FPGA, ASIC, or ASIC structured or deployed by one or more quantum circuits. The method may further include steps for processing the data to produce HMM result data, which results may then be fed back to the CPU and/or GPU and/or QPU.

[0295] Particularly, in one embodiment, as can be seen with respect to Figure 3A, a bioinformatics segmentation system that includes an HMM accelerator is provided. For example, in one case, the bioinformatics segmentation system can be configured as a variant calling system 1. The system is illustrated as being implemented in hardware, but can also be implemented via one or more quantum circuits, such as a quantum computing platform. Specifically, Figure 3A provides a high-level view of an HMM interface structure. In particular embodiments, the variant calling system 1 is configured to accelerate at least a portion of a variant calling operation, such as an HMM operation. Thus, in various cases, the variant calling system may be referred to herein as an HMM system 1. The system 1 includes a server having one or more central processing units (CPU/GPU/QPU) 1000 configured to perform one or more routines related to sequencing and/or processing genetic information, such as comparing a sequenced genetic sequence to one or more reference sequences.

[0296] Additionally, system 1 includes a peripheral device 2, such as an expansion card, which includes a microchip 7, such as an FPGA, ASIC or sASIC. In some cases, one or more quantum circuits may be provided and configured to perform the various operations presented herein. It should also be understood that the term ASIC may refer to a structured ASIC (sASIC), where appropriate. The peripheral device 2 includes an interconnect 3 and a bus interface 4, such as a parallel or serial bus, that connects the CPU/GPU/QPU 1000 to the chip 7. For example, the device 2 may comprise a peripheral component interconnect, such as a PCI, PCI-X, PCIe, or QPI (Quick Interconnect), and may include a bus interface 4, which is adapted to operatively and/or communicatively connect the CPU/GPU/QPU 1000 to the peripheral device 2, such as for high data transfer rates and low latency. Accordingly, in particular cases, the interface may be a peripheral component interconnect express (PCIe) 4 that is associated with the microchip 7, wherein the microchip includes an HMM accelerator 8. For example, in particular cases, the HMM accelerator 8 is configured to perform an accelerated HMM function, such as where the HMM function, in certain embodiments, may be at least partially implemented in the hardware of the FPGA, AISC, or sASIC or via one or more suitably configured quantum circuits.

[0297] Specifically, Figure 3A presents a high-level figure of an HMM accelerator 8 having an exemplary organization of one or more engines 13, such as a plurality of processing engines 13a through 13m+i, for performing one or more processes of a variant call function, such as including an HMM task. Accordingly, the HMM accelerator 8 may be composed of a data distributor 9, e.g., CentCom, and one or a plurality of processing clusters 11 through 11n+i that may be organized or otherwise include one or more instances 13, such as wherein each instance may be configured as a processing engine, such as a small engine 13a through 13m+1. For example, distributor 9 may be configured to receive data, such as from CPU/GPU/QPU 1000, and distribute or otherwise transfer that data to one or more of the plurality of HMM processing clusters 11.

[0298] Particularly, in certain embodiments, the distributor 9 may be positioned locally between the integrated PCIe interface 4 and the HMM accelerator module 8, such as when the interface 4 communicates with the distributor 9, such as via an interconnect or other suitably configured bus 5, e.g., PCIe bus. The distributor module 9 may be adapted to communicate with one or more HMM accelerator clusters 11, such as via one or more cluster buses 10. For example, the HMM accelerator module 8 may be configured as or otherwise include an arrangement of clusters 11a through 11n+1, such as wherein each HMM cluster 11 may be configured as or otherwise include a cluster cube 11, and/or may include one or more instances 13, wherein the instance may be configured as a processing engine 13 that is adapted to perform one or more operations on data received thereby. Accordingly, in various embodiments, each cluster 11 may be formed as or otherwise include a cluster cube 11a through 11n+1, wherein each of the cubes may be operationally associated with multiple HMM accelerator engine instances 13a through 13m+1, such as wherein each cluster cube 11 may be configured to direct data to a plurality of processing engines 13a through 13m+1 within the cluster 11.

[0299] In various instances, the HMM accelerator 8 is configured to compare each base of a subject's sequenced genetic code, such as in read format, to the various known candidate haplotypes generated from a reference sequence and determine the probability that any given base at a position that is considered matches or does not match the relevant haplotype, e.g., the read includes a SNP, an insertion, or a deletion, thereby resulting in a variation of the base at the position that is considered. Particularly, in various embodiments, the HMM accelerator 8 is configured to assign transition probabilities to the sequence of bases of the read passing between each of these states, Match ("M"), Insertion ("I"), or Deletion ("D"), as described in more detail below herein.

[0300] More particularly, depending on the configuration, the HMM acceleration function may be implemented in software, such as by the CPU/GPU/QPU 1000 and/or microchip 7, and/or may be implemented in hardware and may be present within the microchip 7, such as positioned on the peripheral expansion card or board 2. In various embodiments, this functionality may be implemented partially as software, e.g., executed by the CPU/GPU/QPU 1000, and partially as hardware, implemented on the chip 7 or via one or more quantum processing circuits. Accordingly, in various embodiments, the chip 7 may be present on the motherboard of the CPU/GPU/QPU 1000 or may be part of the peripheral device 2, or both. Accordingly, the HMM accelerator module 8 may include or be otherwise associated with multiple interfaces, e.g., 3, 5, 10, and/or 12, so as to allow efficient transfer of data to and from the processing engines 13.

[0301] Accordingly, as can be seen with respect to Figures 2 and 3, in various embodiments, a microchip 7 configured to perform a variant, e.g., haplotype, calling function is provided. The microchip 7 may be associated with a CPU/GPU/QPU 1000, such as directly coupled thereto, e.g., included on the motherboard of a computer, or indirectly coupled thereto, such as being included as part of a peripheral device 2 that is operatively coupled to the CPU/GPU/QPU 1000, such as through one or more interconnects, e.g., 3, 4, 5, 10, and/or 12. In this case, the microchip 7 is present in the peripheral device 2. It should be understood that, although configured as a microchip, the accelerator could also be configured as one or more quantum circuits of a quantum processing unit, wherein the quantum circuits are configured as one or more processing engines for performing one or more of the functions disclosed herein.

[0302] Thus, the peripheral device 2 may include a parallel or serial expansion bus 4, such as for connecting the peripheral device 2 to the central processing unit (CPU/GPU/QPU) 1000 of a computer and/or server, such as via an interface 3, e.g., DMA. In particular cases, peripheral device 2 and/or serial expansion bus 4 may be a Peripheral Component Interconnect Express (PCIe) that is configured to communicate with or otherwise include microchip 7, such as via connection 5. As described herein, microchip 7 may be at least partially configured as or may otherwise include an HMM accelerator 8. HMM accelerator 8 may be configured as part of microchip 7, e.g., as hardware and/or as code to be executed in association therewith, and is configured to perform a variant call function, such as to perform one or more operations of a Hidden Markov Model, on data provided to microchip 7 by CPU/GPU/QPU 1000, such as via PCIe interface 4. Similarly, once one or more variant call functions have been performed, e.g., one or more HMM operations have been executed, the results thereof may be transferred from the HMM accelerator to the microchip 7. 8 of chip 7 via bus 4 to CPU/GPU/QPU 1000, such as via connection 3.

[0303] For example, in particular cases, a CPU/GPU/QPU 1000 for processing and/or transferring information and/or executing instructions is provided together with a microchip 7 that is at least partially configured as an HMM accelerator 8. The CPU/GPU/QPU 1000 communicates with the microchip 7 through an interface 5 that is adapted to facilitate communication between the CPU/GPU/QPU 1000 and the HMM accelerator 8 of the microchip 7 and therefore can communicatively connect the CPU/GPU/QPU 1000 to the HMM accelerator 8 that is part of the microchip 7. To facilitate these functions, the microchip 7 includes a distributor module 9, which may be a CentCom, that is configured to transfer data to a plurality of HMM engines 13, for example, via one or more clusters 11, wherein each engine 13 is configured to receive and process the data, such as by executing a protocol HMM therein, computing the final values, outputting the results thereof, and repeating the same. In various cases, performance of an HMM protocol may include determining one or more transition probabilities, as described below herein. In particular, each HMM engine 13 may be configured to perform work, such as that includes one or more of the generation and/or evaluation of a virtual HMM matrix to produce and output a final sum value with respect thereto, wherein the final sum expresses the probable likelihood that the called base matches or is different from a corresponding base in a hypothetical haplotype sequence, as described below herein.

[0304] Figure 3B presents a detailed representation of the HMM cluster 11 of Figure 3A. In various embodiments, each HMM cluster 11 includes one or more HMM instances 13. One or more clusters may be provided, as desired in accordance with the amount of resources provided, such as in the quantum computing chip or processor. In particular, an HMM cluster may be provided, wherein the cluster is configured as a cluster hub 11. The cluster hub 11 takes data pertaining to one or more jobs 20 from the distributor 9, and is further communicatively connected to one or more, e.g., a plurality of, HMM instances 13, such as via one or more HMM instance buses 12, to which the cluster hub 11 transmits the job data 20.

[0305] The bandwidth for data transfer throughout the system may be a relatively low bandwidth process, and once a job 20 is received, the system 1 may be configured to complete the job, such as having it exit the chip 7 to memory. In various embodiments, a job 20a is sent to a processing engine 13a at any given time, but multiple jobs 20a-n may be distributed across the clustering cube 11 to multiple different processing engines 13a-13m+1, such that each of the processing engines 13 will be working on a single job 20, e.g., a single comparison between one or more reads and one or more haplotype sequences, in parallel and at high speeds. As described below, performance of such a job 20 may typically involve generating a virtual array so that the subject's “read” sequences can be compared to one or more, e.g., two, hypothetical haplotype sequences in order to determine differences between them. In such cases, a single job 20 may involve processing one or more arrays that have a multiplicity of cells therein that need to be processed for each comparison that is performed, such as in a base-by-base manner. Since the human genome has about 3 billion base pairs, there may be on the order of 1 to 2 billion different jobs to be performed when analyzing a 30X oversampling of a human genome (which is equated to about 20 trillion cells in the arrays of all associated HMM jobs).

[0306] Accordingly, as described herein, each HMM instance 13 may be adapted to perform an HMM protocol, e.g., generating and processing an HMM matrix, on sequence data, such as data received thereby from the CPU/GPU/QPU 1000. For example, as explained above, in sequencing a subject's genetic material, such as DNA or RNA, the DNA/RNA is broken into segments, such as up to about 100 bases in length. The identity of these 100-base segments is then determined, such as by an automated sequencer, and the "read" is output to a FASTQ or other text-based file that stores each base identity of the read along with a Phred quality score (e.g., typically a number between 0 and 63 on a logarithmic scale, where a score of 0 indicates the least confidence that the base called is correct, with scores between 20 and 45 generally being acceptable as relatively accurate).

[0307] In particular, as indicated above, a Phred quality score is a quality indicator that measures the quality of the identification of nucleobase identities generated by the sequencing processor, e.g., by the automated DNA/RNA sequencer. Thus, each base in a read includes its own quality score, e.g., Phred, based on which the sequencer has assessed the quality of that particular identification. Phred represents the confidence with which the sequencer estimates that it has the correct base identity called. This Phred score is then used by the deployed HMM module 8, as described in detail below, to further determine the accuracy of each base called in the read compared to the haplotype to which it was mapped and/or aligned, such as by determining its Match, Insertion, and/or Deletion transition probabilities, e.g., into and out of the Match state. It should be appreciated that, in various embodiments, the system 1 may modify or otherwise adjust the initial Phred score prior to performance of an HMM protocol thereon, such as by considering neighboring bases/scores and/or fragments of neighboring DNA and allowing such factors to influence the Phred score of the base, e.g., cell, under examination.

[0308] In such cases, as can be seen with respect to Figure 4, the system 1, e.g., computer/quantum software, can determine and identify active regions 500n within the sequenced genome that can be explored and/or otherwise subjected to further processing as described herein, which can be broken down into jobs 20n that can be parallelized across the various available cores and chains 1007 throughout the system 1. For example, such active regions 500 can be identified as being regions of variation between the sequenced and reference genomes. In particular, the CPU/GPU/QPU 1000 may have multiple threads 1007 executing, identifying active regions 500a, 500b, and 500c, compiling and aggregating multiple different jobs 20n to be executed, e.g., via a suitably configured aggregator 1008, based on the active region(s) 500a through c that are currently being examined. Any suitable number of threads 1007 may be employed so as to allow the system 1 to operate at maximum efficiency, e.g., the more threads present, the less active time is spent waiting.

[0309] Once identified, compiled and/or aggregated, the chains 1007/1008 will then transfer the active jobs 20 to the data distributor 9, e.g., CentCom, of the HMM module 8, such as via the PCIe interface 4, for example, in a fire and forget manner and will then be moved to a different process while waiting for the HMM 8 to send the output data back so as to be matched back to the corresponding active region 500 to which it maps and/or aligns. The data distributor 9 will then distribute the jobs 20 to the various different HMM clusters 11, such as in a job-by-job manner. If everything is working effectively, this may be in a first-in, first-out format, but this need not be the case. For example, in various embodiments, jobs and raw data and processed job results data may be sent through and through the system as they become available.

[0310] In particular, as can be seen with respect to FIGS. 2, 3 and 4, data from multiple jobs 20 may be aggregated into 4K byte pages of data, which may be sent via PCIe 4 to and through CentCom 9 and processing engines 13, e.g., via clusters 11. The amount of data that is sent may be greater or less than 4K bytes, but will typically include about 100 HMM jobs per 4K page (e.g., 1,024) of data. In particular, this data is then digested by data distributor 9 and fed to each cluster 11, such as where a 4K page is sent to a cluster 11. However, this need not be the case since any given job 20 may be sent to any given cluster 11, based on which clusters become available and when.

[0311] Accordingly, the clustering approach 11 as presented herein efficiently distributes incoming data to the processing engines 13 at high speed. Specifically, as data arrives at the PCIe interface 4 from the CPU/GPU/QPU 1000, e.g., via the DMA connection 3, the received data may then be sent via the PCIe bus 5 to the CentCom distributor 9 of the variant calling microchip 7. The distributor 9 then sends the data to one or more HMM processing clusters 11, such as via one or more cluster dedicated buses 10, wherein the cluster 11 may then transmit the data to one or more processing instances 13, e.g., via one or more instance buses 12, such as for processing. In this case, the PCIe interface 4 is adapted to deliver data over the peripheral expansion bus 5, distributor 9, and/or cluster buses 10 and/or instance 12 at a fast rate, such as at a rate that can keep one or more, e.g., all, of the HMM accelerator instances 13a-(m+1) within one or more, e.g., all, of the HMM clusters 11a-(n+1) busy, such as for an extended period of time, e.g., all the time during which the system 1 is running, the jobs 20 are being processed, and while also keeping track of the output of the processed HMM data that is to be sent back to the one or more CPUs 1000 via the PCIe interface 4.

[0312] For example, any inefficiency in interfaces 3, 5, 10, and/or 12 that leads to idle time for one or more of the HMM accelerator instances 13 may add directly to the overall processing time of system 1. In particular, when analyzing a human genome, there may be on the order of two or more billion different jobs 20 that need to be distributed to the various HMM clusters 11 and processed over the course of a period of time, such as less than 1 hour, less than 45 minutes, less than 30 minutes, less than 20 minutes, including 15 minutes, 10 minutes, 5 minutes, or less.

[0313] Accordingly, Figure 4 presents an overview of an exemplary data flow throughout the software and/or hardware of system 1, as generally described above. As can be seen with respect to Figure 4, system 1 may be configured in part to transfer data, such as between PCIe interface 4 and distributor 9, e.g., CentCom, such as via PCIe bus 5. Additionally, system 1 may be further configured in part to transfer received data, such as between distributor 9 and one or more HMM clusters 11, such as via one or more cluster buses 10. Thus, in various embodiments, HMM accelerator 8 may include one or more clusters 11, such as one or more clusters 11 configured to perform one or more processes of an HMM function. In such a case, there is an interface, such as a cluster bus 10, that connects CentCom 9 to HMM cluster 11.

[0314] For example, Figure 5 is a high-level diagram depicting the interface into and out of the HMM module 8, such as into and out of a cluster module. As can be seen with respect to Figure 6, each HMM cluster 11 may be configured to communicate with, e.g., receive data from, and/or send final result data, e.g., sum data, to the CentCom data distributor 9 via a dedicated cluster bus 10. In particular, any suitable interface or bus 5 may be provided so long as it allows the PCIe interface 4 to communicate with the data distributor 9. More particularly, the bus 5 may be an interconnect that includes interpretation logic useful in communicating with the data distributor 9, wherein the interpretation logic may be configured to accommodate any protocol employed to provide its functionality. Specifically, in various instances, the interconnect may be configured as a PCIe bus 5.

[0315] Additionally, cluster 11 may be configured such that a single or multiple clock domains may be employed therein, and thus one or more clocks may be present within cluster 11. In particular cases, multiple clock domains may be provided. For example, a slower clock may be provided, such as for communications, e.g., to and from cluster 11. Additionally, a faster clock, e.g., a high speed, may be provided, which may be employed by the HMM instances 13 for use in performing the various calculations described herein.

[0316] Particularly, in various embodiments, as can be seen with respect to Figure 6, the system 1 may be configured such that, in a first instance, as the data distributor 9 leverages the existing CentCom IP, a collar, such as a joint, may be provided, wherein the joint is configured to translate signals to and from the CentCom interface 5 to and from the HMM cluster interface or bus 10. For example, an HMM cluster bus 10 may communicatively and/or operationally connect the CPU/GPU 1000 to the various clusters 11 of the HMM accelerator module 8. Thus, as can be seen with respect to Figure 6, structured read and/or write data for each haplotype and/or for each read may be sent throughout the system 1.

[0317] After a job 20 is input to the HMM engine, an HMM engine 13 may typically start: a) immediately, if it is IDLE, or b) after it has completed its currently assigned task. It should be appreciated that each HMM accelerator engine 13 may handle ping and pong inputs (e.g., it may be working on one data set while another is being loaded), thereby minimizing downtime between jobs. Additionally, the HMM cluster collar 11 may be configured to automatically take the input job 20 sent by the data distributor 9 and assign it to one of the HMM engine instances 13 in the cluster 11 that can receive new work. There need not be a control on the software side that can select a specific HMM engine instance 13 for a specific job 20. However, in various cases, the software may be configured to control such instances.

[0318] Accordingly, in view of the foregoing, the system 1 can be continuously transmitted when transferring the result data back to the CPU/GPU/QPU and due to this efficiency, not much data needs to go back to the CPU/GPU/QPU to achieve the usefulness of the results. This allows the system to achieve a variant call operation of about 30 minutes or less, such as about 25 or about 20 minutes or less, for example, about 18 or about 15 minutes or less, including about 10 or about 7 minutes or less, even about 5 or about 3 minutes or less, depending on the system configuration.

[0319] Figure 6 presents a high-level view of various functional blocks within an exemplary HMM engine 13 within a hardware accelerator 8, in the FPGA or ASIC 7. Specifically, within the hardware HMM accelerator 8, there are multiple clusters 11, and within each cluster 11 there are multiple engines 13. Figure 6 presents a single instance of an HMM engine 13. As can be seen with respect to Figure 6, the engine 13 may include an instance bus interface 12, a plurality of memories, e.g., an HMEM 16 and an RMEM 18, various other components 17, HMM control logic 15, as well as a result output interface 19. Particularly, on the engine side, the HMM instance bus 12 is operatively connected to the memories, HMEM 16 and RMEM 18, and may include interface logic that communicates with the cluster hub 11, wherein the hub is in communication with the distributor 9, which in turn is in communication with the PCIe interface 4 which communicates with the variant calling software that is executed by the CPU/GPU and/or the server 1000. The HMM instance bus 12 therefore receives the data from the CPU 1000 and loads it into one or more of the memories, e.g., the HMEM and RMEM. This configuration may also be implemented in one or more quantum circuits and adapted accordingly.

[0320] In such cases, sufficient memory space should be allocated so that at least one or two or more haplotypes, e.g., two haplotypes, can be loaded into, e.g., the HMEM 16, per given read sequence that is loaded into, e.g., the RMEM 18, whereby when multiple haplotypes are loaded, it results in an alleviation of the load on the PCIe bus bandwidth 5. In particular cases, two haplotypes and two read sequences can be loaded into their respective memories, which could allow the four sequences to be processed together in all relevant combinations. In other cases, four, or eight, or sixteen sequences, e.g., pairs of sequences, can be loaded and, in a similar manner, be processed in combination, so as to further alleviate bandwidth when desired.

[0321] Additionally, sufficient memory may be reserved so that a ping-pong structure may be implemented therein so that once the memories are loaded with a new job 20a, such as on the ping side of the memory, a new job signal is indicated, and the control logic 15 may begin processing the new job 20a, such as generating the matrix and performing the requirement calculations, as described herein and below. Consequently, this leaves the pong side of the memory available to be loaded with another job 20b, which may be loaded therein while the first job 20a is being processed, so that as the first job 20a is completed, the second job 20b may immediately begin to be processed by the control logic 15.

[0322] In such a case, the matrix for job 20b may be preprocessed so that there is virtually no downtime, e.g., one or two clock cycles, from the end of processing of the first job 20a to the beginning of processing of the second job 20b. Thus, by utilizing the ping and pong side of the memory structures, the HMEM 16 can typically store 4 haplotype sequences, e.g., two per part, and the RMEM 18 can typically store 2 read sequences. This ping-pong configuration is useful in that it simply requires a little extra memory space, but allows a doubling of the throughput of the engine 13.

[0323] During and/or after processing, memories 16, 18 are fed to the transition probability calculator and lookup table (LUT) block 17a, which is configured to calculate various information related to "prior" data, as explained below, which in turn feeds the prior result data to the M, I, and D state calculator block 17b for use in calculating the transition probabilities. One or more scratchpad RAMs 17c may also be included, such as to hold the M, I, and D states at the edge of the range, e.g., the values of the bottom row of the processing range, which, as indicated, in various cases may be any suitable number of cells, e.g., about 10 cells, in length so as to be proportionate to the length of the range 35.

[0324] Additionally, a separate results output interface block 19 may be included so that when the sums are completed, the sums, e.g., a 4 32-bit word, may be immediately transmitted back to the calling software of the CPU/GPU/QPU 1000 variant. It will be appreciated that this configuration may be adapted so that the system 1, specifically the M, I, and D calculator 17b, is not kept waiting for the output interface 19 to be released, for example, so long as it does not take as long to release the results as it does to perform the work 20. Thus, in this configuration, there may be three segmentation steps working together to perform general system segmentation, such as loading memory, performing the MID calculations, and outputting the results. Furthermore, it will be noted that any given HMM engine 13 is one of many with its own output interface 19, however, they may share a common interface 10 back to the data distributor 9. Thus, the clustering cube 11 will include management capabilities to manage the transfer ("xfer") of information through the HMM accelerator 8 so as to avoid collisions.

[0325] Accordingly, the following details the processes that are performed within each module of the HMM engines 13 as it receives the haplotype and read sequence data, processes the same, and outputs result data related thereto, as generally illustrated above. Specifically, the high bandwidth computations in the HMM engine 13, within the HMM cluster 11, are directed to computing and/or updating the match (M), insertion (I), and deletion (D) status values, which are employed to determine whether the particular read being examined matches the haplotype reference as well as the extent thereof, as described above. In particular, the read along with the Phred score and GOP value for each base in the read is transmitted to the cluster 11 from the distributor 9 and is thereby assigned to a particular processing engine 13 for processing. This data is then used by the M, I, and D calculator 17 of the processing engine 13 to determine whether the base called in the read is more or less likely to be correct and/or to match its respective base in the haplotype, or to be the product of a variation, e.g., an insertion or deletion; and/or, if there is a variation, whether such variation is the likely result of true variability in the haplotype or, instead, an artifact of an error in the sequence generation and/or mapping and/or alignment systems.

[0326] As indicated above, a part of such an analysis includes the MID calculator 17 determining the transition probabilities from one base to another in the read that passes from one M, I, or D state to another compared to the reference, such as from a match state to another match state, or a match state to an insertion state or to a deletion state. To make such determinations, each of the associated transition probabilities is determined and considered in evaluating whether any observed variation between the read and the reference is a true variation and not merely a machine or processing error. For these purposes, the Phred score for each base that is considered is useful in determining the transition probabilities into and out of the match state, such as passing from a match state to an insertion or deletion state, e.g., gapped, in the comparison. Similarly, the transition probabilities of continuing from a gapped state, e.g., an insertion or deletion state, back to a match state are determined. In particular cases, the probabilities of entering or exiting the deletion or insertion state, e.g., exiting a gap continuation state, may be a fixed value and may be referred to herein as the gap continuation probability or penalty. However, in various cases, such gap continuation penalties may be oscillating and thus subject to changing dependence on the accuracy demands of the system configuration.

[0327] Accordingly, as depicted with respect to Figures 7 and 8 , each of the M, I, and D state values is computed for each possible read and haplotype base pairing. In such a case, a virtual array 30 of cells containing the read sequence that is evaluated on one axis of the array and the associated haplotype sequence on the other axis may be formed, such that each cell in the array represents a base position in the read and haplotype reference. Thus, if the read and haplotype sequences are each 100 bases in length, the array 30 will include 100 by 100 cells, a given portion of which may need to be processed in order to determine the likelihood and/or the extent to which that particular read matches that particular reference. Thus, once virtually formed, the array 30 can then be used to determine the various state transitions that occur when moving from one base in the read sequence to another and compare them to that of the haplotype sequence, as depicted in Figures 7 and 8. Specifically, the processing engine 13 is configured so that a plurality of cells can be processed in parallel and/or sequentially when traversing the array with the control logic 15. For example, as depicted in Figure 7, a virtual processing track 35 is propagated and moves across the array 30, such as from left to right, processing individual cells of the array 30 across the diagonal from right to left.

[0328] More specifically, as can be seen with respect to Figure 7, each individual virtual cell within the matrix 30 includes a state value of M, I, and D that needs to be calculated to assess the nature of the identity of the called base, and as depicted in Figure 7, the data dependencies for each cell in this process can be clearly seen. Thus, to determine a given M state of a present cell that is processed, the Match, Insert, and Delete states of the cell diagonally above the present cell need to be inserted into the present cell and used in calculating the M state of the cell that is presently calculated (e.g., thus, progression down diagonally, forward through the matrix is indicative of matching).

[0329] However, to determine the state of I, only the Match and Insert states for the cell directly above the present cell need to be inserted into the present cell that is processed (thus the vertical downward "gap" progression when continuing in an insert state). Similarly, to determine the state of D, only the Match and Delete states for the cell directly to the left of the present cell need to be inserted into the present cell (thus the horizontal transverse "gap" progression when continuing in a delete state). As can be seen from Figure 7, after computation of cell 1 (the shaded cell in the topmost row) begins, processing of cell 2 (the shaded cell in the second row) can begin as well, without waiting for any results from cell 1, because there are no data dependencies between that cell in row 2 and the cell in row 1 when processing begins. This forms a reverse diagonal 35 in which processing proceeds downward and to the left, as shown by the red arrow. This 35-reverse diagonal processing approach increases the processing efficiency and throughput of the overall system. Similarly, data generated in cell 1 can be immediately advanced to the cell below and ahead to the right of the topmost cell 1, thus advancing the 35-track forward.

[0330] For example, Figure 7 depicts an exemplary HMM matrix structure 35 that shows the hardware processing flow. The matrix 35 includes the haplotype base index, e.g., containing 36 bases, positioned to run along the upper edge of the horizontal axis, and further includes the read base index, e.g., 10 bases, positioned to extend along the lateral edge of the vertical axis in such a manner as to form a cell structure, in which a selection of the cells may be populated with a state probability of M, I, and D, and transition probabilities of transitioning from the present state to a neighboring state. In such a case, as described in more detail below, a movement from the match state to a match state results in a forward diagonal progression through the matrix 30, while a movement from a match state to an insertion state results in a vertical downward processing gap, and a movement from a match state to a deletion state results in a horizontal progression gap. Therefore, as represented in Figure 8, for a given cell, when determining the matching, insertion, and deletion states for each cell, the matching, insertion, and deletion probabilities of its three contiguous cells are employed.

[0331] The downward split in Figure 7 represents the parallel and sequential nature of the processing engine (or engines) that is configured to produce a processing band or wave 35 that moves progressively along the virtual matrix in accordance with the data dependencies, set out in Figures 7 and 8, to determine the M, I, and D states for each particular cell in the matrix 30. Accordingly, in certain cases, it may be desirable to calculate the identities of each cell in a downward and diagonal manner, as explained above, rather than simply calculating each cell along a vertical or horizontal axis exclusively, although this may be done if desired. This is due to the increased wait time, e.g., latency, that might be required when processing the virtual cells of the matrix 35 individually or sequentially along the vertical or horizontal axis only, such as by means of the hardware configuration.

[0332] For example, in such a case, when moving linearly and sequentially through virtual matrix 30, such as in a row-by-row or column-by-column manner, in order to process each new cell, the state computations of each previous cell would have to be completed, thus increasing the overall latency time. However, by propagating the M, I, D probabilities of each new cell in a downward and diagonal fashion, system 1 does not have to wait for the processing of its previous cell, e.g., cell one, to be completed before beginning processing of a contiguous cell in row two of the matrix. This allows parallel and sequential processing of cells in a diagonal arrangement to occur and further allows the various computational segmentation delays associated with the M, I, and D state computations to be hidden. Consequently, as track 35 moves through matrix 30 from left to right, the computational processing moves diagonally downward, e.g., toward the left (as shown by the arrow in Figure 7). This configuration can be particularly useful for quantum hardware and/or circuit deployments, such as where memory and/or clock-to-clock latency are a primary concern.

[0333] In such configurations, the actual value output of each invocation of an HMM engine 13, for example after having calculated the entire matrix 30, may be a bottom row (e.g., Row 35 of Figure 21) containing M, I, and D states, wherein the M and I states may be summed (the D states may be ignored at this point having already served their function in processing the above calculations) so as to produce a final sum value which may be a single probability that estimates, for each read index and haplotype, the probability of observing the read, for example assuming that the haplotype was the true original DNA sampled.

[0334] In particular, the result of processing array 30, for example, of Figure 7, may be a single value that represents the probability that the read is a true representation of that haplotype. This probability is a value between 0 and 1 and is formed by summing all the M and I states of the bottom row of cells in HMM array 30. Essentially, what is being assessed is the possibility that something could go wrong in the sequencer, or associated DNA preparation methods prior to sequencing, so as to incorrectly produce a mismatch, insertion or deletion in the read that is not actually present within the subject's genetic sequence. In such a case, the read is not a true reflection of the subject's actual DNA.

[0335] Thus, by taking such production errors into account, it can be determined what any given read actually represents relative to the haplotype, and thus allow the system to better determine how the subject's genetic sequence, e.g., en masse, may differ from that of a reference sequence. For example, many haplotypes can be run against many read sequences, generating scores for all of them, and determining, based on which matches have the best scores, what the individual's actual genomic sequence identity is and/or how it actually varies relative to a reference genome.

[0336] More particularly, Figure 8 depicts an enlarged view of a portion of the HMM state matrix 30 of Figure 7. As shown in Figure 8, given the internal composition of each cell in the matrix 30, as well as the structure of the matrix as a whole, the probability of the M, I, and D states for any given "new" cell that is computed is dependent on the M, I, and D states of several of its surrounding neighbors that have already been computed. In particular, as shown in more detail with respect to Figures 1 and 16, in an exemplary configuration, there may be an approximately 0.9998 probability of moving from one match state to another match state, and there is only a 0.0001 probability (open gap penalty) of moving from a match state to an insertion or deletion state, e.g., with a gap. Furthermore, when in a gapped insertion or gapped deletion state, there may be only a 0.1 probability (gap extension or gap continuation penalty) of remaining in the gapped state, while there is a 0.9 probability of returning to a matching state. It should be noted that according to this model, all probabilities into or out of a given state must sum to one. In particular, the processing of matrix 30 revolves around calculating the transition probabilities given the various gap opening or gap continuation penalties, and a final sum is calculated.

[0337] Thus, these calculated state transition probabilities are derived primarily from directly contiguous cells in the array 30, such as cells that are immediately to the left, above, or diagonally above and to the left of the given cell currently being calculated, as seen in Figure 16. Additionally, the state transition probabilities may in part be derived from the "Phred" quality score that accompanies each read base. These transition probabilities are therefore useful for computing the state values of M, I, and D for that particular cell, and similarly for any new cell that is calculated. It should be appreciated that, as described herein, the gap opening and gap continuation penalties may be fixed values, however, in many cases, the gap opening and gap continuation penalties may be variable and therefore programmable within the system, albeit employing additional hardware resources dedicated to determining such variable transition probability calculations. Such cases may be useful when greater accuracy is desired. However, when such values are assumed to be constant, lower resource usage and/or chip size can be achieved, leading to higher processing speeds, as explained below.

[0338] Accordingly, there are multiplication calculations and/or other mathematical computations, such as multiplications and/or additions, that are involved in deriving each new state value of M, I, and D. In such a case, such as for calculating maximum throughput, the primitive mathematical computations involved in each transition state calculation of M, I, and D may be segmented. Such segmentation may be configured in a way that the corresponding clock frequencies are high, but where the segmentation depth may be non-trivial. Furthermore, such segmentation may be configured to have a finite depth, and in such cases may take more than one clock cycle to complete the operations.

[0339] For example, these computations may be performed at high speeds within the processor 7, such as about 300 MHz. This may be achieved such as by segmenting the FPGA or ASIC heavily with registers so that small mathematical computation occurs between each flip-flop. This segmentation structure results in multiple cycles of latency in going from the input of the matching state to the output, but given the reverse diagonal computation structure shown in Figure 7 above, these latencies may be hidden throughout the HMM array 30, such as where each cell represents a clock cycle.

[0340] Thus, the number of M, I, and D state calculations may be limited. In such a case, the processing engine 13 may be configured such that a grouping, e.g., range 35, of cells in several rows of the matrix 30 may be processed as a group (such as in a downward and diagonally to the left fashion as illustrated by the arrow in Figure 7) before proceeding to the processing of a second range below, e.g., wherein the second range contains the same number of cells in rows to be processed as the first. In such a manner, a hardware implementation of an accelerator 8 as described herein may be adapted so as to make the overall system more efficient as described above.

[0341] In particular, Figure 9 presents an exemplary computational structure for performing various state processing computations described herein. More particularly, Figure 9 presents three dedicated logic blocks 17 of the processing engine 13 for computing the state computations involved in generating each state value of M, I, and D for each particular cell, or grouping of cells, that is processed in the HMM array 30. These logic blocks may be implemented in hardware, but in some cases may be implemented in software, such as to be performed by one or more quantum circuits. As can be seen from Figure 9, the matching state computation 15a is more involved than either the insertion 15b or deletion 15c computations. This is because in the computation of the matching state 15a of the present cell being processed, all previous matching, insertion and deletion states of the contiguous cells along with various priors are included in the present matching computation (see Figures 9 and 10), while only the matching, insertion and deletion states are included in their respective computations. Therefore, as can be seen from Figure 9, in the computation of a matching state, three state multipliers, as well as two adders, and a final multiplier, which consider the priors, e.g., Phred, are included. However, to compute the I or D state, only two multipliers and one adder are included. It can be seen that in hardware, the multipliers require more resources than the adders.

[0342] Accordingly, to various extents, the M, I, and D state values for processing each new cell in the HMM matrix 30 use knowledge or precomputation of the following values, such as the “prior” M, I, and D state values to the left, above, and/or diagonally to the left and above the cell that is currently computed in the HMM matrix. Additionally, such values representing prior information, or “prioris,” may be, at least in part, based on the “Phred” quality score and whether the read basis and reference basis in a given cell in the matrix 30 match or differ. Such information is particularly useful in determining a matching state. Specifically, as can be seen from Figure 9, in such cases, there are basically seven "transition probabilities" (M to M, I to M, D to M, I to I, M to I, D to D, and M to D) that indicate and/or estimate the probability of seeing a gap opening, e.g., seeing a transition from a match state to an insertion or deletion state; seeing a gap closing, e.g., going from an insertion or deletion state back to a match state; and seeing the next state continuing in the same state as the previous state, e.g., Match to Match, Insert to Insert, Deletion to Deletion.

[0343] The state values (e.g., at any cell to be processed in the HMM array 30), priors, and transition probabilities are all values in the range [0,1]. Additionally, there are also known starting conditions for cells that are on the left or top edge of the HMM array 30. As can be seen from logic 15a of Figure 9, there are four multiplication and two addition computations that can be employed in the particular M state calculation that is determined for any given cell that is processed. Similarly, as can be seen from the logic of 15b and 15c, there are two multiplications and one addition involved for each I state and each D state calculation, respectively. Collectively, together with the prior multipliers, this adds up to a total of eight multiplications and four addition operations for the M, I, and D state calculations associated with each unique cell in the HMM array 8 to be processed.

[0344] The final sum output, e.g., row 34 of Figure 16, of the computation of matrix 30, e.g., for a single job 20 of comparing a read to one or two haplotypes, is the sum of the final M and I states across the entire bottom row 34 of matrix 30, which is the final sum value that is output from HMM accelerator 8 and delivered to CPU/GPU/QPU 1000. This final sum value represents how well the read matches the haplotype (or haplotypes). The value is a probability, e.g. less than one, for a single job 20a that can then be compared to the resulting output of another job 20b to form the same active region 500. Note that there are on the order of 20 trillion HMM cells to evaluate on a "typical" human genome at 30X coverage, where these 20 trillion HMM cells are spread across about 1 to 2 billion HMM matrices 30 from all associated HMM jobs 20.

[0345] The results of such calculations can then be compared against each other in order to determine, in a more precise manner, how the genetic sequence of a subject differs, for example, in a base-by-base comparison, from that of one or more reference genomes. For the final sum calculation, the adders already employed to calculate the M, I and/or D states of the individual cells can be redeployed in order to compute the final sum value, such as by including a multiplexing on a selection of the redeployed adders, thus adding an additional last row, for example, with respect to calculation time, to the matrix in order to calculate this final sum, where, if the read length is 100 bases represents about a 1% overhead. In alternative embodiments, dedicated hardware resources can be used to perform such calculations. In many cases, the logic of the adders for the M and D state calculations can be deployed to calculate the final sum, where the D state adder can be effectively deployed since it is not otherwise being used in the processing leading to the sum values.

[0346] In certain cases, these relevant calculations and processes may be configured to correspond to the output of a given sequencing platform, such as including an array of sequencers, which as a collective may be capable of outputting (on average) a new human genome at 30x coverage every 28 minutes (although resulting from the array of sequencers in clusters of about 150 genomes every three days). In such a case, when the present mapping, alignment, and variant calling operations are configured to fit within a sequencing platform of throughput technologies, a portion of the 28 minutes (e.g., about 10 minutes) that it takes for the sequencing array to sequence a genome may be used by a suitably configured mapper and/or aligner, as described herein, to take the image file/BCL/FASTQ output from the sequencer and perform genome mapping and/or alignment steps, e.g., post-sequencer processing. It takes about 18 minutes of sequencing time to perform the variant calling step, of which the HMM operation is the main computational component, as before the nucleotide sequencer sequences the next genome for the next 28 minutes. Consequently, in such cases, 18 minutes can be budgeted to compute the 20 trillion HMM cells that need to be processed in order to process a genome, as each of the HMM cells to be processed includes about twelve mathematical operations (e.g., eight multiplications and/or addition operations). Such a result allows for the following computational dynamics (20 trillion HMM cells) x (12 mathematical operations per cell) / (18 minutes x 60 seconds/minute), which is about 222 billion operations per second of sustained output.

[0347] Figure 10 shows the logic blocks 17 of the processing engine of Figure 9, including exemplary M, I, and D state update circuits that present a simplification of the circuit provided in Figure 9. The system may be configured so as not to be memory constrained, so that a single instance of HMM engine 13 (e.g., which computes all unique cells in HMM array 30 at a rate of one cell per clock cycle on average, plus overheads) may be replicated multiple times (at least 65 to 70 times to make the result effective as described above). However, to minimize hardware size, e.g., the size of chip 2 and/or its associated resource usage, and/or in a further effort to include as many instances of HMM engine 13 on chip 2 as is desirable and/or possible, simplifications may be made with respect to logic blocks 15a'-c' of processing instance 13 for computing one or more of the transition probabilities to be calculated.

[0348] In particular, it may be assumed that the gap opening penalty (GOP) and gap continuation penalty (GCP), as described above, such as for insertions and deletions, are the same and are known prior to chip configuration. This simplification implies that the I-to-M and D-to-M transition probabilities are identical. In such a case, one or more of the multipliers, e.g., shown in Figure 9, may be eliminated, such as by pre-adding the I and D states before multiplying by a common Indel-to-M transition probability. For example, in many cases, if the I and D state calculations are assumed to be equal, then the per-cell state calculations may be simplified as shown in Figure 10. In particular, if the I and D state values are equal, then the I state and the D state may be added, and then this sum may be multiplied by a single value, thus saving one multiplication. This can be accomplished because, as seen in Figure 10, the continuation and/or gap closure penalties for the I and D states are equal. However, as indicated above, the system may be configured to compute different values for both the I and D transition state probabilities, and in such a case this simplification could not be employed.

[0349] Additionally, in another simplification, rather than dedicating chip or other computing resources specifically configured to perform the final sum operation at the bottom of the HMM array, the present HMM accelerator 8 may be configured to effectively append one or more additional rows to the HMM array 30, with respect to the computational time, e.g., overhead, that it takes to perform the calculation, and may also be configured to "borrow" one or more adders from the M state 15a and D state 15c computation logic, such as by multiplexing the final sum values to the existing adders as needed, in order to perform the actual final sum calculation. In such a case, the final logic, including the blocks of M logic 15a, I logic 15b, and D logic 15c, wherein the blocks together form part of the HMM MID instance 17, may include 7 multipliers and 4 adders along with the various multiplexes involved.

[0350] Accordingly, Figure 10 presents the M, I, and D state update circuits 15a', 15b', and 15c', including the effects of simplifying assumptions related to transition probabilities, as well as the effect of sharing multiple M, I, and/or D resources, e.g., adder resources, for the final addition operations. A delay block may also be added to the M state path in the M state computation block, as shown in Figure 10. This delay may be added to compensate for delays in the actual hardware implementations of the multiplication and addition operations and/or to simplify the control logic, e.g., 15.

[0351] As shown in Figures 9 and 10, these respective multipliers and/or adders may be oscillating point multipliers and adders. However, in many cases, as can be seen with respect to Figure 11, a logarithmic domain configuration may be implemented in which, in such a configuration, all multiplications become additions. Figure 11 shows what a logarithmic-domain computation would look like if all the multipliers became adders, e.g., 15a", 15b", and 15c", as occurs when employing a logarithmic-domain computational setup. In particular, all the multiplier logic becomes an adder, but the adder itself becomes or otherwise includes a function such as: f(a,b) = max(a,b) - log2(1+2A(-[a-b]), such that the logarithmic portion of the equation can be held within a LUT whose depth and physical size are determined by the required accuracy.

[0352] Given typical read and haplotype sequence lengths as well as the values typically seen for read quality score (Phred) and related transition probabilities, the dynamic range demands on the internal HMM state values can be quite severe. For example, when implementing the HMM module in software, multiple HMM jobs 20 may result in stalls, such as when implementing single-precision (32-bit) swing point state values. This implies a dynamic range that is greater than 80 powers of 10, thus requiring the variant calling software to scale up to double-precision (64-bit) swing point state values. However, the full 64-bit double-precision swing point representation in many cases has some negative implications, such as if compact, high-speed hardware is to be implemented, both compute and storage segmentation demands need to be increased, taking up more chip space and/or slow timing. In such cases, a single fixed-point linear domain number representation can be implemented. However, the dynamic range demands on the state values in this embodiment make the bit widths involved in certain circumstances smaller than desirable. Consequently, in such cases, the single fixed-point logarithmic domain number representation can be implemented as described herein.

[0353] In such a scheme, as can be seen with respect to Figure 11, instead of representing the current state value in memory and computations, the logarithmic base 2 of the number may be represented. This may have several advantages, including employing multiplication operations in linear space that translate to addition operations in logarithmic space; and/or such a logarithmic domain representation of numbers inherently supports a wider dynamic range with only small increases in the number of integer bits. Such logarithmic domain M, I, D state update computations are shown in Figures 11 and 12.

[0354] As can be seen by comparing the logic configuration 17 of Figure 11 with that of Figure 9, multiplication operations sum in the logarithmic domain. Instead, they are replaced by addition operations, and the addition operations are transformed into a function that can be expressed as a maximum operation followed by a correction factor addition, e.g., by means of a LUT, where the correction factor is a function of the difference between the two values that are summed in the logarithmic domain. Such a correction factor can be computed or generated from a lookup table. Whether a correction factor computation or a lookup table implementation is more efficient to use depends on the required precision (bit width) in the difference between the sum values. In particular cases, therefore, the number of logarithmic domain bits for state representation can be in the vicinity of 8 to 12 integer bits plus 6 to 24 fractional bits, depending on the level of quality desired for any given implementation. This implies somewhere between 14 and 36 total bits for logarithmic domain state value representation. Furthermore, it has been determined that there are logarithmic domain fixed-point representations that can provide acceptable quality and acceptable hardware size and speed.

[0355] In various cases, a read sequence is typically processed for each HMM job 20, which, as indicated, may include a comparison against two haplotype sequences. And as above for the haplotype memory, a ping-pong structure may also be used in the read sequence memory 18 to allow various software-implemented functions the ability to write new HMM job information 20b while a current job 20a is still being processed by the HMM engine instance 13. Thus, a read sequence storage requirement may be for a single 1,024 x 32 dual-port memory (such as one port for writing, one port for reading, and/or separate clocks for the write and read ports).

[0356] Particularly, as described above, in various instances, the architecture employed by system 1 is configured such that to determine whether a given base in a sequenced sample genome matches that of a corresponding base in one or more reference genomes, a virtual array 30 is formed in which the reference genome is theoretically arrayed across a horizontal axis while the sequenced reads representing the sample genome are theoretically arrayed in a decreasing order down the vertical axis. Accordingly, to perform an HMM calculation, the HMM processing engine 13 as described herein is configured to traverse this virtual HMM array 30. Such processing may be depicted as in Figure 7 as a track 35 moves diagonally down and across the virtual array performing the various HMM calculations for each cell of the virtual array as seen in Figure 8.

[0357] More particularly, this theoretical traversal involves processing a first cluster of cell rows 35a of array 30 in its entirety, such as for all haplotype and read bases within the cluster, before proceeding to the next cluster of rows 35b (e.g., the next group of read bases). In such a case, the M, I, and D state values for the first cluster are stored at the lower edge of this initial cluster of rows so that these M, I, and D state values can then be used to feed the top row of the next cluster (range) down in array 30. In various cases, system 1 may be configured to allow up to 1,008 haplotypes and/or reads in length in HMM accelerator 8, and since the numerical representation employs W bits for each state, this implies a memory of 1,008 words x W bits for M, I, and D state storage.

[0358] Accordingly, as indicated, such memory could be a single-port or dual-port memory. Additionally, a cluster-level scratchpad memory, e.g., for storing range limit results, may also be provided. For example, in accordance with the above disclosure, the memories discussed are already configured on a per-engine 13 instance basis. In particular HMM implementations, multiple engine instances 13a-(n+1) may be grouped into a cluster 11 that is served by a single connection, e.g., PCIe bus 5, to PCIe interface 4 and DMA 3 via CentCom 9. Multiple clusters 11a-(n+1) may be instantiated so as to more effectively utilize PCIe bandwidth using existing CentCom 9 functionality.

[0359] Thus, in a typical configuration, somewhere between 16 and 64 13m engines are instantiated within a cluster 11n, and one to four clusters may be instantiated in a typical HMM FPGA/ASIC implementation 8 (e.g., depending on whether it is a dedicated HMM FPGA image or whether the HMM has to share FPGA real state with the sequencer/mapper/aligner and/or other modules, as disclosed herein). In particular cases, there may be a small amount of memory used at the cluster level 11 in the HMM hardware. This memory can be used as an elastic First In First Out ("FIFO") to capture output data from the HMM engine instances 13 in the cluster and pass it to CentCom 9 for further transmission back to the CPU 1000 software via DMA 3 and PCIe 4. In theory, this FIFO could be quite small (on the order of two 32-bit words), as data is typically passed to CentCom 9 almost immediately after arriving at the FIFO. However, to absorb potential disruptions in the output data path, the size of this FIFO can be parameterizable. In particular cases, the FIFO can be used with a depth of 512 words. Thus, the cluster-level storage requirements could be a single 512 x 32 dual-port memory (separate read and write ports, same clock domain).

[0360] Figure 12 presents various HMM state transitions 17b that represent the relationship between Gap Opening Penalties (GOP), Gap Closing Penalties (GCP), and transition probabilities involved in determining whether and how well a given read sequence matches a particular haplotype sequence. In performing such an analysis, the HMM engine 13 includes at least three logical blocks 17b, such as a logical block for determining a matching state 15a, a logical block for determining an insertion stage 15b, and a logical block for determining a deletion state 15c. This M, I, and D state calculation logic 17, when properly configured, functions effectively to avoid high-bandwidth bottlenecks, such as from the HMM computational throughput. However, once the core computing architecture of M, I, D is determined, other system enhancements can also be configured and deployed in order to avoid the development of other bottlenecks within the system.

[0361] In particular, system 1 may be configured to maximize the process of efficiently feeding information from computation core 1000 to variant caller module 2 and back again, so as not to produce other bottlenecks that could limit overall throughput. Such a block that feeds the M, I, D state computation logic of HMM core 17 is the transition probabilities and priors computation block. For example, as can be seen with respect to Figure 9, each block cycle employs the presentation of seven transition probabilities and one Priori at the input to the M, I, D state computation block 15a. However, after simplifications that result in the architecture of Figure 10, only four unique transition probabilities and one Priori are employed for each clock cycle at the input to the M, I, D state computation block. Accordingly, in many cases, these computations can be simplified and the resulting values generated. Thus, increasing yield, efficiency and reducing the possibility of a bottleneck forming at this stage of the process.

[0362] Additionally, as described above, the Priors are values generated by the read quality, e.g., Phred score, of the particular base being investigated and whether or not that base matches the hypothetical haplotype base for the current cell that is evaluated in the virtual HMM matrix 30. The relationship can be described by the equations below: First, the Phred of the read in question can be expressed as a probability = 10A(-(Phred of read/10)). Then, the Priori can be computed based on whether the read base matches the hypothetical haplotype base: If the read base and the hypothetical haplotype base match: Priori = 1 - Phred of read expressed as a probability. Otherwise: Priori = (Phred of read expressed as probability)/3. The division by three operation in this last equation reflects the fact that there are only four possible bases (A, C, G, T). Therefore, if the read base and haplotype do not match, then it must be one of the three remaining possible bases that do match, and each of the three possibilities is modeled as being equally possible.

[0363] The Phred scores per read base are delivered to the HMM 8 hardware accelerator as 6-bit values. The equations for deriving the Priors then have 64 possible outcomes for the "match" case, and an additional 64 possible outcomes for the "no match" case. This can be effectively implemented in hardware as a 128-word lookup table, where the address in the lookup table is a 7-bit quantity formed by concatenating the Phred value with a single bit indicating whether or not the read base matches the hypothesized haplotype base.

[0364] Further, with respect to determining the match-to-insert and/or match-to-delete probabilities, in various implementations of the architecture for the HMM hardware accelerator 8, separate gap opening (GOP) penalties may be specified for the Match-to-insert state transition and the Match-to-delete state transition, as indicated above. This is the same as the M2I and M2D values in the state transition diagram of Figure 12 which are different. As the GOP values are delivered to the HMM hardware accelerator 8 as 6-bit Phred-like values, the gap opening transition probabilities may be computed in accordance with the following equations: transition probability M2I = 10A(-(Read GOP(I)/10)) and transition probability M2D = 10A(-(Read GOP(D)/10)). Similar to the derivation of priors in hardware, a simple 64-word lookup table can be used to derive the values of M2I and M2D. If GOP(I) and GOP(D) are entered into the HMM 8 hardware as potentially different values, then two such lookup tables (or one shared lookup table per resource, potentially clocked at twice the frequency of the rest of the circuit) can be used.

[0365] Further, with respect to determining the match-to-match transition probabilities, in various cases, the match-to-match transition probability may be calculated as: M2M transition probability = 1 - (M2I transition probability + M2D transition probability). If the M2I and M2D transition probabilities can be configured to be less than or equal to a value of 1/2, then in various embodiments, the above equation may be implemented in hardware in a manner that increases overall efficiency and throughput, such as by rewriting the equation to be: M2M transition probability = (0.5 - M2I transition probability) + (0.05 - M2D transition probability). This rewriting of the equation allows M2M to be derived using two 64-element lookup tables followed by an adder, where the lookup tables store the results.

[0366] Further, with respect to determining the Insertion-to-Insertion and/or Deletion-to-Deletion transition probabilities, the I2I and D2D transition probabilities are functions of the gap continuation probability (GCP) values input to the HMM hardware accelerator 8. In various cases, these GCP values may be 6-bit Phred-like values given on a per-read basis. The I2I and D2D values may then be derived as shown: I2I transition probability = 10A(-(Read GCP(I)/10)) and D2D transition probability = 10A(-(Read GCP(D)/10)). Similar to some of the other transition probabilities discussed above, I2I and D2D values can be effectively implemented in hardware and can include two lookup tables (or one shared lookup table per resource), such as having the same form and content as the Indel Match lookup tables discussed earlier. That is, each lookup table can have 64 words.

[0367] Additionally, with respect to determining the Insertion and/or Deletion probabilities for Matching, the I2M and D2M transition probabilities are functions of the gap continuation probability (GCP) values and can be computed as: I2M transition probability = 1 - I2I transition probability and D2M transition probability = 1 - D2D transition probability, where the I2I and D2D transition probabilities can be derived as discussed above. A simple subtraction operation to implement the above equations may be more expensive in hardware resources than simply implementing another 64-word lookup table and using two copies of it to implement the I2M and D2M derivations. In such cases, each lookup table may be 64 words long. Of course, in all relevant embodiments, simple or complex subtraction operations can be formed with appropriately configured hardware.

[0368] Figure 13 provides circuitry 17a for a simple computation for HMM transition probabilities and Priors as described above that supports the general state transition diagram of Figure 12. As can be seen with respect to Figure 13, in various cases, a simple HMM hardware accelerator architecture 17a is presented, wherein the accelerator may be configured to include separate GOP values for Insertion and Deletion transitions and/or there may be separate GCP values for Insertion and Deletion transitions. In such a case, the cost of generating seven unique transition probabilities and a Priori each clock cycle may be configured as follows: eight 64-word lookup tables, one 128-word lookup table, and an adder.

[0369] Furthermore, in various cases, the hardware 2 as disclosed herein may be configured to fit as many instances of HMM engine 13 as possible on the given target chip (such as in an FPGA, sASIC, or ASIC). In such a case, the cost of implementing the transition probabilities and prior generation logic 17a may be substantially reduced relative to the costs as provided by the configurations below. First, rather than supporting a more general version of state transitions as disclosed in Figure 13, for example, where there may be separate values for GOP(I) and GOP(D), it may instead, in various cases, be assumed that the GOP values for insertion and deletion transitions are the same for a given basis. This results in several simplifications for the hardware, as noted above.

[0370] In such cases, only one 64-word lookup table can be employed to generate a single M2Indel value, replacing both the M2I and M2D transition probability values, while two tables are typically employed in the more general case. Similarly, only one 64-word lookup table can be used to generate the M2M transition probability value, while both tables and one addition can typically be employed in the general case, since M2M can now be calculated as l-2xM2Indel.

[0371] Second, the assumption can be made that the sequencer-dependent GCP value for both insertion and deletion are equal and that this value does not change over the course of an HMM 20 job. This means that: a single Indel2Indel transition probability can be calculated instead of separate I2I and D2D values, using a 64-word lookup table instead of two tables; and a single Indel2Match transition probability can be calculated instead of separate I2M and D2M values, using a 64-word lookup table instead of two tables.

[0372] Additionally, another simplifying assumption can be made that assumes that the Insert2Insert and Delete2Delete (I2I and D2D) and Insert2Match and Delete2Match (I2M and D2M) values are not identical between insert and delete transitions, but may be static for the particular HMM job 20. Thus, the four associated lookup tables in the more general architecture with I2I, D2D, I2M, and D2M transition probabilities can be eliminated altogether. In many such cases, the static Indel2Indel and Indel2Match probabilities could be entered via software or via an RTL parameter (and thus could be bitstream programmable in an FPGA). In certain cases, these values may be made bitwise programmable, and in certain cases, a training mode may be implemented employing a training sequence in order to further refine the transition probability accuracy for a given run of sequencing or genome analysis.

[0373] Figure 14 presents what the new state transition diagram 17b might look like when deploying these various simplifying assumptions. Specifically, Figure 14 presents the simplified HMM state transition diagram that represents the relationship between GOP, GCP, and transition probabilities with the simplifications presented above.

[0374] Similarly, Figure 15 presents the circuitry 17a,b for HMM transition probabilities and prior generation, which supports the simplified state transition diagram of Figure 14. As noted with respect to Figure 15, a circuit realization of this state transition diagram is provided. Thus, in various cases, for the HMM hardware accelerator 8, the cost of generating the transition probabilities and a Prior each clock cycle reduces to: Two 64-word lookup tables and one 128-word lookup table.

[0375] As set forth above, engine control logic 15 is configured to generate the virtual array and/or traverse the array so as to reach the range edge, e.g., via high-level engine state mechanisms, wherein the result data may be finally summed, e.g., via final sum control logic 19, and stored, e.g., via put/get logic.

[0376] Accordingly, as can be seen with respect to Figure 16, in various embodiments, a method for producing and/or traversing an array of HMM cells 30 is provided. Specifically, Figure 16 presents an example of how the HMM accelerator control logic 15 traverses the virtual cells in the HMM array. For example, assuming for exemplary purposes a latency of 5 clock cycles for each multiplication and addition operation, the worst-case latency through the M, I, D state update calculations could be the 20 clock cycles that would be taken to propagate through the M update calculation. There are half as many operations in the I and D state calculations, implying a latency of 10 clock cycles for these operations.

[0377] These latency implications of M, I, and D computation operations can be understood with respect to Figure 16, which presents several examples of cell-to-cell data dependencies. In such cases, the M and D state information for a given cell feeds the D state computations of the cell in the HMM matrix that is immediately to the right (e.g., that has the same read base as the given cell, but that has the next haplotype base). Similarly, the M and I state information for the given cell feeds the I state computations of the cell in the HMM matrix that is immediately below (e.g., that has the same read base as the given cell, but that has the next read base). Thus, in particular cases, the M, I, and D states of a given cell feed the D and I state computations of cells on the next diagonal of the HMM cell matrix.

[0378] Similarly, the M, I, and D states of a given cell feed the M state computation of the cell that is to the right and below it (e.g., that has both the next haplotype base and the next read base). This cell is actually two diagonals away from the cell that feeds it (whereas the I and D state computations are based on states of a cell that is one diagonal away). This quality of the I and D state computations being based on cells one diagonal away even though the M state computations are based on cells two diagonals away has a beneficial result for hardware design.

[0379] In particular, given these configurations, the I and D state computations can be tailored to take half the time (e.g., 10 cycles) of the M state computations (e.g., 20 cycles). Thus, if the M state computations are initiated 10 cycles before the I and D state computations for the same cell, then the M, I, and D state computations for a cell in the HMM array 30 will all complete at the same time. Additionally, if the array 30 is traversed in a diagonal fashion, such as having a range 35 of about 10 cells each within it (e.g., spanning ten read bases), then: The M and D states produced by a given cell at coordinates (hap, rd) (i, j) can be used by the D state computations of cell (i+1, j) once they are on their way to the computation segment of the cell at (i, j).

[0380] The M and D states produced by a given cell at coordinates (hap, rd)(i, j) may be used by the I state computations of cell (i, j+1) one clock cycle after they are en route to the cell's computation segment at (i, j). Similarly, the M, I, and D states produced by a given cell at coordinates (hap, rd)(i, j) may be used by the M state computations of cell (i+1, j+1) one clock cycle after they are en route to the cell's computation segment at (i, j). Taken together, the above points establish that very little dedicated storage is required for the M, I, and D states along the diagonal of the stripe path that spans the stripe length of, say, ten reads. In such a case, only the registers needed to delay the M, I, and D state values of cell (i, j) M, I, and D by one clock cycle for use in M-cell (i+1, j+1) computations and I-cell (i, j+1) computations by one clock cycle). Furthermore, there is a somewhat virtuous cycle here as the M-state computations for a given cell are started 10 clock cycles before the I and D state computations for that same cell, natively emitting the new M, I, and D states for any given cell simultaneously.

[0381] In view of the foregoing and as can be seen with respect to Figure 16, the HMM accelerator control logic 15 may be configured to evaluate the data within each of the cells of the virtual matrix 30 in a manner that traverses the matrix. In particular, in various embodiments, operations begin at cell (0,0), with the M state calculations beginning 10 clock cycles before the I and D state calculations begin. The next cell to traverse should be cell (1,0). However, there is a latency of ten cycles after the start of I and D calculations before the results from cell (0,0) are available. The hardware therefore inserts nine "dead" cycles into the computation bucket. These are shown as the cells with haplotype index less than zero in Figure 16.

[0382] After completing the dead cycle that has an effective cell position in the array of (-9,-9), the M, I, and D state values for cell (0,0) are available. They (e.g., the M and D state outputs of cell (0,0)) can now be used directly to initiate the D state computations of cell (0,1). One cycle later, the M, I, and D state values of cell (0,0) can be used to initiate the I state computations of cell (0,1) and the M state computations of cell (1,1).

[0383] The next cell to be traversed may be cell (2,0). However, there is a latency of ten cycles after the start of I and D computations before the results of cell (1,0) are available. The hardware therefore inserts eight dead cycles into the computation segment. These are shown as the cells with haplotype index less than zero, as in Figure 16 along the same diagonal as cells (1,0) and (0,1). After completing the dead cycle that has an effective cell position in the matrix of (-8, -9), the M, I, and D state values for cell (1,0) are available. These (i.e., the M and D state outputs of cell (1,0)) are now used directly to start the D state computations of cell (2,0).

[0384] One cycle later, the M, I, and D state values of cell (1,0) may be used to initiate the I state computations of cell (1,1) and the M state computations of cell (2,1). The M and D state values of cell (0,1) may then be used at the same time to initiate the D state computations of cell (1,1). One cycle later, the M, I, and D state values of cell (0,1) are used to initiate the I state computations of cell (0,2) and the M state computations of cell (1,2).

[0385] Now the next cell to traverse may be cell (3,0). However, there is a latency of ten cycles after the start of I and D computations before the results of cell (2,0) are available. The hardware therefore inserts seven dead cycles into the computation segment. These are again shown as the cells with haplotype index less than zero in Figure 16 along the same diagonal as cells (2,0), (1,1) and (0,2). After completing the dead cycle that has an effective cell position in the matrix of (-7,-9), the M, I and D state values for cell (2,0) are available. These (i.e. the M and D state outputs of cell (2,0)) are now used directly to start the D state computations of cell (3,0). And so the computation for another ten cells on the diagonal begins.

[0386] Such processing may continue until the end of the last complete diagonal in strip 35a, which in this example (which has a read com of 35 and haplotype length of 14) will occur after the diagonal starting with the cell at coordinates (hap, rd) of (13,0) is completed. After cell (4,9) in Figure 16 is traversed, the next cell to traverse could be cell (13,1). However, there is a latency of ten cycles after the start of the I and D calculations before the results from cell (12,1) are available.

[0387] The hardware may therefore be configured to begin operations associated with the first cell in the next strip 35b, such as at coordinates (0, 10). After processing cell (0, 10), then cell (13, 1) may be traversed. The entire diagonal of cells starting with cell (13, 1) is then traversed until cell (5, 9) is reached. Similarly, after cell (5, 9) is traversed, the next cell to traverse could be cell (13,2). However, as before, there may be a latency of ten cycles after the start of I and D calculations before the results of cell (12, 2) are available. Therefore, the hardware may be configured to begin operations associated with the first cell in the second diagonal of the next strip 35b, such as at coordinates (1, 10), followed by cell (0, 11).

[0388] After processing cell (0, 11), cell (13, 2) may be traversed in accordance with the methods disclosed above. The entire diagonal 35 of cells beginning with cell (13, 2) is then traversed until cell (6, 9) is reached. Additionally, after cell (6, 9) is traversed, the next cell to be traversed could be cell (13, 3). However, here again, there may be a latency period of ten cycles after the start of the I and D calculations before the results of cell (12, 3) are available. The hardware may therefore be configured to begin operations associated with the first cell on the third diagonal of the next range 35c, such as at coordinates (2, 10), followed by cell (1, 11) and (0, 12), and the like.

[0389] This continues as indicated, in accordance with the above, until the last cell in the first strip 35a (the cell at coordinates (hap, rd) (13, 9)) is traversed, at which point the logic can be completely devoted to traversing the diagonals in the second strip 35b, starting with the cell at (9, 10). The pattern depicted above repeats for as many strips of 10 reads as necessary, until the bottom strip 35c (where those cells in this example that are associated with read bases have an index of 30 or greater) is reached.

[0390] In the bottom lane 35, more dead cells may be inserted, as shown in Figure 16 as cells with read indices greater than 35 and with haplotype indices greater than 13. Additionally, in the final lane 35c, an additional row of cells may be effectively added. These cells are indicated in row 35 in Figure 16 and refer to a dedicated clock cycle on each diagonal of the final lane in which the final sum operations are occurring. In these cycles, the M and I states of the cell immediately above are added together, and the result itself is summed with a running final sum (which is started at zero on the left edge of the HMM array 30).

[0391] Taking the above discussion as context, and in view of Figure 16, it can be seen that for this example of read length 35 and haplotype length 14, there are 102 dead cycles, 14 cycles associated with final sum operations, and 20 cycles of segmentation latency, for a total of 102+14+20 = 146 cycles of overhead. It can also be seen that for any HMM 20 job with a read length greater than 10, the dead cycles in the upper left corner of Figure 16 are independent of read length. It can also be seen that the dead cycles in the bottom and lower right portions of Figure 16 are dependent on read length, with fewer dead cycles for reads that have mod(read length, 10) = 9 and more dead cycles for mod(read length, 10) = 0. It can further be seen that the overhead cycles become less than a total percentage of HMM matrix evaluation cycles 30 as haplotype lengths increase (larger matrix, partially fixed number of overhead cycles) or as read lengths increase (note: this refers to the overhead percentage associated with the final sum row in the matrix that decreases as read length—row—count—increases). Using such histogram data from representative whole-genome human run-throughs, it was determined that traversing the HMM matrix in the manner described above results in less than 10% overhead for the entire genome processing.

[0392] Other methods may be employed to reduce the amount of overhead cycles, including: Having dedicated logic for the final sum operations instead of sharing adders with the M and D state calculation logic. This eliminates a row of the HMM array 30. Using dead cycles to start the HMM array operations for the next HMM job in the queue.

[0393] Each grouping of ten rows of the HMM array 30 constitutes a "strip" 35 in the HMM accelerator function. It will be noted that the length of the strip can be increased or decreased to meet the efficiency and/or throughput demands of the system. Thus, the strip length can be about five rows to about fifty rows or more, such as about ten rows to about forty-five rows, for example, about fifteen or about twenty rows to about forty rows, or about thirty-five rows, including about twenty-five rows to about thirty rows of cells in length.

[0394] With the exceptions noted in the above section relating to collect cycles that might otherwise be dead cycles on the forward edge of the array of Figure 16, the HMM array may be processed one lane at a time. As can be seen with respect to Figure 16, the states of the cells in the bottom row of each lane 35a feed the state computation logic in the top row of the following lane 35b. Accordingly, there may be a need to store (put) and retrieve (get) the state information for these cells in the bottom row, or edge, of each lane.

[0395] The logic for doing so may include one or more of the following: when M, I, and D state computations for a cell in HMM array 30 are completed for a cell with mod(read index, 10) = 9, saving the result to M, I, D state storage memory. When M, I state computations (e.g., where D state computations do not require information from cells above them in the array) for a cell in HMM array 30 begin for a cell with mod(read index, 10) = 0, retrieving the M, I, and D state information previously stored in M, I, D state storage memory. It will be noted in these cases that the M, I, and D state values that feed the M, I state computations of row 0 (the top row) in HMM array 30 are simply a predetermined constant value and do not need to be recalled from memory, as is true for the M, I, and D state values that are stored in the M, I, D state storage memory. of M and D that feed the D state calculations from column 0 (the left column).

[0396] As indicated above, the HMM accelerator may or may not include a dedicated summation resource in the HMM hardware accelerator such that it exists simply for the purpose of final summation operations. However, in particular cases, as described herein, an additional row may be added to the bottom of the HMM array 30, and the clock cycles associated with that extra row may be used for final summation operations. For example, the summation itself may be achieved by borrowing (e.g., in accordance with Figure 13) an adder from the M state computation logic for the M+I operation, and further borrowing an adder from the D state computation logic to add the newly formed M+I sum to the ongoing final sum accumulation value. In such a case, the control logic for activating the final summation operation may begin whenever the read index driving the HMM traversal operation equals the length of the read sequence input for the job. These operations can be seen in line 34 toward the bottom of the 30-sample HMM matrix of Figure 16.

[0397] Thus, as can be seen above, in one implementation, the variant caller may make use of the mapper and/or aligner mechanisms to determine the likelihood as to when multiple reads originated, such as relative to a given location, e.g., chromosomal location. In such cases, the variant caller may be configured to detect the underlying sequence at that location, such as independently of other regions not immediately adjacent to it. This is particularly useful and works well when the region of interest does not resemble any other region of the genome by the breadth of a single read (or a pair of reads for paired-end sequencing). However, a significant fraction of the human genome does not meet this criterion, which can make variant calling, e.g., the process of reconstructing a subject's genome from the reads that an NGS produces, challenging.

[0398] In particular, although DNA sequencing has improved dramatically, variant calling remains a difficult problem due to the redundant structure of the genome. As disclosed herein, however, the complexities presented by genome redundancy can be overcome, at least in part, from a short-read data-driven perspective. More particularly, the devices, systems, and methods for employing the same as disclosed herein can be configured in such a manner as to target homologous or similar regions that may otherwise have been characterized by low variant calling accuracy. In certain cases, such low variant calling accuracy may arise from difficulties observed in mapping and read alignments to homologous regions that typically result in very low read MAPQs. Accordingly, strategic implementations that accurately call variants (SNPs, Indels, and the like) in homologous regions, such as by jointly considering the information present in those homologous regions, are disclosed herein.

[0399] For example, many regions of the genome are homologous, e.g., they have non-identical copies located elsewhere in the genome, e.g., at multiple sites, and as a result, the true source site of a read may be subject to considerable uncertainty. Specifically, if a group of reads map with low confidence, e.g., due to apparent homology, a typical variant caller may ignore and not process the reads, even though they may contain useful information. In other cases, if a read maps erroneously (e.g., the primary alignment is not the true source of the read), detection errors may result. More specifically, previously deployed short-read sequencing technologies have been susceptible to these issues, and conventional detection methods often leave large regions of the genome in the dark.

[0400] In some cases, long-read sequencing can be employed to mitigate these issues, but typically has much higher cost and/or higher error rates, takes longer, and/or suffers from other drawbacks. Therefore, in many cases, it may be beneficial to perform a multi-region joint operation as described herein. For example, rather than considering each region in isolation and/or rather than performing and analyzing long-read sequencing, multi-region joint detection (MRJD) methodologies may be employed, such as where the MRJD protocol considers multiple, e.g., all, sites from which a group of reads may originate, and attempts to detect the underlying sequences together, e.g., in aggregate, using all available information, which may be related to low or abnormal confidence and/or certainty scores.

[0401] For example, for a diploid organism with statistically uniform coverage, a brute-force Bayesian computation as described above can be performed in a variant calling analysis. However, in a brute-force MLRD computation, the complexity of the computation grows rapidly with the number of regions N and the number of candidate haplotypes K to be considered. In particular, to consider all combinations of candidate haplotypes, the number of candidate solutions for which to compute probability can often be exponential. For example, as described in more detail below, in a brute-force implementation, the number of candidate haplotypes includes the number of active positions, where if a graph fitting technique is used to generate the list of candidate haplotypes in a variant calling operation, such as in the construction of a De Brujin graph as disclosed herein, then the number of active positions is the number of independent "blobs" in the graph. Hence, such a brute force calculation may be prohibitively expensive to implement, and like such Bayesian calculations may be prohibitively complex.

[0402] Accordingly, in one aspect, as set forth in Figure 17A, a method for reducing the complexity of such brute force calculations is provided herein. For example, as disclosed above, although the speed and accuracy of DNA/RNA sequencing has been dramatically improved, especially relative to the methods disclosed herein, variant calling, e.g., the process of reconstructing a subject's genome from the reads that a sequencer produces, remains a difficult problem, largely due to the redundant structure of the genome. The devices, systems, and methods disclosed herein are therefore configured to reduce the complexities presented by genome redundancy from a perspective driven by short-read data as opposed to long-read sequencing. In particular, methods for performing very long-read detection representing homologous and/or similar regions of the genome that are typically characterized by low variant calling accuracy without necessarily having to perform long-read sequencing are provided herein.

[0403] For example, in one embodiment, a system and method for performing joint detection of multiple regions are provided. Specifically, in a first case, a general variant calling operation may be performed such as employing the methods disclosed herein. In particular, a general variant caller may employ a reference genome sequence, wherein the reference genome presents all of the bases in a template genome. This reference forms the backbone of an analysis by which a subject's genome is compared to the reference genome. For example, as discussed above, employing a next-generation sequencer, a subject's genome may be broken down into subsequences, e.g., reads, typically about 100 to 1,000 bases each, wherein the reads may be mapped and aligned to the reference, much like assembling a jigsaw puzzle.

[0404] Once the subject's genome has been mapped and/or aligned, using this reference genome in comparison to the subject's actual genome, it can be determined to what extent and how the subject's genome differs from the reference genome, e.g., in a base-by-base manner. In particular, when comparing the subject's genome to one or more reference genomes, such as in a base-by-base manner, the analysis moves iteratively through the sequences, comparing one to the other (or others) to determine whether they agree or diverge. Accordingly, each base within the sequences represents a position to be called, as represented by position A in Figure 18A

[0405] Specifically, for each position A of the reference to be called relative to the subject genome, an oversampling of sequences, e.g., reads, will be mapped and aligned in such a way that a large sample set of reads can fully overlap each other at any given position A. In particular, such oversampling may include a number of reads, e.g., from one to a hundred or more, where each of the reads in the oversampling has nucleotides that overlap the region being called. The calling of these reads base by base therefore involves the formation of a processing window that slides along the calling sequences, where the length of the window, e.g., the number of bases under examination at any given time, forms the active region of determination. Thus, the window represents the active region of bases in the sample that is called, where calling involves comparing each base at a given position, e.g., A, across all reads in the stack within the active region, where the identity of the base at that position in the stack number of reads provides evidence of the true identity of the base at the position that is called.

[0406] For this purpose, based on the relevant MAPQ confidence score derived for each read segment, it can be generally determined, within a certain confidence score, that the mapping and alignment were performed accurately. However, the question still remains, no matter how subtle, as to whether or not the mapping and alignment of the reads are accurate, or whether one or more of the reads actually belong elsewhere. Accordingly, in one aspect, devices and methods for improving confidence in variant calling performance are provided herein.

[0407] In particular, in various cases, the variant caller may be configured to perform one or more joint multi-region detection operations as described herein, which may be employed to generate greater confidence in the achievable results. For example, in such a case, the variant caller may be configured to analyze the various regions in the genome in order to determine particular regions that appear to be similar. For example, as can be seen with respect to Figure 18A, there may be a reference region A, and a reference region B, where the referenced sequences are very similar to each other, for example, but with a few regions of dissimilar base pair matching, such as where example Ref A has an "A", example Ref B has a "T", but outside of these small dissimilarities, all other locations within the region in question may appear to match. Due to the extent of the similarities, these two regions, e.g., Ref A and Ref B, will typically be considered homologous or paralogous regions.

[0408] As depicted, the two reference regions A and B are 99% similar. There may be other regions, e.g., Refs C and D, that are relatively similar, e.g., about 93% similar, but compared to the 99% similarity between reference regions A and B, reference regions C and D might not be considered homologous or at least might have a lower chance of actually being homologous. In such a case, variant calling procedures might be able to adequately call differences between reference regions C and D, but might in certain cases have difficulty calling differences between the highly homologous regions of reference regions A and B, e.g., due to their high homology. In particular, due to the extent of dissimilarity between reference sequences A and B and reference sequences C and D, it would not be expected that reads that map and align to Ref Seq A or B would erroneously map to Ref Seq C or D. However, it would be expected that reads that map and align to Ref Seq A would erroneously map to Ref Seq B.

[0409] Given the extent of homology, erroneous mapping between regions A and B may be quite likely. Accordingly, to increase accuracy, it may be desirable for the system to be able to distinguish and/or account for the difference between homologous regions, such as when performing a mapping, alignment, and/or variant calling procedure. Specifically, when generating a backlog of reads that map and align to a region within Ref A and generating a backlog of reads that map and align to a region within Ref B, any of the reads may in fact be erroneously mapped to the wrong location, and as such, to effect improved accuracy, when performing the variant calling operations disclosed herein, these homologous regions, and the reads mapped and aligned to them, should be considered together, such as in a joint detection protocol, e.g., a multi-region joint detection protocol as described herein.

[0410] Accordingly, presented herein are devices, systems, as well as methods of their use, which are directed to multiple region joint detection (MRJD), such as where a plurality, e.g., all, of the reads from various clusters of the various identified homologous regions are considered together, such as where, instead of making a single call for each site, a joint call is made for all sites that appear to be homologous. Making such joint calls is advantageous due to the fact that, before attempting to make a call for each reference individually, it would first have to be determined to which region, of which reference, the various reads in question actually map and align and this is inherently uncertain, and this very problem is solved by the proposed joint detection. Thus, due to the fact that the regions of the two references are so similar, it is very difficult to determine which reads map to which regions. However, if these regions are called together, there is no need to make an advance decision about which homologous reads map to which reference region. Therefore, when performing joint calling, the assumption can be made that any reads in a cluster from a region in one reference, e.g. A, that are homologous to another region in a second region, e.g. B, could belong to either Ref. A or Ref. B.

[0411] Accordingly, when desired, an MRJD protocol may be deployed in addition to the variant calling algorithm deployed in the devices, systems, and methods herein. For example, in one iteration, a variant calling algorithm takes the evidence presented in the reads mapped and/or aligned to a given region in the sample and reference genomes, analyzes the likelihood that what appears to be in the sample genome is in fact present, based on a comparison to the reference genome, and makes a decision given the evidence as to how the sample actually differs from the reference, e.g., given this evidence, the variant caller algorithm determines the most likely answer of what is different between the read and the reference. However, MRJD is another algorithm that may be deployed in conjunction with the VC algorithm, wherein MRJD is configured to help the variant caller more accurately determine whether an observed difference, e.g., in the subject's read, is in fact a true deviation from the reference.

[0412] Accordingly, the first step in an MJRD analysis involves identifying homologous regions, based on a percentage match between the sequence in a plurality of regions from one or more references, e.g., Ref. A and Ref. B, and the sequences accumulated in one or more regions from the subject reads. In particular, Ref. A and Ref. B may actually be diploid forms of the same genetic material, such as where there are two copies of a given region of the chromosome. Thus, when diploid references are being analyzed, at various positions, Ref. A may have a particular nucleotide, and at the same position in Ref. B, another nucleotide may be present. In this example, Ref. A and Ref. B are homozygous at position A for "A". However, as can be seen in Figure 18A, the subject's DNA is heterozygous at this A position, such that, relative to the reads from the accumulation of Ref. A, one allele of the subject's chromosome has an "A" but the other allele has a "C", yet relative to Ref. B, another copy of the subject's chromosome has an "A" for both alleles at the A position. This also becomes more complicated when the sample being analyzed contains a mutation, for example, at one of these naturally occurring variable positions, such as a heterozygous SNP at the A position (not shown).

[0413] As can be seen with respect to Reference A of Figure 18B, at position A, the subject sample may include reads that indicate heterozygosity at position A, such as where some of the reads include a "C" at that position, and some of the reads indicate an "A" at that position (e.g., haplotype a1 = "A", Ha2 = "C"); whereas, with respect to Ref. B, the reads at position A indicate homozygosity, such as where all of the reads in the pool have an "A" at that position (e.g., HBI = "A", HB2 = "A"). However, MRJD overcomes these difficulties by performing joint calling simultaneously, analyzing all reads that map to both regions of the reference, while considering the possibility that any of the reads may be in the wrong location. Once the various homologous regions have been identified, the next step is to determine the correspondence between the homologous reference regions, and then, relative to MRJD, the mapper and/or aligners' determination as to where the various applicable reads should "map" between the two homologous regions can be discarded and instead all reads in any of the clusters in these homologous regions can be considered collectively, knowing that any of these reads could belong to any of the homologous regions being compared. Therefore, the calculations to determine these joint calls, as detailed below, consider the possibility that any of these reads came from any of the homologous reference regions, and, where applicable, from any haplotype in either reference region.

[0414] It should be realized that although the foregoing has been with reference to multiple regions of homology within a reference, the same analysis can be applied to single region detection as well. For example, as can be seen with respect to Figure 18B, even for a single region, for any given region, there may be two separate haplotypes present, e.g., HI and H2, that the subject's genetic sample may have for a particular region, and because they are haplotypes, they will likely be very similar to each other. Consequently, if these positions are analyzed in isolation from each other, it may be difficult to determine whether there are true variations being considered. Thus, the calculations that are performed with respect to homologous regions are useful for non-homologous regions as well, because any given region is likely to be diploid, e.g., having both a first haplotype (H1) and a second haplotype (H2), and thus, analyzing the regions together will increase the accuracy of the system. Similarly, for a region of reference days, for example a homologous region as described above, what is being called is an HAI and HA2 for the first region, and an HAI and HA2 for the second region (which is equivalent to two strands for each chromosome and two regions for each strand = 4 diploid types, overall.

[0415] Accordingly, MRJD can be employed to determine an initial response, with respect to one or more, e.g., all, of the homologous regions, and then single region detection can be applied back to one or more, e.g., all, of the unique or non-homologous regions, e.g., employing the same basic analysis, and thus better accuracy can be achieved. Thus, non-joint single region detection can also be performed. For example, with respect to single region detection, for candidate haplotypes, HAI, in current iterations, the reference region may be about 300 to 500 base pairs long, and over the reference, a graph, e.g., a De Bruijn graph as shown in Figure 18C, is constructed, such as of K-mers from the reads, where any site that differs from the reference forms a divergent trajectory or "bubble" in the graph, from which haplotypes are extracted, where each extracted haplotype, e.g., divergent trajectory, forms a potential hypothesis for which may lie on one of the two strands of the chromosomes at a particular site in the active region under examination.

[0416] However, if there are many divergent trajectories, for example many bubbles across the graph are formed, as seen with respect to Figure 18C, and a large number of haplotypes have been extracted, then a maximum cutoff may be introduced to keep the calculations manageable. The cutoff may be at any statistically significant number, such as 35, 50, 100, 125 to 128, 150, 175, 200 or more, etc. However, in certain cases, a substantially larger number, for example all, of the haplotypes may be considered.

[0417] In such a case, instead of extracting the entire source to narrow down haplotypes from start to finish, e.g. from the beginning of the sequence to the end, only the sequences associated with the individual bubbles need to be extracted, e.g. only the bubbles need to be aligned to the reference. Accordingly, the bubbles are extracted from the DBG, the sequences aligned to the reference, and from these alignments, specific SNPs, insertions, deletions and the like can be determined, as to why the sequences of the various bubbles differ from the reference. Hence, in this sense, all the different hypothetical haplotypes for analysis can be derived by mixing and matching the sequences belonging to all the various bubbles in different combinations. In such a manner, all the haplotypes to be extracted do not need to be enumerated. Such methods for performing joint detection of multiple regions are described in more detail below in this document.

[0418] Furthermore, abstractly, although all of these candidate haplotypes can be tested, a tree growth algorithm can be performed when the graph that is produced begins to look like a growing tree. For example, a branching tree graph of joint haplotypes/diplotypes can be constructed in such a way that as the tree grows, the underlying algorithm works to both grow and prune the tree as more and more calculations are performed, and it becomes apparent that several different candidate hypotheses are simply too unlikely. Thus, as the tree grows and is pruned, not all of the hypothesized haplotypes need to be calculated.

[0419] Specifically, with respect to growing the tree function, when there is a discrepancy between two references, or between references and reads, as to which base is present at given positions that are separated, it must be determined that the base actually belongs at that position, and in view of such discrepancies, it needs to be determined which differences may be caused by SNPs, Indels, or the like, versus which are machine errors. Consequently, when causing tree growth, for example by extracting bubbles from the De Bruijn graph, such as through SW or NW alignment, and positioning them within the emerging tree graph, each bubble to be extracted becomes an event in the tree graph, which represents possible SNPs, Indels, and/or other differences from the reference. See Figure 18C.

[0420] In particular, in a DBG, the bubbles represent divergences from the reference, e.g., representative of Indels (which bases were added or deleted), SNPs (which bases are different), and the like. Consequently, as the bubbles are aligned to the reference (or references), the various differences between the two are categorized as events, and a list of the various events, e.g., bubbles, is generated. Therefore, the determination then becomes: which combination of the possible events, e.g., possible SNPs and Indels, led to the actual variations in the subject's genetic sequence, e.g., is the true one in each of the various actual haplotypes, e.g., 4, based on probability. More particularly, any candidate, e.g., joint diplotype candidate, forming a GO root (which represents events for a given segment) may have 4 haplotypes, and each of the four haplotypes will form an identified subset of the events.

[0421] However, as can be seen with respect to Figure 18D, when performing a tree function growth and/or pruning, a complete list of the entire subset of all event combinations can be, but need not be, determined all at once. Instead, the determination starts at a single GO position, e.g., an event, and the tree grows from there one event at a time, which, through the pruning function, can leave multiple probability events unresolved. Thus, with respect to a tree function growth, as can be seen with respect to Figure 18D, the computation starts with the determination of haplotypes, e.g., HAI, HA2, HBI, HB2 (for a diploid organism), where the initial haplotypes are assumed to be all unresolved with respect to their respective references, e.g., Ref. A and Ref. B, with essentially none of the events present.

[0422] Accordingly, the initial starting point is with the root of the tree being GO, and the diplotype set that has all unresolved events. Then, a particular event, e.g. an initial bubble, is selected as the origin for determination, such that the initial event must be resolved for all haplotypes, where the event may be a first point of divergence from the reference, such as with respect to the potential presence of a SNP or Indel at position one. As exemplified in Figure 18E, at position one, there is an event or bubble, such as a SNP, where a "C" has been replaced by an "A", such that the reference has an "A" at position one, but the read in question has a "C". In such a case, since for that position in the stack there are 4 haplotypes, and each may have an "A", as in the reference, or the "C" event, there are potentially 24=16 possibilities to resolve that position. Therefore, the calculation immediately moves from the root to 16 branches, representing the potential resolutions to the event at position one.

[0423] Thus, as can be seen from Figure 18D, all potential sequences for all four haplotypes can be presented, e.g., HAI, HA2, HBI, HB2, where at position one there is either an "A" as in agreement with the reference, or event "C" indicating the presence of a SNP, for that event one, where event "C" is determined by examining the various bubble trajectories through the graph. Thus, for each branch or child node, each branch may differ based on the likelihood of the base at position one agreeing with or deviating from the reference, while the remainder of the events remain unresolved. This process will then be repeated for each branch node and for each base within the variation bubbles, so as to resolve all events for all haplotypes. The probabilities can then be recalculated for observing any particular readout given the various potential haplotypes.

[0424] In particular, for each node, there may be four haplotypes, and each haplotype may be compared against each read in the backlog. For example, in one embodiment, the SW, NW, and/or HMM engine analyzes each node and considers each of the four haplotypes for each node. Accordingly, generating each node triggers the SW and/or HMM engine to analyze that node considering all haplotypes, e.g., 4, for that node compared to each of the reads, wherein the SW and/or HMM engine considers one haplotype for one read for each of the haplotypes and each of the reads for all viable nodes.

[0425] Thus, if, for the exemplary purposes of this example, it is the case that there is a heterozygous "C" SNP for a region of a haplotype, e.g., one strand of a chromosome has a "C", but all the other bases at that position for the other strands do not, e.g., all match the reference "A", then all of the reads in the cluster would be expected to support that conclusion, such as having a majority of "A"s at position one, and a minority, e.g., about 1/4, of the reads having a "C" at position one, for the true node. Thus, if any subsequent observable reads at a different node show a multiplicity of "Cs" at position one, then that node is unlikely to be the true node, e.g., it will have a low probability, due to the fact that there are not enough reads with Cs at that position in the cluster to make its occurrence likely. Specifically, it is more likely that the existence of a "C" at that position in the reads in question is evidence of a sequencing or other scientific error, rather than being a true haplotype candidate. Consequently, if certain nodes end up having low probabilities compared to the true node, this is because they are not supported by a majority of the reads, e.g. in the backlog, and so these nodes can be pruned, thereby discarding the low-probability nodes but in a way that preserves the true node (or nodes).

[0426] Accordingly, once event position one has been determined, the next event position can be determined, and the processes described herein can then be repeated for that new position relative to any of the surviving nodes that were not previously pruned. In particular, event two can be selected from the existing available nodes, and that event can serve as the GI root to determine the likely identity of the base at position two, such as by once again defining new haplotypes, e.g., 4, as well as their various branches, e.g., 16, accounting for possible variations relative to position 2. Thus, by repeating this same process, event 2 can now be resolved. Therefore, as can be seen with respect to Figure 18D, once position 1 has been determined, a new node for position 2 can be selected, and its 16 potential haplotype candidates can be considered. In such a case, candidates for each of HAI, HA2, HBI, HB2 can be determined, but in this case, since position 1 has already been resolved, in terms of determining the nucleotide identity for each of the haplotypes at position 1, it is position 2 that will now be resolved, for each of the haplotypes at position 2, as shown in Figure 18D, which shows the resolution of position 2.

[0427] Once this process is complete, once all events have been processed and resolved, e.g. including all child nodes and children of child nodes that were not pruned, then the nodes in the tree that were not pruned can be examined, and it can be determined based on the probability scores, that the tree represents the joint diplotype, e.g. where the sequence has the highest probability of being true. So in this way, because of the pruning function, the entire tree does not need to be constructed, e.g. most of the tree will end up being pruned as the analysis continues, so the overall amount of computation is greatly reduced relative to non-pruning functions, although substantially more than performing non-joint diplotype calling, e.g. single region calling. Consequently, the present analytical modules can determine and resolve two or more regions of high homology with a high degree of accuracy, e.g., employing joint diplotype analysis, where traditional methods simply cannot resolve such regions, e.g., due to false positives and irresolutions.

[0428] In particular, various variant caller implementations may be configured to simply not perform an analysis on regions of high homology. The present iterations overcome these and other such problems in the field. More particularly, the present devices, systems, and methods of use thereof may be configured to consider a larger proportion, e.g., all haplotypes, despite the occurrence of regions of high homology. Of course, the speed of such calculations may be further increased by not performing certain calculations where it may be determined that the results of such calculations have a low probability of being true, such as by implementing a pruning function as described herein.

[0429] A benefit of these configurations, e.g., joint diplotype resolution and pruning, is that now the size of the active region window, e.g., of bases that are analyzed, can be increased from about a few hundred bases being processed to a few thousand, or even tens or hundreds of thousands of bases can be processed together, such as in a contiguous active region. This increase in size of the active window of analysis allows more evidence to be considered when determining the identity of any particular nucleotide at any given position, thus allowing a larger context within which a more accurate determination of nucleotide identity can be made. Similarly, a larger context allows supporting evidence to be better chained together when comparing one or more reads that cover one or more regions that have one or more deviations from the reference. Thus, in such a manner, an event can be connected to another event, which itself can be connected to another event, etc., and from these connections a more precise call regarding a particular event currently under consideration can be made, thus allowing more distant evidence, e.g. hundreds to thousands of bases or more away, to be informative in producing a given variant call (despite the fact that any given read is typically only hundreds of bases long), thus making the processes herein much more precise.

[0430] Particularly, in such a manner, the active region can further be produced to include thousands, even tens of thousands, even hundreds of thousands of bases or more, and consequently the method for forming a De Bruijn graph by extracting all haplotypes can be avoided, since only a limited number of haplotypes, those with bubbles that may be viable, need to be explored, and even those that are viable, once it becomes clear that they are no longer viable, can be pruned, and for those that remain viable, chaining can be employed in order to improve the accuracy of the eventual variant calls that are made. All of this is made possible by quantum and/or hardware computing. The same can also be done in software by a CPU or a GPU, but it will be slower.

[0431] It should be appreciated that with respect to the above examples, it is the probabilities of the input data, e.g. the reads, that are being determined, given these haplotype theories produced by the De Bruijn plot. However, it may also be useful to employ Bayes' theorem, such as to determine the probability of reads given a joint diplotype, to the converse probability of determining from the joint diplotype theory a best fit given the reads and the evidence evaluated. Consequently, as can be seen with respect to Figure 18C, from the generated De Bruijn plot, once joint detection of multiple regions and/or pruning has occurred, a set of potential haplotypes will result, and these haplotypes will then be tested against the subject's actual reads. Specifically, each horizontal cross-section represents a haplotype, e.g. B1, which can then be subjected to another HMM protocol to retest the reads to determine the probability of a particular read given the B1 haplotype.

[0432] However, in certain cases, the haplotype, e.g., B1, may not yet be completely determined, but HMM may still be useful to perform, and in such a case, a modified HMM calculation, e.g., a partially determined (PD) HMM operation, discussed below, may be performed where the haplotype is allowed to have undetermined variants, e.g., SNPs and/or indels, where the same have not yet been determined, and as such the calculation is similar to calculating the best possible probability for an achievable answer given any combination of variants at the unresolved positions. This therefore further facilitates iterative growth of the tree function, where the actual growth of the tree, e.g., performing the PD HMM operation, need not be restricted to only those calculations where all possible variants are known. Thus, in this way, several HMM PD calculations can be performed, in an iterative manner, to grow the node tree, despite the fact that there are still undetermined regions of unknown possible events in particular candidate haplotypes, and when it becomes possible to prune the tree, HMM PD resources can be fluidly shifted from calculating pruned nodes to processing only those possibilities that have the highest probability of successfully characterizing the true genotype.

[0433] Accordingly, in determining the probability that a particular base is actually present at any one position, the identity of the base at that position can be determined based on the identity at that position in each region of each chromosome, e.g., each haplotype, which represents a viable candidate. Thus, for any given candidate, what is being determined is the identity of the given base at the position in question in each of the four haplotypes simultaneously. In particular, what is being determined is the probability of observing the reads from each of the stacks given the given likelihood. Specifically, each candidate represents a joint diplotype, and so each candidate includes about four haplotypes, which can be presented in the following equations as G = genotype, where G = the four haplotypes of a single diploid region of a chromosome of the genome, e.g., a joint diplotype. In such a case, what must be calculated is the probability of actually observing each of the candidate read bases identified from the sequences in the stacks that are assumed to be in fact the truth. This initial determination can be accomplished by an HMM haplotype calculation as presented above in this document.

[0434] For example, for a "Diplotype Set" = 4 Haplotypes: (Region A: HAIHA2, and Region B: HBIHB2 ) = G - P(R/G) as determined by an HMM (Error Model) = n P(r/G) =

[0435] Therefore, if it is assumed that the specific haplotype Ha1 is the true sequence in that region, and the read comes from it, then what is the chance that this Ha1 read sequence was actually observed. Consequently, the HMM calculator works to determine, assuming that the Ha1 haplotype is the true one, what is the probability of actually observing the given read sequence in question.

[0436] Specifically, if the read does match the haplotype, then this is obviously a very high probability. However, if the particular read in question does not match the haplotype, then any deviation from it must be explainable by a scientific error, such as a sequencing error or sequencing mechanism, and not by actual variation. Thus, the calculation of HMM is a function of error patterns. Specifically, it asks what is the probability of the necessary combination of errors that could have occurred in order to observe the particular reads being analyzed. Consequently, in this model not just one region is being considered, but a multiplicity of positions in a multiplicity of regions in a multiplicity of strands are being considered simultaneously (e.g., instead of considering at most possibly two haplotypes in a region, now what is being considered is simultaneously the possibility of four haplotypes for any given position in any given region, simultaneously, using all the read data from all the regions in question. These processes, e.g., tree pruning, joint detection of multiple regions, and HMM PD, will now be described in more detail.

[0437] Specifically, as can be seen with respect to Figures 17 and 18 , a high-level processing chain is provided, such as wherein the processing chain may include one or more of the following steps: Identifying and inserting homologous regions, performing processing of the inserted homologous regions, performing a very long read pruning (VLRD) or a multiple region joint detection (MJRD), and outputting a variant call file. Particularly, with respect to identifying homologous regions, a mapped, aligned, and/or classified SAM and/or BAM file, e.g., a CRAM, may be used as the primary input to a multiple region joint detection processing engine that implements an MRJD algorithm as described herein. The MJRD processing engine may be part of an integrated circuit, such as a CPU and/or GPU and/or quantum computing platform, executing software, e.g., a quantum algorithm, or deployed within an FPGA, ASIC, or the like. For example, the mapper and/or aligner disclosed above can be used to generate a CRAM file, e.g., with settings to output N secondary alignments for each read along with the primary alignments. These primary and secondary reads can then be used to identify a list of homologous regions, where homologous regions can be computed based on a user-defined similarity threshold between the N regions of the reference genome. This list of identified homologous regions can then be fed to the preprocessing stage of an appropriately configured MRJD module.

[0438] Accordingly, in the preprocessing stage, for each set of homologous regions, a joint accumulation may first be generated such as using primary alignments from one or more, e.g., each, region in the set. See, e.g., Figure 19. Using this joint accumulation, a list of active/candidate variant positions (SNPS/INDELs) may then be generated such that each of these candidate variants may be processed and evaluated by the MRJD preprocessing engine(s). To reduce computational complexity, a connection matrix may be computed that may be used to define the order of processing of the candidate variants.

[0439] In such implementations, the multiple region joint detection algorithm evaluates each identified candidate variant based on the processing order defined in the generated connection matrix. First, one or more candidate joint diplotypes (Gi) may be generated and given a candidate variant. Then, the posterior probabilities of each of the joint diplotypes (P(Gi|R)) may be calculated. From these posterior probabilities, a genotype matrix may be computed. Then, N diplotypes with the lowest posterior probabilities may be pruned so as to reduce the computational complexity of the calculations. Then, the next candidate variant that provides evidence for the current candidate variant being evaluated may be included in the above repeated process. Having included information such as from one or more, e.g., all, candidate variants from one or more, e.g., all, regions in the homologous region defined for the current variant, a variant call may be performed from the final genotyping matrix. Each of the active positions can therefore be evaluated in the above manner, thus resulting in a final VCF file.

[0440] In particular, as can be seen with respect to Figure 17B, an MJRD processing step may be implemented, such as including one or more of the following steps and blocks: The identified and assembled joint accumulation is loaded, a list of candidate variants is then created from the assembled joint accumulation, and a connection matrix is computed. In particular, in various cases, a pre-processing methodology may be performed, such as prior to performing one or more variant calling operations, such as a multiple read joint detection operation. Such operations may include one or more pre-processing blocks, including: the steps pertaining to loading joint accumulations, generating a list of variant candidates from the joint accumulations, and computing the connection matrix. Each of the blocks and potential steps associated therewith will now be discussed in more detail.

[0441] Specifically, a first joint accumulation preprocessing block may be included in the analysis procedure. For example, multiple reference regions for a given identified range may be extracted, such as from the mapped and/or aligned reads. In particular, using the list of homologous regions, a joint accumulation for each set of homologous regions may be generated. Then, a user-defined range may be used to extract the N reference regions corresponding to N homologous regions within a range. Subsequently, one or more, e.g., all, of the reference regions may be aligned, such as using a Smith-Waterman alignment, which may be used to generate a world coordinate system of all bases in the N reference regions. Furthermore, all primary reads corresponding to each region may then be extracted from the input SAM or BAM file and mapped to the world coordinates. This mapping can be performed, as described in this document, such as using the alignment information (CIGAR) present in a CRAM file for each read. In the scenario where some read pairs have not been previously mapped, the reads can be mapped and/or aligned, e.g., aligned by Smith-Waterman, to their respective reference region.

[0442] More particularly, once a pooled build has been generated and loaded, see e.g. Figure 19, a list of candidate variants may be created, such as from the pooled build. For example, a De Bruijn graph (DBG) or other assembly graph may be produced in order to extract multiple candidate variants (SNPs/Indels) that can be identified from the pooled build. Once the DBG is produced, the various bubbles in the graph may be extracted in order to derive a list of variant candidates.

[0443] In particular, given all reads, a graph can be generated using each reference region as a backbone. All identified candidate variant positions can then be aligned to world coordinates. A connection matrix can then be computed, where the matrix defines the processing order of the active positions, which can be a function of read length and/or insert size. As mentioned herein, Figure 19 shows an example of a joint stacking of two homologous regions on chromosome 1. Although this stacking is with reference to two homologous regions of chromosome 1, it is for exemplary purposes only, since the output of the stacking process can be used for any and all homologous regions regardless of chromosome.

[0444] As can be seen from Figure 20, a list of candidate variants can be created as follows. First, a cluster can be formed and a De Bruijn graph (DBG) or other assembly graph can be constructed, in accordance with the methods disclosed herein. The DBG can then be used to extract candidate variants from the clusters. The construction of the DBG is performed in such a manner as to generate bubbles, indicating variations, representing alternative trajectories through the graph, where each alternative path is a candidate haplotype. See, for example, Figures 20 and 21.

[0445] Accordingly, the various bubbles in the graph represent the list of candidate variant haplotype positions. Then, given all the reads, the DBG can be generated using each reference region as a backbone. Then, all candidate variant positions can be aligned to world coordinates. Specifically, Figure 20 illustrates a flowchart that presents the process of generating a DBG and using it to produce candidate haplotypes. More specifically, the De Bruijn graph can be employed to create the list of candidate variants from SNPs and INDELs. Given that there are N regions that are being jointly processed by MRJD, N de-bruijn graphs can be constructed. In such a case, each graph can use one reference region as a backbone and all the reads corresponding to the N regions.

[0446] For example, in one methodological implementation, after the DBG is constructed, candidate haplotypes can be extracted from the De Bruijn graph based on the candidate events. However, by employing an MRJD preprocessing protocol as described herein, N regions can be processed together, such as where the length of the regions can be a few thousand bases or more, and the number of haplotypes to be extracted can grow exponentially very quickly. Consequently, in order to reduce computational complexity, instead of extracting entire haplotypes, only the blobs need to be extracted from the graphs that are representative of the candidate variants.

[0447] An example of bubble structures formed in a De Bruijn graph is shown in Figure 21. Several regions to be processed together are identified. This determines one of two processing trajectories that can be followed. If joint regions are identified, all reads can be used to form a DBG. Bubbles showing possible variants can be extracted in order to identify the various candidate haplotypes. Specifically, for each bubble, a SW alignment can be performed on the alternative paths to the reference backbone. For this, candidate variants can be extracted and the events of each graph can be stored.

[0448] However, in other cases, once the first process has been performed, so as to generate one or more DBGs, and/or i is now equal to 0, then the union of all candidate events from all DBGs can be generated, where any duplicates can be removed. In such a case, all candidate variants can be mapped, such as to a universal coordinate system, so as to produce the list of candidates, and the list of candidate variants can be sent as an input to a pruning module, such as the MJRD module. An example of only performing bubble extraction, rather than extracting the entire haplotypes, is shown in Figure 22. In this case, it is only the bubble region showing possible variants that is extracted and processed, as described herein.

[0449] Specifically, once the representative blobs have been extracted, global alignment, e.g. Smith-Waterman alignment, of the blob path and the corresponding reference backbone can be performed to obtain the candidate variant(s) and their position in the reference. This can be performed for all extracted blobs in the entire De Bruijn graph. Then, the union of all extracted candidate variants can be taken from the N graphs, duplicate candidates, if any, can be removed, and the unique candidate variant positions can be mapped to the universal coordinate system obtained from the joint accumulation. This results in a final list of candidate variant positions for the N regions that can act as an input to a "Pruned" MRJD algorithm.

[0450] In particular preprocessing blocks, as described above herein, a connection matrix may be computed. For example, a connection matrix may be used to define the order of processing of active, e.g., candidate positions, such as a function of read length and insert size. For example, to further reduce computational complexity, a connection matrix may be computed so as to define the order of processing of identified candidate variants that are obtained from the De Bruijn graph. Such a matrix may be constructed and employed in conjunction with or as a ranking function to determine which candidate variants to process first. Such a connection matrix, therefore, may be a function of the average read length and the insert size of the paired-end reads. Consequently, for a given candidate variant, other candidate variant positions that are at integral multiples of the insert size or within the read length have higher weights compared to candidate variants at other positions. This is because these candidate variants are more likely to provide evidence for the current variant being evaluated. An exemplary ranking function, as deployed in this paper, is shown in Figure 23 for an average read length of 101 and insert size of 300.

[0451] With respect to an MJRD pruning function, exemplary steps of a pruned MRJD algorithm as referred to above are shown in Figure 24. For example, the input to the MRJD platform and algorithm is the joint accumulation of N regions, e.g., all candidate variants (SNPs/INDELs), the a priori probabilities based on a mutation model, and the connection matrix. Accordingly, the input to the pruned MRJD processing platform may be the joint accumulation, the identified active positions, the generated connection matrix, and the a posteriori probability model and/or the results thereof.

[0452] Next, each candidate variant in the list can be processed and other variants can be successively added as evidence for a current candidate that is processed using the connection matrix. Consequently, given the current candidate variant and any supporting candidates, candidate joint diplotypes can be generated. For example, a joint diplotype is a set of 2N haplotypes, where N is the number of regions that are processed together. The number of candidate joint diplotypes M is a function of the number of regions that are processed together, the number of active/candidate variants that are considered, and the number of phases. An example for generating joint diplotypes is shown below. For: P = 1 , Number of active/candidate variant positions that are considered; N = 2, Number of regions that are processed together; M = 2<2 N P>= 24= 16 candidate joint diplotypes

[0453] Therefore, for a single candidate active position, given all reads and both reference regions, let the two haplotypes be 'A' and 'G'.Unique haplotypes = 'A' and 'G'Candidate diplotypes = 'AA', 'AG', 'GA' and 'GG', (4 candidates for 1 region).Candidate joint diplotypes =

[0454] Accordingly, using the candidate diplotype sets, read probabilities can be calculated given a haplotype for each haplotype in the entire set of candidate diplotype sets. This can be accomplished using an HMM algorithm as described herein. However, in doing so, the HMM algorithm can be modified from its standard use case to allow candidate variants (SNPs/INDELs) in the haplotype that have not yet been processed to be considered. Subsequently, read probabilities can be calculated given a diplotype set (P(ri|Gm)) using the results of the modified HMM. This can be accomplished using the formula below.

[0455] For the case of joint detection of 2 regions: where i is the region and j is the phase, it is straightforward to calculate P(R|Gm) for all readings. Then, using Bayes' formula, the posterior probability (P(Gi|R)) can be computed from P(R|Gi) and the prior probabilities (P(Gj)).

[0456] Furthermore, an intermediate genotype matrix can be computed for each region given the posterior probabilities for all candidate joint diplotypes. For each combination of events in the genotype matrix, the posterior probabilities of all joint diplotypes supporting that event can be summed. At this point, the genotype matrix can be considered as "intermediate" due to the fact that all candidate variants supporting the current candidate have been included. However, as seen previously, the number of joint diplotype candidates grows exponentially with the number of candidate variant positions and the number of regions. This in turn exponentially increases the computation required to compute the posterior probabilities. Therefore, in order to reduce computational complexity, at this stage, the number of joint diplotypes based on the posterior probabilities can be pruned such that the number of joint diplotypes maintained is user-defined and programmable. Finally, the final genotype matrix can be updated based on a user-defined confidence metric of variants that is computed using the intermediate genotype matrix. The various steps of these processes are shown in the process flowchart in Figure 24.

[0457] The above process may be repeated until all candidate variants are included as evidence for the current candidates that are processed using the connection matrix. Once all candidates have been included, processing of the current candidate is performed. Other stopping criteria for processing candidate variants are also possible. For example, the process may be stopped when the confidence has stopped increasing as more candidate variants are added. This analysis, as exemplified in Figure 24, may be restarted and repeated in the same manner for all other candidate variants in the list, thus resulting in a final variant call file in the MRJD output. Consequently, rather than considering each region in isolation, a multi-region joint detection protocol, as described herein, may be employed to consider all locations from which a group of reads may have originated as it attempts to detect the underlying sequences together using all available information.

[0458] Accordingly, for joint detection of multiple regions, an exemplary MRJD protocol may employ one or more of the following equations in accordance with the methods disclosed herein. Specifically, rather than considering each region to be evaluated in isolation, MRJD considers a plurality of locations from which a group of reads may have originated and attempts to detect the underlying sequences jointly, such as using as much, e.g., all, of the available information that is useful. For example, in an exemplary embodiment:

[0459] N is the number of regions to be processed together. And Hk is a candidate haplotype, k = 1...K, each of which may include multiple SNPs, insertions and/or deletions relative to a reference sequence. Each Hk haplotype represents a single region along a single strand (or "phase", e.g., maternal or paternal), and need not be contiguous (e.g., may include gaps or "unimportant" sequences).

[0460] Gm is a candidate solution for both phases Φ = 1,2 (for a diploid organism) and all regions n = 1...N:where each element Gm,Φ,n is a haplotype chosen from the candidate set {HI ...Hk}.

[0461] First, the probability of each read can be calculated for each candidate haplotype P(ri|Hk), for example using a Hidden Markov Model (HMM). In the case of datasets with paired reads, ri denotes the pair {ri,1, ri;2}, and P(ri|Hk) = P(ri,1|Hk) P(ri,2|Hk). In the case of datasets with linked reads (e.g. barcoded reads), n denotes the group of reads {ri,1...ri;NL} that originate from the same long molecule, and .

[0462] Then, for each candidate solution Gm, m=1 ...M, the conditional probability of each reading and the conditional probability of the entire accumulation are calculated.

[0463] Next, the a posteriori probability of each candidate solution given the observed accumulation is calculated: where P(Gm) indicates the a priori probability of the candidate solution, which is presented in detail below.

[0464] Finally, the relative probability of each candidate variant Vj is calculated, such as where Gm Vj indicates that Gm supports variant Vj, and Gm ref indicates that Gm supports the reference. In a VCF file, this can be reported as a quality score on a phred scale:

[0465] An exemplary process for performing various variant calling operations is presented herein with respect to Figure 25, where a conventional detection and MRJD process are compared. Specifically, Figure 25 illustrates a joint accumulation of paired reads for two regions whose reference sequences differ by only 3 bases relative to the range of interest. All reads are known to originate from either region #1 or region #2, but it is not known for certain which region any individual read originated from. Note, as described above, that bases are only shown for positions where the two references differ, e.g., bubble regions, or where the reads differ from the reference. These regions are referred to as the active positions. All other positions can be ignored, as they do not affect the calculation.

[0466] Consequently, as can be seen from Figure 25, in a conventional detector, read pairs 1-16 could be mapped to region #2, and these alone could be used for variant calling in region #2. All of these reads match the reference for region #2, so no variants could be called. Similarly, read pairs 17-23 could be mapped to region #1, and these alone could be used for variant calling in region #1. As can be seen, all of these reads match the reference for region #1, so no variants would be called. However, read pairs 24-32 map equally well to region #1 and region #2 (each has a one base difference from ref #1 and ref #2), so the mapping is indeterminate, and a typical variant caller could simply ignore these reads. As such, a conventional variant caller might not produce any variant calls for any region, as seen in Figure 25.

[0467] However, with MRJD, Figure 25 illustrates that the result is completely different from that obtained using conventional methods. The relevant calculations are presented below. In this case, N = 2 regions. Additionally, there are three positions, each with 2 candidate bases (one can safely ignore bases whose count is sufficiently low, and in this example the count is zero in all but 2 bases at each position). If all combinations are considered, this yields = 23= 8 candidate haplotypes: Hi = CAT, H2= CAA, H3= CCT, H4= CCA, H5= GAT, H6= GAA, H7= GCT, H8= GCA.

[0468] In a brute force calculation, where all combinations of all candidate haplotypes are considered, the number of candidate solutions is M = K2N = 82.2 = 4096, and P(Gm/R) can be calculated for each candidate solution Gm. The following illustrates this calculation for two candidate solutions: Where Gmi has no variants (this is the solution found by a conventional detector), and Gm2 has a single heterozygous SNP A > C at position #2 of region #1.

[0469] The probability P(ri|Hk) depends on several factors, including base quality and other HMM parameters. It can be assumed that only base calling errors are present and all base calling errors are equally likely, thus , where pe is the probability of a base calling error, Np(i) is the number of active base positions overlapped by read i, and Ne(i) is the number of errors for read i, assuming haplotype Hk. Consequently, it can be assumed that pe = 0.01, which corresponds to a base quality of phred 20. The table presented in Figure 26 shows P(ri|Hk), for all read pairs and all candidate haplotypes. The two rightmost columns show P(ri|Gmi) and P(ri|Gm2 ), with the product at the bottom. Figure 26 shows that P(R|Gmi) = 3.5'30 and P(R|Gm2 ) = 2.2-15, a difference of 15 orders of magnitude in favor of Gm2

[0470] The posterior probabilities P(Gm|R) depend on the prior probabilities P(Gm). To complete this example, a simple independent identically distributed (IID) model can be assumed, so that the prior probability of a candidate solution with Nv variants is , where Np is the number of active positions (3 in this case) and Pv is the probability of a variant, assumed to be 0.01 in this example. This yields P(Gm) = 7.22e-13, and P(Gm2 ) = 0.500. Note that Gm2 is heterozygous over the no 1 region, and all heterozygous pairs of haplotypes have a mirror image representation with the same probability (obtained by simply sweeping these). In this case, the sum of the probabilities for Gm2 and its mirror image sums to 1.000. By calculating the probabilities of individual variants, a heterozygous SNP AC at position no. 2 of region no. 1, with a phred quality score of 50.4, can be seen.

[0471] Consequently, as can be seen, there is immense computational complexity to performing a brute-force variant calling operation, where the complexity can be reduced by performing joint detection of multiple regions, as described herein. For example, the complexity of the above calculations grows rapidly with the number of regions N and the number of candidate haplotypes K. To consider all combinations of candidate haplotypes, the number of candidate solutions for which to calculate probabilities is M = K2N. In a brute-force implementation, the number of candidate haplotypes is K = 2Np, where Np is the number of active positions (e.g., as exemplified above, if graph-fitting techniques are used, the list of candidate haplotypes, then Np is the number of independent blobs in the graph). Thus, a mere brute-force calculation can be prohibitively expensive to implement. For example, if N = 3 and Np = 10, the number of candidate solutions is M = 23.2.10 = 260 = 1018. However, in practice, it is not uncommon to have values of Np much larger than this.

[0472] Accordingly, because a brute-force Bayesian computation can be prohibitively complex, the following description presents other methods for reducing the complexity of such computations. For example, in a first step of another embodiment, starting with a small number of positions , (or even a single position = 1), the Bayesian computation can be performed through these positions. At the end of the computation, candidates whose probability is below a predefined threshold can be eliminated, such as in a tree function pruning as described above. In such a case, the threshold can be adaptive.

[0473] Then, in a second step, the number of positions may be increased by a small number ΔNp (such as one: ), and the surviving candidates may be combined with one or more, e.g., all, of the possible candidates at the new position(s), as in a tree function growth. These steps of (1) performing the Bayesian computation, (2) pruning the tree, and (3) growing the tree may then be repeated, e.g., sequentially, until a stopping criterion is met. The threshold history may then be used to determine the confidence of the result (e.g., the probability that the true solution was found or not). This process is illustrated in the flowchart shown in Figure 27.

[0474] It should be understood that there are a variety of possible variations to this approach. For example, as indicated, the pruning threshold may be adaptive, such as based on the number of surviving candidates. For example, a simple implementation may set the threshold to keep the number of candidates below a fixed number, while a more sophisticated implementation may set the threshold based on a cost-benefit analysis of including additional candidates. Furthermore, a simple stopping criterion may be that a result has been found with a sufficient level of confidence, or that the confidence in the initial position has stopped growing as more positions are added. Furthermore, a more sophisticated implementation may perform some type of cost-benefit analysis of continuing to add more positions. Additionally, as can be seen with respect to Figure 27, the order in which new positions are added may depend on a variety of criteria, such as the distance to the initial position (or positions) or how highly connected those positions are to the positions already included (e.g., the amount of overlap with paired reads).

[0475] A useful feature of this algorithm is that the probability that the true solution was not found can be quantified. For example, a useful estimate is obtained by simply summing the probabilities of all pruned branches at each step: Ppruned = Ppruned + ∑mesetpruned . Such an estimate is useful for calculating the confidence of the resulting variant calls: . Good confidence estimates are essential for producing good Receiver Operating Characteristic (ROC) curves. This is a key advantage of this pruning method over other ad hoc complexity reductions.

[0476] Returning to the example accumulation of Figure 25 and starting with the leftmost position (position #1) and working toward the rightmost base position one in time, using a pruning threshold of phred 60 in each iteration: represents the candidate solutions at the j-th iteration. Figure 28 shows the candidate solutions at the first iteration, representing all combinations of bases C and G, listed in order of decreasing probability. For any solution with equivalent mirror-image representations (obtained by scanning the phases), only a single representation is shown. The probabilities for all candidate solutions can be calculated, and those probabilities beyond the pruning threshold (indicated by the solid line in Figure 28) can be disregarded. As can be seen from Figure 28, as a result of the pruning methods disclosed herein, six candidates survive.

[0477] Next, as can be seen from Figure 29, the tree can be grown by finding all combinations of surviving candidates from iteration #1 and candidate bases (C and A) at position #2. A partial list of new candidates is shown in Figure 29, again shown in order to decrease the probability. Again, the probabilities can be calculated and compared to the pruning limit, and in this case, 5 candidates survive.

[0478] Finally, all combinations of the surviving candidates from iteration #2 and the candidate bases at position #3 (A and T) can be determined. The final candidates and their associated probabilities are shown in Figure 30. Consequently, when calculating the probabilities of individual variants, a heterozygous AC SNP at position #2 of region #1 is determined, with a phred score of 50.4, which is the same result found in the brute force calculation. In this example, pruning had no significant effect on the final result, but in general, pruning can affect the calculation, often resulting in a higher confidence score.

[0479] There are many possible variations for implementations of this approach, which may affect the performance and complexity of the system, and different variations may be suitable for different scenarios. For example, there may be variations in the decision of which regions to include. For example, before performing joint detection of multiple regions, the variant caller may be configured to determine whether a given active region should be processed individually or together with other regions, and if together, it may then determine which regions to include. In other cases, some implementations may rely on a list of secondary alignments provided by the mapper to inform or otherwise make this decision. Other implementations may use a database of homologous regions, computed offline, such as based on a reference genome search.

[0480] Consequently, a useful step in such operations is to decide which positions to include. For example, it should be realized that many regions of interest may not be self-contained and/or isolated from adjacent regions. Thus, the information in the accumulation may influence the probability of separated bases by much more than the total read length (e.g., the paired read length or long molecule length). As such, it needs to be decided which positions to include in the MRJD calculation, and the number of positions is not unrestricted (even with pruning). For example, some implementations may process overlapping blocks of positions and update the results for a subset of the positions based on the confidence levels in those positions or the completeness of the evidence at those positions (e.g., positions near the middle of the block typically have more complete evidence than those near the edge).

[0481] Another determining factor may be the order in which new positions are added. For example, for pruned MRJD, the order in which new positions are added may affect performance. For example, some deployments may add new positions based on the distance to already included positions or the degree of connectivity to those positions (e.g., the number of reads that overlap both positions). Additionally, there are also many variations on how pruning can be performed. In the example presented above, pruning was based on a fixed probability threshold, but in general, the pruning threshold can be adaptive or based on the number of surviving candidates. For example, a simple deployment may set the threshold to keep the number of candidates below a fixed number, while a more sophisticated deployment may adjust the threshold based on a cost-benefit analysis of including additional candidates.

[0482] Several implementations may perform pruning based on probabilities P(R|Gm) rather than prior probabilities P(Gm|R). This has the advantage of allowing elimination of equivalent mirror image representations across regions (in addition to phases). This advantage is at least partially offset by the disadvantage of not pruning candidates with very low prior probabilities, which in some cases may be beneficial. As such, a useful solution may depend on the scenario. If pruning is performed, such as based on P(R|Gm), then the Bayesian computation could be performed once after the final iteration.

[0483] Furthermore, in the above example, the process was stopped after processing all the base positions in the backlog shown, but other stopping criteria are also possible. For example, if only a subset of the base positions (e.g., when processing overlapping blocks) is being solved, the process could stop when the result for the subset has been found with a sufficient level of confidence, or when the confidence has stopped increasing as more positions are added. A more sophisticated implementation, however, might perform some sort of cost-benefit analysis, weighing the computational cost against the potential value of adding more positions.

[0484] Prior probabilities can also be useful. For example, in the examples above, a simple IID model was used, but other models can also be used. For example, one might notice that clusters of variants are more common than would be predicted by an IID model. One might also note that variants are more likely to occur at positions where references differ. Therefore, incorporating such knowledge into the prior probabilities P(Gm) can improve detection performance and produce better ROC curves. In particular, it should be noted that prior probabilities for homologous regions are not well understood in the genomics community, and this knowledge is still evolving. As such, some implementations may update the prior models as better information becomes available. This may be done automatically as more results are produced. Such updates may be based on other biological samples or other regions of the genome for the same sample, the learnings from which can be applied to the methods in this document to further promote faster and more accurate analysis.

[0485] Accordingly, in some cases, an iterative MJRD process may be implemented. Specifically, the methodology described herein may be extended to allow message passing between related regions in order to further reduce complexity and/or increase detection performance of the system. For example, the output of the computation at one location may be used as an a priori input probability for the computation at a nearby location. Additionally, some implementations may use a combination of pruning and iteration to achieve the desired performance/complexity tradeoff.

[0486] Furthermore, sample preparation can be implemented to optimize the MRJD process. For example, for paired-end sequencing, it may be useful to have a concise distribution in insert size when using conventional detection. However, in many cases, introducing variation in insert size could significantly improve performance for MRJD. For example, the sample can be prepared to intentionally introduce a bimodal distribution, a multimodal distribution, or a bell curve-like distribution with a higher variance than would typically be implemented for conventional detection.

[0487] Figure 31 illustrates the ROC curves for MRJD and a conventional detector for the human sample NA12878 over selected regions of the genome with a single homologous copy, such that N = 2, with varying degrees of reference sequence similarity. This dataset used paired-end sequencing with a read length of 101 and an average insert size of approximately 400. As can be seen from Figure 31, MRJD provides dramatically increased sensitivity and specificity over these regions relative to conventional detection methods. Figure 32 illustrates the same results displayed as a function of the sequence similarity of the references, measured over a 1,000-base window (e.g., if the references differ by 10 bases out of 1,000, then the similarity is 99.0 percent). For this dataset, it may be obvious that conventional detection starts to perform poorly at sequence similarity -0.98, while MRJD performs quite well up to 0.995 and beyond.

[0488] Additionally, in various cases, this methodology may be extended to allow message passing between related regions to further reduce complexity and/or increase detection performance. For example, the output of the computation at one location may be used as an a priori input probability for the computation at a nearby location, and in some implementations, may use a combination of pruning and iteration to achieve the desired performance/complexity tradeoff. In particular cases, as indicated above, prior to performing joint detection of multiple regions, the variant caller may determine whether a given active region should be processed individually or jointly with other regions. Additionally, as indicated above, some implementations may rely on a list of secondary alignments provided by the mapper to make such a decision. Other implementations may use a database of homologous regions, computed offline based on a search of the reference genome.

[0489] In view of the foregoing, a Hidden Markov Model with Pairwise Determination (HMM PD) can be implemented in a manner that takes advantage of the benefits of MRJD. For example, MRJD can separately estimate the probability of observing a portion or all of the reads given each possible diplotype set, which comprises one haplotype per ploidy per homologous reference region. For example, for two homologous regions on diploid chromosomes, each diplotype set will include four haplotypes. In such cases, all or a portion of the possible haplotypes can be considered, such as sense constructed, for example, by modifying each reference region with each possible subset of all variants for which there is nontrivial evidence. However, for long homologous reference regions, the number of possible variants is large, so the number of haplotypes (combinations of variants) becomes exponentially large, and the number of diplotype sets (combinations of variants) becomes exponentially large. haplotypes) can be astronomical.

[0490] Consequently, to keep MRJD calculations tractable, it may not be useful to test all possible joint diplotypes. Instead, in some cases, the system may be configured such that only a small subset of “most likely” joint diplotypes are tested. These “most likely” joint diplotypes may be determined by incrementally constructing a tree of partially determined joint diplotypes. In such a case, each node of the tree may be a partially determined joint diplotype that includes a partially determined haplotype by ploidy per homologous reference region. In such a case, a partially determined haplotype may include a reference region modified by a partially determined subset of the possible variants. Accordingly, a partially determined subset of the possible variants may include an indication, for each possible variant, of one of three states: that the variant is determined to be present, or the variant is determined to be absent, or the variant is not yet determined, e.g., it may be present or absent. At the root of the tree, all variants are undetermined in all haplotypes; tree nodes that branch off further successfully from the root have successively more variants determined to be present or absent in each haplotype of each diplotype set from the node.

[0491] Furthermore, in the context of such a joint diplotype tree as described above, the amount of MRJD computations is kept limited and tractable by pruning branches of the tree in which all joint diplotype nodes are unlikely, e.g., moderately to extremely unlikely, relative to other branches or more likely nodes. Accordingly, such pruning may be performed on branches at nodes that are as yet only partially determined; e.g., several or many variants not yet determined to be present or absent from the haplotypes of a pruned node joint diplotype. Thus, in such a case, it is useful to estimate or bound the probability of observing each read assuming a partially determined haplotype to be true. A modified pairwise hidden Markov model (pHMM) calculation, denoted "PDHMM" for "partially determined pairwise hidden Markov model", is useful for estimating the probability P(R|H) of observing read R assuming that the true haplotype H* is consistent with the partially determined haplotype H. Consistent in this context means that some partially determined haplotype H* agrees with the partially determined haplotype H with respect to all variants whose presence or absence is determined in H, but for variants undetermined in H, H* may agree with the reference sequence modified or unmodified by each undetermined variant.

[0492] It is noted that it is not generally appropriate to perform a common pHMM calculation for some shorter subhaplotype of H chosen to encompass only determined variant positions. It is generally important to construct the joint diplotype tree with undetermined variants that are resolved in an effective order that is generally quite different from their geometric order, so that a partially determined haplotype H typically has many undetermined variant positions interspersed with determined ones. To adequately account for PCR indel errors, it is useful to use a pHMM-like calculation that encompasses all determined variants and a significant radius around them, which may not be compatible with attempts to avoid undetermined variant positions.

[0493] Accordingly, inputs to the HMM PD may include the nucleotide sequence called the read R, the base quality scores (e.g., phred scale) of the nucleotides called R, a baseline haplotype HO, and a list of undetermined variants (edits) of HO. The undetermined variants may include single base substitutions (SNPs), multiple base substitutions (MNPs), insertions, and deletions. Advantageously, it may be appropriate to support both undetermined SNPs and deletions. An undetermined MNP may be imperfectly but adequately represented as multiple independent SNPs. An undetermined insertion may be represented by first editing the insertion in the baseline haplotype, then indicating the corresponding undetermined deletion that would undo the insertion.

[0494] Constraints may be placed on the undetermined deletions to facilitate the implementation of hardware engines with limited memory and state logic, such as that two undetermined deletions may not overlap (delete the same baseline haplotype bases). If a partially determined haplotype needs to be tested with undetermined variants that violate such constraints, this may be resolved by converting one or more undetermined variants to determined variants in a larger number of HMM PD operations, covering cases with those variants present or absent. For example, if two undetermined deletions A and B violate overlapping each other in baseline haplotype HO, then deletion B can be edited into HO to produce HOB, and two HMM PD operations can be performed using undetermined deletion A only, one for baseline haplotype HO and the other for baseline haplotype HOB, and the maximum likelihood output of the two HMM PD operations can be maintained.

[0495] The result of a HMM PD operation may be an estimate of the maximum P(R|H*) across all H* haplotypes that can be formed by editing HO with any subset of the undetermined variants. Maximization may be performed locally by contributing to pHMM-like dynamic programming in a given cell as if an adjacent undetermined variant were present or absent from the haplotype, whichever has the better score, i.e., contributes the greater partial probability. Such local maximization during dynamic programming may result in larger estimates of the maximum P(R|H*) than true maximization across individual pure H* haplotypes, but the difference is generally irrelevant.

[0496] Undetermined SNPs can be incorporated into HMM PD by allowing one or more matching nucleotide values to be specified for each haplotype position. For example, if base 30 of HO is “C” and an undetermined SNP replaces that “C” with a “T”, then the haplotype from HMM PD operation might indicate position 30 as corresponding to both bases “C” and “T”. In usual pHMM dynamic programming, any transition to an “M” state results in multiplying the path probability by the probability of a correct base call (if the haplotype position matches the read position) or by the probability of a specific base call error (if the haplotype position diverges from the read position); for HMM PD, this is modified by using the probability of correct call if the read position matches the possible haplotype base (e.g., “C” or “T”) and the probability of base call error otherwise.

[0497] Undetermined haplotype deletions can be incorporated into HMM PD by flagging optionally deleted haplotype positions and modifying the pHMM dynamic programming to allow alignment paths to skip horizontally across the undetermined deletion haplotype segments without loss of probability. This can be accomplished in a variety of ways, but with the common property that probability values in M, I, and/or D states can stream horizontally (along the haplotype axis) across the amplitude of an undetermined deletion without being reduced by common gap-opening and gap-extending probabilities.

[0498] In a particular embodiment, haplotype positions at which indeterminate deletions begin are flagged "F1" and positions at which indeterminate deletions end are flagged "F2". In addition to the M, I, and D “states” (partial probability representations) for each cell of the HMM matrix (horizontal haplotype/vertical readout), each HMM PD cell may additionally include BM, BI, and BD “deviation” states. In haplotype columns flagged by F1, the BM, BI, and BD states receive values copied from the M, I, and D states of the cell to the left, respectively. In haplotype columns not flagged by F2, particularly columns that begin with an F1-flagged column end that extends into an indeterminate deletion, the BM, BI, and BD states transmit their values to the BM, BI, and BD states of the cell to the right, respectively. In F2-signaled haplotype columns, instead of the M, I, and D states used to calculate adjacent cell states, the maximum of M and BM is used, and the maximum of I and BI is used, and the maximum of D and BD is used, respectively. This is exemplified in an F2 column as multiplexed selection of M and BM signals, of I and BI signals, and of D and BD records.

[0499] Note that although the BM, BI, and DB state records may be represented in columns F1 through F2, and the maximization of the M/BM, I/BI, and D/BD multiplexers may be shown in column F2, these components may be present for all cell calculations, allowing an indeterminate deletion to be handled at any position, and allowing indeterminate deletions with corresponding F1 and F2 flags throughout the haplotype. Note also that the F1 and F2 flags may be in the same column, for the case of a single base indeterminate deletion. It is further noted that the HMM PD array of cells may be represented as a schematic representation of the M, I, D, BM, BI, and BD logic state calculations, but in a hardware implementation, a smaller number of cell calculation logic elements may be present and appropriately segmented to calculate M, D, I, BM, BI, and BD state values at high clock frequencies, and the array cells may be calculated with varying degrees of hardware parallelism, in various orders consistent with the inherent logical dependencies of the HMM PD calculation.

[0500] Thus, in this embodiment, pHMM state values in a column may be immediately to the left of an indeterminate deletion that may be captured and passed rightward, unchanged, to the rightmost column of that indeterminate deletion, where they replace pHMM calculations whenever they achieve normal path scores. When these maxima are chosen, the "deviation" state values BM, BI, and BD represent local dynamic programming results where the indeterminate deletion is taken to be present, while "normal" state values M, I, and D represent results where the indeterminate deletion is taken to be the absent state.

[0501] In another embodiment, a single deviant state may be used, such as a BM state that receives from a state of M in columns flagged by F1 or that receives a sum of states of M, D, and/or I. In another embodiment, instead of using "deviant" states, gap opening and/or gap extension penalties are eliminated within columns of undetermined deletions. In another embodiment, deviant states contribute additively to rightward dynamic programming of undetermined deletions, rather than local maximization being used. In another embodiment, more or less or differently defined or differently located haplotype position flags are used to trigger deviance or similar behavior, such as a single flag indicating membership in an undetermined deletion. In a further embodiment, two or more overlapping undetermined deletions may participate, such as with the use of additional flags and/or deviant states. Additionally, undetermined insertions in the haplotype are supported instead of or in addition to undetermined deletions. Similarly, undetermined insertions and/or deletions in the reading axis are supported instead of or in addition to undetermined deletions and/or insertions in the haplotype axis. In another embodiment, undetermined multinucleotide substitutions are supported as atomic variants (all present or all absent). In another embodiment, undetermined variable length substitutions are supported as atomic variants. In another embodiment, undetermined variants are penalized with fixed or configurable probability or scoring adjustments.

[0502] This HMM PD computation may be implemented as a hardware engine, such as in FPGA or ASIC technology, by extension of a hardware engine architecture for "common" pHMM computation, or it may be implemented by one or more quantum circuits in a quantum computing platform. In addition to an engine segmentation logic for computing, transmitting, and storing M, I, and D state values for multiple, successive cells, parallel segmentation logic may be constructed for computing, transmitting, and storing BM, BI, and BD state values, as described herein and above. The memory resources and ports for storing and retrieving M, I, and D state values may be accompanied by similar or broader or deeper resources and ports for storing and retrieving BM, BI, and BD state values. Flags, such as F1 and F2, may be stored in memories along with associated haplotype bases.

[0503] Multiple nucleotides corresponding to, for example, undetermined SNP haplotype positions may be encoded in any manner, such as using a one-bit vector per possible nucleotide value. The cell calculation dependencies in the pHMM matrix are unchanged in HMM PD, so the order and segmentation of multiple cell calculations may remain the same for HMM PD. However, the latency in time and/or clock cycles to complete cell calculation increases somewhat for HMM PD, due to the requirement to compare "normal" and "deviation" state values and select the larger ones. Consequently, it may be advantageous to include one or more extra segmentation stages for HMM PD cell calculation, resulting in additional clock cycles of latency. Additionally, it may be additionally advantageous to extend each "range" of calculated cells by one or more rows, to keep the longer segmentation filled without dependency problems.

[0504] This HMM PD calculation tracks twice as many state values (BM, BI, and BD in addition to M, I, and D) as a regular pHMM calculation, and may require about twice the hardware resources for an equivalent throughput engine modality. However, a HMM PD engine has exponential speed and efficiency advantages for increasing numbers of undetermined variants, versus a regular pHMM engine run once for each haplotype that represents a distinct combination of undetermined variants that are present or absent. For example, if a partially determined haplotype has 30 undetermined variants, each of which may be independently present or absent, there are 2A30, or over 1 billion, distinct specific haplotypes that pHMM might otherwise need to process.

[0505] Accordingly, these and other such operations disclosed herein can be performed in order to better understand and accurately predict what happened to the subject's genome such that the reads varied from the reference. For example, although the occurrence of mutations may be random, there are cases in which the probability of their occurrence appears to be potentially predictable to some extent. In particular, in some cases, when mutations do occur, they may occur at certain defined locations and in certain ways. More particularly, mutations, if they occur, will occur at one allele or the other or both and will tend to occur at certain locations relative to others, such as at the ends of chromosomes. Accordingly, this and other associated information can be used to develop mutation models, which can be generated and used to better assess the likely presence of a mutation and one or more regions of the genome. For example, considering several a priori knowledge, for example, one or more mutation models, when performing genomic variation analyses, better and more accurate genomic analysis results can be obtained, such as with more precise demarcations of genetic mutation.

[0506] Such mutation models may generate an account of the frequency and/or location of multiple known mutations and/or mutations that appear to occur in conjunction with one another or otherwise non-randomly. For example, it has been determined that toward the ends of a given chromosome, variations occur predominantly. Thus, known mutation models may be generated, stored in a database herein, and used by the system to produce a better prediction of the presence of one or more variations within the genomic data being analyzed. Additionally, a machine learning process, as described in more detail below herein, may also be implemented such that multiple output data derived by the analyses performed herein may be analyzed and used to better inform the system as to whether to make a specific variance call, such as in accordance with the machine learning principles disclosed herein. Specifically, machine learning can be deployed on collective datasets, especially regarding the variances determined, and this learning can be used to generate more comprehensive mutation models, which in turn can be employed to produce more accurate variance determinations.

[0507] Thus, the system may be configured to observe all of the various variation data, mine that data for various correlations, and when correlations are found, such information may be used to better weight and therefore more accurately determine the presence of other variations in other genome samples, such as on an ongoing basis. Accordingly, in such a manner, the system, especially the variant calling engine, may be constantly updated with respect to the learned variant correlation data so as to produce better variant calls moving forward, so as to obtain better and more accurate result data.

[0508] Specifically, telemetry may be employed to update the growing mutation model of moo to achieve better analysis in the system. This may be of particular utility when analyzing samples that are in some way connected to each other, such as by being within the same geographic population, and/or may be used to determine which reference genome among a plurality of reference genomes may be a better reference genome by which a particular sample should be analyzed. Furthermore, in various cases, mutation modeling and/or telemetry may be employed in order to better select the reference genome to be used in the system processes and thus increase the accuracy and efficiency of the system results. In particular, when a plurality of reference genomes may be used in one or more of the analyses herein, a particular reference genome may be selected for use over others, such as by applying a mutation model in order to select the most appropriate reference genome to apply.

[0509] It is noted that when performing secondary analysis, the underlying structure for each region of the genome that is mapped and aligned may include one or more underlying genes. Accordingly, in many cases, this understanding of the underlying genes and/or the functions of the proteins they encode may be informative when performing secondary analysis. In particular, tertiary cues and/or results may be useful in secondary analysis protocols that are performed by the present system, such as in a biological contextually sensitive mutation modeling process. More particularly, since DNA encodes genes, where genes encode proteins, information about such proteins that result in repulsive mutations and/or functions may be used to inform mutation models that are employed in performing secondary and/or tertiary analyses on the subject's genome.

[0510] For example, tertiary analysis, such as on a set of sample genes encoding mutated proteins, may be informative when performing secondary analysis of genomic regions known to encode such mutations. Thus, as discussed above, various tertiary processing results may be used to inform and/or update the mutation models used herein to achieve improved accuracy and efficiency when performing the various secondary analysis operations disclosed herein. Specifically, information about mutated proteins, e.g., contextual tertiary analysis, may be used to update the mutation model when performing secondary analysis of those regions known to encode proteins and/or potentially include such mutations.

[0511] Accordingly, in view of the foregoing, for embodiments involving FPGA-accelerated mapping, alignment, classification, and/or variant calling applications, one or more of these functions may be implemented in one or both of the software and hardware (HW) processing components, such as software running on a CPU, GPU, QPU, and/or traditional firmware, as may be embedded in an FPGA, ASIC, sASIC, and the like. In such cases, the CPU and FPGA need to be able to communicate in order to pass the results of a step on one device, e.g., the CPU or the FPGA, to be processed in a subsequent step on the other device. For example, when a mapping function is executed, the construction of large data structures, such as an index of the reference, may be implemented by the CPU, wherein the execution of a hash function against the same may be implemented by the FPGA. In such a case, the CPU can construct the data structure, store it in an associated memory, such as DRAM, where the memory can then be accessed by the processing engines running on the FPGA.

[0512] For example, in some embodiments, communications between the CPU and the FPGA may be implemented over any suitable interconnect, such as a peripheral bus, such as a PCIe bus, USB, or a network interface, such as Ethernet. However, a PCIe bus may be a comparatively loose integration between the CPU and the FPGA, such that transmission latencies between the two may be relatively high. Consequently, although one device, for example, (the CPU or the FPGA) may access memory attached to the other device (e.g., by a DMA transfer), the memory region (or regions) accessed is not cacheable, due to the fact that there is no facility to maintain cache coherence between the two devices. As a consequence, transmissions between the CPU and the FPGA are restricted to occurring between large high-level processing steps, and a large amount of input and output needs to be queued between the devices so that they do not slow each other down waiting for high-latency operations. This slows down the various processing operations revealed in this paper. Furthermore, when the FPGA accesses non-cacheable CPU memory, the full burden of such access is imposed on the CPU's external memory interfaces, which are bandwidth-limited compared to its internal cache interfaces.

[0513] Consequently, due to such loose CPU/FPGA integrations, it is generally necessary to have "centralized" software control over the FPGA interface. In such cases, the various software threads may be processing multiple data units, but when these threads generate work for the FPGA engine to perform, the work needs to be aggregated into "central" temporary stores, such as by a single aggregator software thread, or by multiple threads blocking aggregation access via semaphores, with aggregated work being passed via DMA packets managed by a central software module such as a kernel space driver. Then, as results are produced by the HW engines, the reverse process occurs, with a software driver receiving DMA packets from the HW, and a deaggregator thread distributing the results to the various waiting software worker threads. However, this centralized software control of communication with the HW FPGA logic is impractical and expensive in resource usage, reduces the effectiveness of software threading and HW/software communication, limits the practical HW/software communication bandwidth, and dramatically increases its latency.

[0514] Additionally, as can be seen with respect to Figure 33A, loose integration between the CPU 1000 and the FPGA 7 may require each device to have its own dedicated external memory, such as DRAMs 1014, 14. As depicted in Figure 33A, the CPU(s) 1000 has its own DRAM 1014 on the system motherboard, such as DDR3 or DDR4 DIMMs, while the FPGA 7 has its own dedicated DRAMs 14, such as four 8 GB SODIMMs, which may be directly connected to the FPGA 7 via one or more DDR3 buses 6, such as a high latency PCIe bus. Similarly, the CPU 1000 may be communicatively coupled to its own DRAM 1014, such as by a suitably configured bus 1006. As noted above, the FPGA 7 may be configured to include one or more processing engines 13, wherein the processing engines may be configured to perform one or more functions in a bioinformatics segmentation as described herein, such as when the FPGA 7 includes a mapping engine 13a, an alignment engine 13b, and a variant calling engine 13c. Other engines as described herein may also be included. In various embodiments, one or both of the CPU may be configured to include a cache 1014a, 14a, respectively, that has the capacity to store data, such as result data that is transferred thereto by one or more of the various components of the system, such as one or more memories and/or processing engines.

[0515] Many of the operations disclosed herein to be performed by the FPGA 7 for genomic processing require large memory accesses to perform the underlying operations. Specifically, due to the large units of data involved, e.g., 3+ billion nucleotide reference genomes, 100+ billion nucleotides of sequencer read data, etc., the FPGA 7 may need to access the host memory 1014 a large number of times such as to access an index, such as a 30 GB hash table or other reference genome index, such as for the purpose of mapping the seeds of a sequenced DNA/RNA query to a 3 Gbp reference genome and/or to search for candidate segments, e.g., of the reference genome, to align against.

[0516] Accordingly, in various implementations of the system disclosed herein, many fast random memory accesses may need to occur by one or more of the wired processing engines 13, such as in the performance of a mapping, alignment, and/or variant calling operation. However, it may be prohibitively impractical for the FPGA 7 to perform so many small random accesses via the peripheral bus 3 or other network link to the memory 1014 attached to the host CPU 1000. For example, in such cases, the return data latencies may be very high, the bus efficiency may be very low, e.g., for such small random accesses, and the load on the CPU's external memory interface 1006 may be prohibitively large.

[0517] Additionally, as a result of each device needing its own dedicated external memory, the typical form factor of the complete CPU 1000 + FPGA 7 platform is forced to be larger than may be desirable, for example, for some applications. In such cases, in addition to a standard system motherboard for one or more CPUs 1000 and chips 7 and supporting memories, 1014 and/or 14, space is required on the board for a large FPGA package (which may need to be even larger in order to have enough pins for multiple external memory buses) and multiple memory modules, 1014, 14. Standard motherboards, however, do not include these components, nor could they easily have space for them, so a practical embodiment may be configured to utilize an expansion card 2, containing the FPGA 7, its memory 14, and other supporting components, such as a power supply, for example, connected to the PCIe expansion slot on the CPU motherboard. To make room for expansion card 2, the system can be manufactured to be in a large enough chassis, such as a 1U or 2U or larger rackmount server.

[0518] In view of the foregoing, in many instances, as can be seen with respect to Figure 33B, to overcome these factors, it may be desirable to configure the CPU 1000 to be in a tightly coupled arrangement with the FPGA 7. Particularly, in many instances, the FPGA 7 may be tightly coupled to the CPU 1000, such as by a low latency interconnect 3, such as a fast path interconnect (QPI). Specifically, to establish tighter CPU+FPGA integration, the two devices may be connected by any suitable low latency interface, such as a "processor interconnect" or the like, such as INTELS® Fast Path Interconnect (QPI) or HyperTransport (HT).

[0519] Accordingly, as seen with respect to Figure 33B, there is provided a system 1 wherein the system includes both a CPU 1000 and a processor, such as an FPGA 7, wherein both devices are associated with one or more memory modules. For example, as depicted, the CPU 1000 may be coupled, such as via a suitably configured bus 1006, to a DRAM 1014, and similarly, the FPGA 7 is communicatively coupled to an associated memory 14 via a DDR3 bus 6. However, in this case, rather than being coupled to each other, such as via a typical high latency interconnect, e.g., PCIe interface, the CPU 1000 is coupled to the FPGA 7 via a low latency hyper transport interconnect 3, such as a QPI. In such a case, due to the inherent low latency nature of such interconnects, the associated memories 1014, 14 of the CPU 1000 and the FPGA 7 are readily accessible to each other. Additionally, in various cases, due to this tight coupling configuration, one or more caches 1114a/14a associated with the devices may be configured to be coherent with respect to each other.

[0520] Some key properties of such a tightly coupled CPU/FPGA interconnect include a high bandwidth, e.g., 12.8 GB/s; low latency, e.g., 100 to 300 ns; an adapted protocol designed to enable efficient remote memory accesses and efficient small memory transfers, e.g., on the order of 64 bytes or less; and a supported protocol and CPU integration for cache access and cache coherence. In such cases, a natural interconnect to use for such tight integration with a given CPU 1000 may be its native CPU-to-CPU interconnect 1003, which may be employed herein to enable multiple cores and multiple CPUs to operate in parallel on a shared memory space 1014, thereby enabling access of each other's cache stacks and external memory in a cache-coherent manner.

[0521] Accordingly, as can be seen with respect to Figures 34A and 34B, a board 2 may be provided, such as wherein the board may be configured to receive one or more CPUs 1000, such as via a plurality of interconnects 1003, such as native CPU-CPU interconnects 1003a and 1003b. However, in this case, as depicted in Figure 34A, a CPU 1000 is configured to be coupled to interconnect 1003a, but instead of another CPU being coupled thereto via interconnect 1003b, an FPGA 7 of the disclosure is configured to be coupled thereto. Additionally, system 1 is configured such that CPU 1000 may be coupled to associated FPGA 7, such as by a low-latency tight-coupled interconnect 3. In such cases, each memory 1014, 14 associated with respective devices 1000, 7 may be designed to be accessible by one another, such as in a high-bandwidth cache-coherent manner.

[0522] Similarly, as can be seen with respect to Figure 34B, the system can also be configured to receive packets 1002a and/or 1002b, such as wherein each of the packets includes one or more CPUs 1000a, 1000b that are tightly coupled, for example, via low latency interconnects 3a and 3b, to one or more FPGAs 7a, 7b, such as wherein, given the system architecture, each packet 2a and 2b can be coupled to one another, such as via a tightly coupled interconnect 3. Furthermore, as can be seen with respect to Figure 35, in various instances, a package 1002a can be provided, wherein the package 1002a includes a CPU 1000 that has been manufactured in such a manner that it is tightly coupled to an integrated circuit, such as an FPGA 7. In such a case, Due to the close coupling of the CPU 1000 and the FPGA 7, the system may be constructed such that they may directly share a cache 1014a in a manner that is consistent, coherent, and readily accessible by any device, such as with respect to the data stored therein.

[0523] Thus, in such cases, the FPGA 7 and/or package 2a/2b may in effect masquerade as another CPU and thus operate in a cache-coherent shared memory environment with one or more CPUs, such as multiple CPUs on a multi-socket motherboard 1002 or multiple CPU cores within a multi-core CPU device. With such an FPGA/CPU interconnect, the FPGA 7 may effectively share CPU memory 1014, rather than having its own dedicated external memory 14, which may or may not be included and accessed. Thus, in such a configuration, fast, short, random accesses are effectively supported by the interconnect 3, such as with low latency. This makes it practical for the various processing engines 13 in the FPGA 7 to access large data structures in the memory of the CPU 1000.

[0524] For example, as can be seen with respect to Figure 37 , a system for performing one or more of the methods disclosed herein is provided, such as wherein the method includes one or more steps for performing the functions of the disclosure, such as one or more mapping and/or alignment and/or variant calling functions, as described herein, in a shared manner. Particularly, in a step (1), a data structure may be generated or otherwise provided, such as by an NGS and/or CPU 1000, wherein the data structure may then be stored in an associated memory (2), such as a DRAM 1014. The data structure may be any data structure, such as with respect to those described herein, but in this case may be a plurality of sequenced data reads and/or a reference genome and/or an index of the reference genome, such as for performing mapping and/or alignment and/or variant calling functions.

[0525] In a second step (2), such as in relation to mapping and/or alignment functions, etc., an FPGA 7 associated with the CPU 1000, such as via a tight coupling interface 3, may access CPU associated memory 1014 so as to perform one or more actions in relation to the stored sequenced reads, reference genome (or genomes), and/or an index thereof. Particularly, in a step (3), for example, in an exemplary mapping operation, the FPGA 7 may access a data structure, e.g., the sequenced reads and/or reference sequences, so as to produce one or more seeds therefrom, such as wherein the data structure includes one or more genome reference reads and/or sequences. In such a case, the seeds, for example, or the reference sequences and/or reads may be employed for the purposes of performing a hash function in relation thereto, such as to produce one or more reads that have been mapped to one or more positions in relation to the reference genome.

[0526] In another step (3), the mapped result data may be stored, for example, in host memory 1014 or in an associated DRAM 14. Additionally, once the data has been mapped, the FPGA 7, or a processing engine 13 thereof, may be reconfigured, for example, partially reconfigured, as an alignment engine, which may then access the stored mapped data structure so as to perform an alignment function thereon, so as to produce one or more reads that have been aligned to the reference genome. In a further step (4), the host CPU may then access the mapped and/or aligned data so as to perform one or more functions thereon, such as for producing a De Brujin Graph (“DBG”), wherein the DBG may then be stored in its associated memory. Similarly, in one or more additional steps, the FPGA 7 may again access the host CPU memory 1014 in order to access the DBG and perform an HMM analysis on it in order to produce one or more variant call files.

[0527] In particular cases, the CPU 1000 and/or the FPGA 7 may have one or more memory caches, which due to the tight coupling of the interface between the two devices, will allow separate caches to be coherent, such as with respect to transition data, e.g., results data, stored therein, such as the results of the performance of one or more functions herein. In such a manner, data may be shared in a substantially seamless manner between the tightly coupled devices, thereby allowing a range of functions to be interwoven together, such as in a bioinformatics range. Thus, in such a case, it may no longer be necessary for the FPGA 7 to have its own dedicated external memory, and thus, due to such a tightly coupled configuration, the stored reads, reference genome, and/or reference genomic index, as described herein, may be intensively shared, e.g., in a cache-coherent manner, such as for read mapping and alignment and other genomic data processing operations.

[0528] Additionally, as can be seen with respect to Figure 38 , low latency and cache coherency configurations, as well as other component configurations discussed herein, allow for smaller, lower-level operations to be performed on one device (e.g., on a CPU or FPGA) before returning a data structure or processing chain 20 to the other device, such as for further processing. For example, in one case, a CPU chain 20a may be configured to queue large amounts of work for the FPGA hardware logic 13 to process, and the same or different chain 20b may be configured to then process the large queue of result data thus generated, such as at a substantially later time. However, in many cases, it may be more efficient, as disclosed herein, for a single CPU chain 20 to perform a blocking "function call" to an attached FPGA hardware engine 13, where the CPU can be set to resume software execution once the FPGA hardware function is complete. Thus, rather than packetizing data structures for continuous transmission via DMA 14 to FPGA 7 and stripping the results from the packets when they return, a software chain 20 could simply provide a memory pointer to FPGA engine 13, which could access and modify shared memory 1014/14 in-place in a cache-coherent manner.

[0529] In particular, given the relationship between the structures provided herein, the granularity of the software/hardware cooperation may be much finer, with smaller lower-level operations being allocated to be performed by multiple hardware engines 13, such as function calls from multiple allocated software chains 20. For example, in a loose CPU/FPGA interconnect platform, for effective acceleration of DNA/RNA read mapping, alignment, and/or variant calling, a complete mapping/alignment/variant calling pipeline may be constructed as one or more software and/or FPGA engines, with unmapped and unaligned reads being continuously streamed from software to hardware, and fully mapped and aligned reads being continuously streamed from hardware back to hardware, where the process may be repeated, such as for variant calling. With respect to the configurations described herein, this may be very fast. However, in many cases, such a system may suffer from limitations in flexibility, complexity and/or programmability, such as due to the fact that the entire mapping/alignment and/or variant calling segmentation is implemented in hardware circuitry, which, although reconfigurable in an FPGA, is generally much less flexible and programmable than software and may therefore be limited to lower algorithmic complexity.

[0530] In contrast, using a tight CPU/FPGA interconnect, such as a QPI or other interconnect in the configurations disclosed herein, various resource-intensive discrete operations, such as seed generation and/or mapping, rescue scanning, gapless alignment, gapped alignment, e.g., Smith-Waterman alignment, etc., may be implemented as distinct, separately accessible hardware engines 13, e.g., see Figure 38 , and the general mapping/alignment and/or variant calling algorithms may be implemented in software, with low-level acceleration calls to the FPGA for the specific expensive processing steps. This structure allows for full software programmability outside of the specific acceleration calls, and allows for greater algorithmic complexity and flexibility than operations implemented in hardware.

[0531] Furthermore, in such a software execution framework accelerated by discrete low-level FPGA hardware acceleration calls, hardware acceleration functions may be more easily shared for multiple purposes. For example, when hardware engines 13 form large monolithic segmentations, the individual segmentation subcomponents may generally be specialized for their environment and interconnected only within a segmentation, which, unless tightly coupled, may not generally be accessible for any purpose. But many genomic data processing operations, such as Smith-Waterman alignment, gapless alignment, de Bruijn graph construction or assembly, and other such operations, may be used in multiple original higher-level algorithms. For example, as described herein, Smith-Waterman alignment can be used in DNA/RNA read alignment and mapping, such as against a reference genome, but can also be configured to be used by haplotype-based variant callers to align candidate haplotypes to a reference genome, or to each other, or to sequenced reads, such as in an HMM analysis and/or variant calling function. Thus, exposing multiple discrete low-level hardware acceleration functions through general software function calls can enable the same acceleration logic, e.g., 13, to be leveraged throughout a genomic data processing application, such as in the performance of both alignment and variant calling operations, e.g., HMM.

[0532] It is also practical, with tight CPU/FPGA interconnection, to have software control of CPU 1000 distributed rather than centralized through communication with the various FPGA hardware engines 13 described herein. In widespread multi-thread, multi-core, multi-CPU software design practices, many software threads and processes communicate and cooperate continuously, without any central software modules, drivers, or threads to manage the intercommunication. In such a format, this is practical due to cache-coherent shared memory, which is visible to all threads on all cores of the CPUs; while physically coherent memory sharing between cores and CPUs occurs by intercommunication via processor interconnect, e.g., QPI or HT.

[0533] In a similar manner, as can be seen with respect to Figures 36-38, the systems provided herein may have multiple CPUs and/or FPGAs that may be in a CPU/FPGA tight interconnect configuration that incorporates a plurality of chains, e.g., 20a, b, c, and a plurality of processes executing on one or multiple cores and/or CPUs, e.g., 1000a, 100b, and 1000c. As such, the system components are configured to communicate and cooperate in a distributed manner with each other, e.g., among the various different CPU and/or FPGA hardware acceleration engines, such as using cache-coherent memory sharing among the various CPUs and FPGAs. For example, as can be seen with respect to Figure 36 , a plurality of CPU cores 1000a, 1000b, and 1000c can be coupled together in a manner that they share one or more memories, e.g., DRAMs 1014, and/or one or more caches having one or more layers, e.g., L1, L2, L3, etc., or levels associated therewith. Similarly, with respect to Figure 38 , in another embodiment, a single CPU 1000 can be configured to include multiple cores 1000a, 1000b, and 1000c that can be coupled together in such a manner that they share one or more memories, e.g., DRAMs 1014, and/or one or more caches, 1014a, having one or more layers or levels associated therewith.

[0534] Thus, in any embodiment, data to be passed from one or more software chains 20 of one or more CPU cores 1000 to a hardware engine 13, e.g., an FPGA, or vice versa, may be continuously and/or uninterruptedly updated in shared memory 1014, or a cache and/or layer thereof, that is visible to each device. Additionally, requests to process data in shared memory 1014, or notification of updated results therein, may be signaled between the software and/or hardware, such as through a suitably configured bus, e.g., DDR4 bus, or in queues that may be implemented within the shared memory itself. Standard software mechanisms for data control, transfer, and protection, such as semaphores, mutexes, and atomic integers, may also be similarly implemented for software/hardware coordination.

[0535] Accordingly, in some embodiments, as exemplified in Figure 36, without the need for the FPGA 7 to have its own dedicated memory 14, or other external sources, due to cache-coherent memory sharing via a tight CPU/FPGA interconnect, it becomes much more practical to package the FPGA 7 more compactly and natively within traditional CPU motherboards 1000, without the use of expansion cards. See, for example, Figures 34A and 34B and Figure 35. Several package design alternatives are available. Specifically, an FPGA 7 may be installed on a multiple CPU motherboard in a CPU socket as shown in Figures 34A and 34B, such as by using a suitable interposer such as a small PC board 2, or by creating an alternative wire-bond package of the FPGA die within the CPU chip package 2a, wherein the CPU socket pins are suitably routed to the FPGA pins and include power and ground connections, a processor interconnect 3 (QPI, HT, etc.), and other system connections. Accordingly, an FPGA die and a CPU die may be included in the same multi-chip package (MCP) with the necessary connections, including power, ground, and CPU/FPGA interconnect, made within the package 2a. Inter-die connections may be by wire-bonding from die to die, or by connection to a common substrate or interposer, or by bonded pads or vias through silicon between stacked dies.

[0536] Additionally, in various implementations, the FPGA and CPU cores may be fabricated on a single die, see Figure 35, using a system-on-a-chip (SOC) methodology. In any such case, custom logic, e.g., 17, may be instantiated within the FPGA 7 to either communicate over the CPU/FPGA interconnect 3, e.g., via suitably dedicated protocols, or to service, convert, and/or route memory access requests from internal FPGA engines 13 to the CPU/FPGA interconnect 3, via suitably dedicated protocols, to shared memory 1014a. Additionally, some or all of such logic may be hardened in custom silicon to avoid using real FPGA logic state for this purpose, such as where the hardened logic may persist on the CPU die, and/or on the FPGA die, or separate die. Furthermore, in either of these cases, power supply and heat dissipation requirements can be adequately met, such as within a single package (MCP or SOC). Furthermore, the FPGA size and CPU core count can be selected to be within a safe power envelope, and/or dynamic methods (clock rate management, clock switching, core disabling, power islands, etc.) can be used to regulate power consumption according to the switching demands of the CPU and/or FPGA.

[0537] All of these packaging design options share several advantages. The tightly integrated CPU/FPGA platform becomes compatible with standard motherboards and/or system chassis of a variety of sizes. If the FPGA is installed via an interposer into a CPU socket, see Figures 34A and 34B, then at least one dual-socket motherboard 1002 may be employed. In other cases, a quad-socket motherboard may be employed to allow configurations of 3 CPUs + 1 FPGA, 2 CPUs + 2 FPGAs, or 1 CPU + 3 FPGAs, etc., to be deployed. If each FPGA resides in the same chip package as a CPU (MCP or SOC), then a single-socket motherboard may be employed, potentially in a very small chassis (although a dual-chassis motherboard is depicted); the same can also be scaled up very well, for example 4 FPGAs and 4 multi-core CPUs on a 4-socket motherboard, which could however operate in a compact chassis such as a 1U rackmount server.

[0538] Accordingly, in many cases, therefore, there may not be a need for an expansion card to be installed in order to integrate CPU and FPGA acceleration, because the FPGA 7 can be integrated into the CPU socket 1003. This implementation avoids the extra space and power requirements of an expansion card and avoids several additional points of failure that expansion cards sometimes have with respect to relatively low-reliability components. Furthermore, standard CPU cooling solutions (heat exchangers, heat pipes, and/or fans), which are effective and yet low-cost since they are manufactured in high volumes, can be applied to FPGAs or CPU/FPGA packages in CPU sockets, whereas cooling for expansion cards can be expensive and ineffective.

[0539] Similarly, an FPGA/intermediate and/or CPU/FPGA package may include the full power supply from a CPU socket, e.g. 150 W, while a standard expansion card may be power-limited to e.g. 25 W or 75 W from a PCIe bus. In many cases, for genomic data processing applications, all of these packaging options can facilitate easy installation of a tightly integrated CPU+FPGA computing platform, such as within a DNA sequencer. For example, typical modern "next-generation" DNA sequencers contain the sequencing apparatus (sample and reagent storage, fluidics tubing and control, sensor arrays, primary image and/or signal processing) within a chassis that also contains a standard or custom server motherboard, wired to the sequencing apparatus for sequencing control and data acquisition. A tightly integrated CPU+FPGA platform as described herein may be achieved in such a sequencer such as by simply installing one or more FPGA/intermediate and/or FPGA/CPU packages into CPU sockets on your existing motherboard, or alternatively by installing a new motherboard with both the CPU(s) and FPGA(s), for example, tightly coupled, as disclosed herein. Furthermore, all of these package design options may be configured to facilitate easy deployment of the tightly integrated CPU+FPGA platform, such as in a cloud-accessible and/or data center server shelf that includes compact/dense servers with very high reliability/availability.

[0540] Thus, in accordance with the teachings herein, there may be many processing stages for DNA (or RNA) sequencing data from mapping and alignment to classification and/or deduplication to variant calling, which may vary depending on the primary and/or secondary and/or tertiary processing technologies employed and their applications. Such processing steps may include one or more of: signal processing on electrical measurements from a sequencer, image processing on optical measurements from the sequencer, base calling using processed signal or image data to determine the most likely nucleotide sequence and confidence scores, filtering low-quality sequenced reads or polyclonal clusters, detecting and trimming adapters, key sequences, barcodes, and low-quality read ends, as well as de novo sequence assembly, generating and/or utilizing De Bruijn graphs and/or sequence graphs, e.g., De Bruijn and sequence graph construction, editing, trimming, cleaning, repairing, coloring, annotating, comparing, transforming, splitting, splicing, parsing, subgraph selection, traversal, iteration, recursion, searching, filtering, importing, exporting, including mapping reads to a reference genome, aligning reads to candidate mapping sites in the reference genome, locally assembling reads mapped to a reference region, sorting reads by aligned position, tagging and/or removing duplicate reads, including PCR or optical duplicates, realignment of multiple overlapping reads for indel consistency, base quality score recalibration, variant calling (single or pooled sample), structural variant analysis, copy number variant analysis, somatic variant calling (e.g., tumor sample only, matched tumor/normal or mismatched tumor/normal, etc.), RNA splice junction detection, alternative RNA splicing analysis, RNA transcript assembly, RNA transcript expression analysis, RNA differential expression analysis, RNA variant calling, DNA/RNA mismatch analysis, DNA methylation analysis and calling, variant quality score recalibration, variant filtration, variant annotation from known variant databases, sample contamination detection and estimation, phenotype prediction, disease testing, treatment response prediction, customized treatment protection, ancestry and mutation history analysis, population DNA analysis, marker identification genetic, encoding genomic data into standard formats and/or compression files (e.g., FASTA, FASTQ, SAM, BAM, VCF, BCF), decoding genomic data from standard formats, querying, selecting or filtering subsets of genomic data, general compression and decompression for genomic files (gzip, BAM compression), specialized compression and decompression for genomic data (CRAM), encryption and decryption of genomic data, calculation of statistics, comparison and presentation of genomic data, comparison of genomic data results, accuracy analysis and reporting, genomic file storage, archiving, rescue, backup, retrieval and transmission, as well as genomic database construction, searching, access management, data extraction and the like.

[0541] All of these operations can be quite slow and expensive when implemented on traditional computing platforms. The slowness of such operations implemented solely in software can be due in part to the complexity of the algorithms, but is typically due to very large input and output data sets that result in high latency in moving the data. The devices and systems disclosed herein overcome these problems in part due to the configuration of the various hardware processing engines, acceleration by the various hardware implementations, and/or in part due to tightly coupled CPU/FPGA configurations. Accordingly, as can be seen with respect to Figure 39, one or more, e.g., all of these operations, can be accelerated by cooperation of CPUs 1000 and FPGAs 7, such as in a distributed processing model as described herein. For example, in some cases (encryption, general compression, read mapping and/or alignment), an entire operational function may be implemented substantially entirely in custom FPGA logic (such as per hardware design methodology, e.g., RTL), such as where the CPU software primarily serves the function of compiling large packets of data for preprocessing by worker chains 20, such as aggregating the data into multiple jobs to be processed by one or more hardware-implemented processing engines, and feeding the multiple data inputs, such as in a first format, to one or more of the FPGA engine(s) 13, and/or receiving the results thereof.

[0542] For example, as can be seen with respect to Figure 39 , in various embodiments, a worker chain generates multiple packets of work data that may be compiled and/or continuously transmitted into larger work packets that may be queued and/or further aggregated in preparation for transfer, e.g., via a DDR3 to the FPGA 7, such as via a high-bandwidth, low-latency, point-to-point interconnect protocol, e.g., QPI 3. In particular cases, data may be temporarily stored in accordance with the particular data sets that are transferred to the FPGA. Once the packet data is received by the FPGA 7, such as in a cache-coherent manner, it may be processed and sent to one or more specialized clusters 11 so that it may be further directed to one or more processing pools for processing thereby in accordance with one or more of the segmentation operations described herein.

[0543] Once processed, the result data may then be sent back to the cluster and queued to be sent back through the tightly coupled point-to-point interconnect to the CPU for post-processing. In certain embodiments, the data may be sent to a disaggregator chain prior to post-processing. Once post-processing has occurred, the data may be sent back to the initial worker chain 20 that may be waiting for the data. Such distributed processing is particularly beneficial for the functions disclosed above herein. In particular, these functions are distinguishable by the facts that their algorithmic complexity (while having a very high net computational load) is quite limited and they may each be configured so as to have a reasonably uniform computation cost across their various sub-operations.

[0544] However, in many cases, rather than processing the data in large packets, smaller subroutines or protocols or discrete function elements may be realized, such as pertaining to one or more functions of a segmentation, rather than performing the entire processing functions for that segmentation on that data. Thus, a useful strategy may be to identify one or more critical computationally intensive subfunctions in any given operation, and then implement that subfunction in custom FPGA logic (hardware acceleration), such as for the intensive subfunction(s), while implementing the balance of the operation, and ideally much or most of the logarithmic complexity, in software for execution on CPUs/GPUs/QPUs, as described herein, such as with respect to Figure 39.

[0545] In general, it is typical of genomic data processing operations that a small percentage of the algorithmic complexity represents a large percentage of the overall computational load. For example, as a typical example, 20% of the algorithmic complexity for performing a given function may represent 90% of the computational load, while the remaining 80% of the algorithmic complexity may only represent 10% of the computational load. Thus, in many cases, the system components described herein may be configured to deploy the high complexity portion, e.g., 20% or more, to be executed very efficiently on custom FPGA logic that may be tractable and maintainable in a hardware design, and thus may be configured to execute the same on the FPGA; which in turn may reduce the CPU computational load by 90%, thereby enabling 10x speedups. Other typical examples may be even more extreme, such as where 10% of the algorithmic complexity may represent 98% of the computational burden, in which case applying FPGA acceleration as described in this paper to the 10% complexity portion will be even easier, but may also enable a net speedup of 50x. In many cases where extremely accelerated processing is desired, one or more of these functions may be performed by a quantum processing unit.

[0546] However, such "chunked" or distributed processing acceleration approaches may be more practical when implemented on a highly integrated CPU/GPU+FPGA platform, rather than a loosely integrated CPU/GPU+FPGA platform. Particularly, on a loosely integrated platform, the portion, e.g., functions, to be implemented in FPGA logic may be selected so as to minimize the size of the input data to the FPGA engine (or engines) and to minimize the output data from the FPGA engine (or engines), such as for each unit of data processed, and further may be configured so as to maintain the high-latency tolerant software/hardware boundary. In such cases, the boundary between the hardware and software portions may be forced, e.g., on the loosely integrated platform, until it is extracted through certain low-bandwidth/high-latency cutoff points, the divisions of which may not otherwise be desirable when optimizing the partitioning of algorithmic complexity and computational loads. This can often result in either pushing the boundaries of the hardware portion, encompassing an undesirably large portion of the algorithmic complexity in the wired format, or shrinking the boundaries of the hardware portion, undesirably excluding portions with dense computation load.

[0547] In contrast, in a tightly integrated CPU/GPU+FPGA platform, due to the cache-coherent shared memory and the high-bandwidth/low-latency CPU/GPU/FPGA interconnect, the low complexity/high compute load portions of a genomic data processing operation can be very precisely selected for deployment in custom FPGA logic (e.g., via the hardware engine(s) described herein) with optimized software/hardware constraints. In such a case, even if a data unit is large within the desired software/hardware constraint, it can still be efficiently delivered to the FPGA hardware engine for processing by simply passing a pointer to the particular data unit. In particular, in such a case, in accordance with Figure 33B, the hardware engine 13 of the FPGA 7 may not need to access each element of the data unit stored within the DRAM 1014; instead, it can access the necessary elements, for example, within cache 1014a, with small efficient accesses via the low latency interconnect 3' served by the CPU/GPU cache, thereby consuming less aggregate bandwidth than if the entire data unit had to be accessed and/or transferred to FPGA 7, such as by DMA from DRAM 1014, via a loose interconnect 3, in accordance with Figure 33A.

[0548] In such cases, the hardware engine 13 may write the processing results to the data unit in place in CPU/GPU memory 1014, without continuously DMAing an entire copy of the data unit to the CPU/GPU memory. Even if the desired software/hardware limit is not suitable for a software chain 20 to perform a high-latency, non-blocking queued delivery to the hardware engine 13, it may potentially perform a blocking function call to the hardware engine 13, suspending for a short latency until the hardware engine completes, wherein the latency is dramatically reduced by the cache-coherent shared memory, the low-latency/high-bandwidth interconnect, and the distributed software/hardware coordination model, as in FIG. 33B.

[0549] In particular cases, because specific signal/image processing and base calling algorithms and requirements vary from one sequencer technology to another, and because the amount of raw sequencer sensor data is typically gigantic (the same being reduced to huge after signal/image processing and to merely large after base calling), such signal/image processing and base calling can be efficiently performed within the sequencer itself or on a computing server connected by a high-bandwidth transmission channel to the sequencer. However, DNA sequencers have been achieving increasingly high throughput, at a rate of increase exceeding Moore's Law, such that existing central processing unit ("CPU") and/or Graphics Processing Unit ("GPU")-based signal/image processing and base calling, when deployed individually and alone, have become increasingly inadequate for the task. However, since a tightly integrated CPU+FPGA and/or GPU+FPGA and/or GPU/CPU+FPGA platform can be configured to be compact and easily instantiated within such a sequencer, e.g., as a CPU and/or GPU and/or FPGA chip positioned on the sequencer's motherboard, or easily installed on a server adjacent to the sequencer, or a cloud-based server system remotely accessible from the sequencer, such a sequencer may be an ideal platform for delivering the massive compute acceleration offered by the custom FPGA/ASIC hardware engines described herein.

[0550] For example, the system provided herein may be configured to perform primary, secondary, and/or tertiary processing, or portions thereof, to be implemented by a CPU, GPU, and/or FPGA; CPU + FPGA; GPU + FPGA; GPU/CPU + FPGA; QPU; CPU/QPU; GPU/QPU; CPU and/or GPU and/or QPU + accelerated FPGA platform. Furthermore, such accelerated platforms, e.g., including one or more FPGA and/or QPU hardware engines, are useful for deployment in cloud-based systems as described herein. For example, signal/image processing, base calling, mapping, alignment, classification, deduplication, and/or variant calling algorithms, or portions thereof, generally require large amounts of oscillating point and/or fixed point computations, notably additions and multiplications. These functions can also be configured to be performed by one or more quantum processing circuits, such as to be implemented in a quantum processing platform.

[0551] In particular, large modern FPGAs/quantum circuits contain thousands of high-speed multiplication and addition resources. More particularly, these circuits may include custom engines that may be deployed on or by them, wherein the custom engines may be configured to perform parallel arithmetic operations at rates that far exceed the capabilities of simple general-purpose CPUs. Similarly, simple GPUs have more comparable parallel arithmetic resources. However, GPUs often have problematic architectural limitations and programming constraints that may prevent them from being fully utilized. Accordingly, these FPGA and/or quantum processing and/or GPU arithmetic resources may be hard-wired or otherwise configured by design to operate in exactly the same manner as designed with near 100% efficiency, such as to perform the computations required to perform the functions herein. Consequently, GPU cards can be added to expansion slots on a motherboard with a tightly integrated CPU and/or FPGA, thus allowing all three types of processor to cooperate, although the GPU can still cooperate with all of its limitations and the limitations of loose integration.

[0552] More particularly, in various instances, with respect to Graphics Processing Units (GPUs), a GPU may be configured to implement one or more of the functions as described herein so as to accelerate the processing speed of the underlying computations required to perform that function, in whole or in part. More particularly, a GPU may be configured to perform one or more tasks in a mapping, alignment, sorting, deduplication, and/or variant calling protocol, such as to accelerate one or more of the computations, e.g., the large amounts of oscillating point and/or fixed point computations such as additions and multiplications involved therein, so as to work in conjunction with a server CPU and/or FPGA to accelerate application and processing performance and shorten the computational cycles required to perform such functions. Cloud servers as described herein with GPU/CPU/FPGA cards may be configured so as to easily handle compute-intensive tasks and deliver a smoother user experience when leveraged for virtualization. Such computationally intensive tasks can also be offloaded to the cloud, such as to be performed by a quantum processing unit.

[0553] Accordingly, if a tightly integrated CPU+FPGA or GPU+FPGA and/or CPU/GPU/FPGA platform with shared memory is employed with a sequencer, or a fixed or cloud-based server, such as for signal/image processing, base calling, mapping, alignment, classification, deduplication, and/or variant calling functions, there may be an advantage achieved, such as in an incremental development process. For example, initially, a limited portion of the compute load, such as a dynamic programming function for base calling, mapping, alignment, classification, deduplication, and/or variant calling, may be deployed on one or more FPGA engines, where other work may be performed on the CPU and/or GPU expansion cards. However, the tight integration model of CPU/GPU/FPGA and shared memory presented herein can be further configured later to make it easy to incrementally select additional compute-intensive functions for GPU, FPGA, and/or quantum acceleration, which can then be deployed as processing engines, and several of their functions can be offloaded for execution on the FPGA(s) and/or in some cases can be offloaded to the cloud, e.g., for performance by a QPU, thereby accelerating signal/image/base calling/mapping/alignment/variant processing. Such incremental advancements can be deployed as needed to keep pace with the increasing throughput of various primary and/or secondary and/or tertiary processing technologies.

[0554] Thus, read mapping and alignment, e.g., of one or more reads to a reference genome, as well as classification, deduplication, and/or variant calling may benefit from such GPU and/or FPGA and/or QPU acceleration. Specifically, mapping and alignment and/or variant calling, or portions thereof, may be implemented partially or entirely as custom FPGA logic, such as by continuously streaming "to-be-mapped and/or aligned and/or variant-called" reads from CPU/GPU memory to FPGA mapping/alignment/variant-calling engines, and continuously streaming mapped and/or aligned and/or variant-called read records back, which may further be continuously streamed back on-board, such as in the performance of classification and/or variant calling. This type of FPGA acceleration runs on a loosely integrated CPU/GPU+FPGA platform and in the configurations described in this paper can be extremely fast. However, there are some additional advantages that can be gained by moving to a tightly integrated CPU/GPU/QPU+FPGA platform.

[0555] Accordingly, with respect to mapping and alignment and variant calling, in some embodiments, a shared advantage of a tightly integrated CPU/GPU+FPGA and/or quantum processing platform as described herein is that the acceleration of mapping/alignment/variant calling, e.g., hardware acceleration, can be effectively divided into several discrete compute-intensive operations, such as seed generation and/or mapping, seed chain formation, paired-end rescue scans, gapless alignment and gapped alignment (Smith-Waterman or Needleman-Wunsch), De Bruijn graph formation, performing an HMM computation, and the like, such that the CPU and/or GPU and/or quantum computing software performs lighter-weight (but not necessarily less complex) tasks and can perform acceleration calls to hardware and/or other discrete quantum computing engines as needed. Such a model may be less effective on a typical loosely integrated CPU/GPU+FPGA platform, for example, due to the large amounts of data to round-trip between stages and high latencies, but may be more effective on a tightly integrated CPU+FPGA, GPU+FPGA, and/or quantum computing platform with cache-coherent shared memory, high-bandwidth/low-latency interconnect, and a distributed software/hardware coordination model. Additionally, as with variant calling, both Hidden Markov Model (HMM) and/or dynamic programming (DP) algorithms, including Viterbi and direct algorithms, may be deployed in association with a base calling/mapping/alignment/sorting/deduplication operation, such as to compute the most likely original sequence that explains the observed sensor measurements, in a configuration so as to be well suited to the parallel cellular model of FPGAs and quantum circuits described herein.

[0556] Specifically, efficient utilization of hardware and/or software resources in a distributed processing configuration may result from reducing hardware acceleration and/or quantum computing to discrete compute-intensive functions. In such cases, several of the functions disclosed herein may be performed in a monolithic pure hardware engine so as to be less compute-intensive, but may nevertheless still be algorithmically complex and thus may consume large amounts of physical FPGA resources (lookup tables, flip-flops, lock RAMs, etc.). In such cases, moving a portion or all of these various discrete functions to software could take up available CPU cycles in return for giving up substantial amounts of FPGA area. In certain such cases, the freed-up FPGA area may be used to establish greater parallelism for the compute-intensive mapping/alignment/variant calling subfunctions, thereby increasing speedup, or for other genomic acceleration functions. Such benefits can also be achieved by deploying computationally intensive functions on one or more quantum circuits dedicated for deployment by a quantum computing platform.

[0557] Thus, in various embodiments, the algorithmic complexity of the one or more functions disclosed herein may be somewhat reduced by being configured in a pure hardware or pure quantum computing implementation. However, some operations, such as comparing candidate alignment pairs for paired-end reads and/or performing fine-grained mapping quality (MAPQ) estimation, represent very low computational overheads and thus could benefit from more complex and precise processing in CPU/GPU and/or quantum computing software. Thus, in general, reducing hardware processing to specific compute-intensive operations could allow more complex and precise algorithms to be employed in the CPU/GPU portions.

[0558] Furthermore, in various embodiments, all or a portion of the mapping/alignment/classification/deduplication/variant calling operations disclosed herein could be configured in such a manner that more algorithmically complex computations can be performed at higher levels in hardware and/or via one or more quantum circuits, such as where the computationally intensive hardware and/or quantum functions called are configured to be performed in a dynamic or iterative order. In particular, a monolithic pure hardware/quantum processing design could be implemented in a manner that functions most effectively as a linear segmentation. For example, if during processing a Smith-Waterman alignment has exhibited evidence of the true alignment path escaping the scoring band, e.g., band as described above, another Smith-Waterman alignment could be called to correct for the same. Thus, these configurations could essentially reduce FPGA hardware/quantum acceleration to discrete functions, as a form of procedural abstraction, which could allow higher-level complexity to be easily built on top of them.

[0559] Additionally, in many cases, flexibility within mapping/alignment/variant calling algorithms and features thereof can be improved by reducing hardware and/or quantum acceleration to discrete compute-intensive functions and configuring the system to perform other, e.g., less intensive portions, in CPU and/or GPU software. For example, although hardware algorithms can be modified and reconfigured in FPGAs, such changes to hardware designs, e.g., via firmware, can generally require several times more design effort than similar changes to software code. In such cases, the compute-intensive portions of mapping and alignment and classification and deduplication and/or variant calling, such as seed mapping, seed chain formation, paired-end rescue scans, gapless alignment, gapped alignment, and HMM, which are relatively well defined, are thus stable functions and do not require frequent algorithmic changes. These functions can therefore be suitably optimized in hardware, while other functions, which could be performed by CPU/GPU software, are better suited for incremental algorithmic improvement, which is significantly easier in software. However, once fully developed, they could be implemented in hardware. One or more of these functions could also be configured to be implemented in one or more quantum circuits of a quantum processing machine.

[0560] Accordingly, in many cases, variant calling (with respect to DNA or RNA, single or pooled sample, germline or somatic, etc.) can also benefit from FPGA and/or quantum acceleration, as can its various computationally intensive functions. For example, haplotype-based callers, which call bases on evidence derived from a provided context within a window around a potential variant, as described above, are often the most computationally intensive operation. Such operations include comparing a candidate haplotype (e.g., a single-stranded nucleotide sequence that represents a theory of the true sequence of at least one of the sampled strands at the genome locus in question) to each sequencer read, such as to estimate a conditional probability of observing the read given the true haplotype.

[0561] Such operation may be performed by one or more of a MRJD, Pairwise Hidden Markov Model (Paired HMM) and/or a Pairwise Hidden Markov Model (PDHMM) calculation that sums the probabilities of possible combinations of errors in sequencing or sample preparation (PCR, etc.) by a dynamic programming algorithm. Thus, in relation thereto, the system may be configured such that a pairwise HMM or PDHMM calculation may be accelerated by one or more, e.g., parallel, FPGA hardware or quantum processing engines, while the CPU/GPU/QPU software may be configured to execute the remainder of the original haplotype-based variant calling algorithm on a loosely integrated or tightly integrated CPU+FPGA, or GPU+FPGA, or CPU and/or GPU+FPGA and/or QPU platform. For example, in loose integration, software chains can construct and prepare a De Bruijn and/or assembly graph from reads overlapping a chosen hot region (a window or contiguous subset of the reference genome), extract candidate haplotypes from the graph, and queue haplotype-read pairs for DMA transfer to FPGA hardware engines, such as for pairwise HMM or PD HMM comparison. The same or other software chains can then receive the queued and DMA-transferred pairwise HMM results back from the FPGA to CPU/GPU memory and receive genotyping and Bayesian probability calculations to produce final variant calls. Of course, one or more of these functions can be configured to run on one or more quantum computing platforms.

[0562] For example, as can be seen with respect to Figure 38 , the CPU/GPU 1000 can include one or more, e.g., a plurality, of chains 20a, 20b, and 20c, which can each have access to an associated DRAM 1014, wherein the DRAM has workspace 1014a, 1014b, and 1014c, within which each chain 20a, 20b, and 20c can have access, respectively, in order to perform one or more operations on one or more data structures, such as large data structures. These memory portions and their data structures may be accessed, such as via respective cache portions 1014a', such as by one or more processing engines 13a, 13b, 13c of the FPGA 7, wherein the processing engines may access said data structures, such as in performing one or more of the operations described herein, such as for mapping, alignment, sorting, and/or variant calling. Due to the high bandwidth, tightly coupled interconnect 3, data pertaining to the data structures and/or related to processing results may be shared in a substantially seamless manner between the CPU and/or GPU and/or QPU and/or the associated FPGA, such as in a cache-coherent manner, so as to optimize processing efficiency.

[0563] Accordingly, in one aspect, as disclosed herein, a system may be provided wherein the system is configured to share memory resources among its component parts, such as with respect to performing some computational tasks and subfunctions via software, as performed by a CPU and/or GPU and/or QPU, and performing other computational tasks or subfunctions via firmware, such as via the hardware of an associated chip, such as an FPGA and/or ASIC or structured ASIC. This may be achieved in different ways, such as by a loose or tight direct coupling between the CPU/GPU/QPU and the chip, e.g., FPGA. Such configurations may be particularly useful when distributing operations related to the processing of large data structures, as described herein, that have functions or subfunctions to be used and accessed by the CPU and/or GPU and/or QPU and the integrated circuit. Particularly, in various embodiments, when processing data through a genomics segmentation as described herein, such as to speed up the overall processing function, timing, and efficiency, a number of different operations may be performed on the data, where the operations may involve both software and hardware processing components.

[0564] Accordingly, data may need to be shared and/or otherwise communicated between the software component executing on the CPU and/or GPU and/or QPU and the hardware component embedded in the chip, e.g., an FPGA or ASIC. Accordingly, one or more of the steps in the processing thread, or a portion thereof, may be performed by one device, e.g., the CPU/GPU/QPU, and one or more of the multiple steps may be performed by the other device, e.g., the FPGA or ASIC. In such a case, the CPU and FPGA need to be communicatively coupled, such as by a point-to-point interconnect, in such a manner as to allow for the efficient transmission of such data, where the coupling may involve the shared use of memory resources. To achieve such distribution of tasks and sharing of information for the performance of such tasks, the CPU and/or GPU and/or QPU may be loosely or tightly coupled to each other and/or to an FPGA, or other chip set, and a workflow management system may be included so as to distribute the workload effectively.

[0565] Thus, in particular embodiments, a genomics analysis platform is provided. For example, the platform may include a motherboard, a memory, and a plurality of integrated circuits, such as forming one or more of a CPU/GPU/QPU, a mapping module, an alignment module, a classification module, and/or a variant calling module. Specifically, in particular embodiments, the platform may include a first integrated circuit, such as an integrated circuit forming a central processing unit (CPU) and/or graphics processing unit (GPU), that is responsive to one or more software or other algorithms that are configured to instruct the CPU/GPU to perform one or more sets of genomics analysis functions as described herein, such as when the CPU/GPU includes a first set of physical electronic interconnects for connection to the motherboard. In other embodiments, a quantum processing unit is provided, wherein the QPU includes one or more quantum circuits that are configured to perform one or more of the functions disclosed herein. In various cases, memory is provided, wherein the memory may also be attached to the motherboard and may further be electronically connected to the CPU and/or GPU and/or QPU, such as via at least a portion of the first set of physical electronic interconnects. In such cases, the memory may be configured to store a plurality of genomic data reads and/or at least one or more reference gene sequences and/or an index, e.g., such as a hash table, of the one or more reference gene sequences.

[0566] Additionally, the platform may include one or more of a second integrated circuit (or circuits), such as where each second integrated circuit (or circuits) forms a field programmable gate array (FPGA) or ASIC or structured ASIC that has a second set of physical electronic interconnects for connection to the CPU and memory, such as via a point-to-point interconnect protocol. In such a case, the FPGA (or structured ASIC) may be firmware programmable to configure a set of hardwired digital logic circuits that are interconnected by a plurality of physical interconnects to perform a second set of genomic analysis functions, e.g., mapping, alignment, classification, deduplication, variant calling, e.g., an HMM function, etc. In particular, the hardwired digital logic circuits of the FPGA may be arranged as a set of processing engines to perform one or more preconfigured steps in a sequence analysis segment of the genomic analysis platform, such as when the set (or sets) of processing engines includes one or more of a mapping and/or alignment and/or classification and/or deduplication and/or variant calling module, wherein the modules may be formed by separate or equal subsets of processing engines.

[0567] For example, with respect to variant calling, a pairwise HMM or PD HMM calculation is one of the most computationally intensive steps of a haplotype-based variant calling protocol. Therefore, the speed of variant calling can be greatly improved by accelerating this step on one or more FPGA or quantum processing engines, as described herein. However, there may be additional benefit in accelerating other computationally intensive steps on additional FPGA and/or QP engines, to achieve greater acceleration of variant calling, or a portion thereof, or to reduce the CPU/GPU load and the number of CPU/GPU cores required, or both, as seen with respect to Figure 38.

[0568] Additional computationally intensive functions, relative to variant calling, that may be implemented in FPGA and/or quantum processing engines include: callable region detection, wherein reference genome regions covered by adequate depth and/or quality of aligned reads are selected for processing; hot region detection, wherein reference genome loci with nontrivial evidence of possible variants are identified, and sufficient context windows around those loci are selected as hot regions for further processing; De-Bruijn or other assembly graph construction, wherein reads overlapping an active region and/or K-mers of those reads are assembled into a graph; assembly graph preparation, such as low coverage or low quality clipping paths, repairing dangling head and tail paths by joining them into a reference backbone in the graph, transforming K-mer to sequence representation of the graph, merging similar branches, and otherwise simplifying the graph; extracting candidate haplotypes from the assembly graph; as well as aligning candidate haplotypes to the reference genome, such as by Smith-Waterman alignment, for example, to determine variants (SNPs and/or indels) from the reference represented by each haplotype, and synchronize their nucleotide positions with the reference.

[0569] All of these functions can be implemented as high-performance hardware engines within the FPGA and/or by one or more quantum circuits of a quantum computing platform. However, calling such a variety of hardware acceleration functions from many integration points in the variant calling software may become inefficient on a loosely integrated CPU/GPU/QPU+FPGA platform, and therefore a tightly integrated CPU/GPU/QPU+FPGA platform may be desirable. For example, various piecemeal processing methods, such as constructing, preparing, and extracting haplotypes from a De Bruijn graph, or other assembly graph, could benefit greatly from a tightly integrated CPU/GPU/QPU+FPGA platform. Additionally, assembly graphs are large and complex data structures, and repeatedly passing them between the CPU and/or GPU and the FPGA could become resource-intensive and inhibit significant acceleration.

[0570] Thus, an ideal model for such graph processing, employing a tightly integrated CPU/GPU/QPU and/or FPGA platform, is to maintain such graphs in cache-coherent shared memory for change processing by CPU and/or GPU and/or QPU software functions and FPGA hardware. In such a case, a software pipeline that processes a given graph could iteratively command multiple compute-intensive graph processing steps through a hardware engine, and then the software could inspect the results and determine the next steps between hardware calls, as exemplified in the process in Figure 39. This processing model could be controlled by a suitably configured workflow management system, and/or could be configured to match software paradigms such as a data structure API or an object-oriented method interface, but with compute-intensive functions being accelerated by custom hardware and/or quantum processing engines, which is most practical to deploy on a tightly integrated CPU and/or GPU and/or QPU + FPGA platform with cache-coherent shared memory and high-bandwidth/low-latency CPU/GPU/QPU/FPGA interconnects.

[0571] Accordingly, in addition to mapping and aligning sequenced reads to a reference genome, the reads may be assembled "de novo", e.g., without a reference genome, such as by detecting apparent overlap between reads, e.g., in an assemblage, in which they are completely or nearly completely in agreement, and joining them into longer, contiguous sequences, scaffolds or graphs. Such assembly may also be performed locally, such as by using all reads determined to map to a given chromosome or portion thereof. Assembly in this manner may also incorporate a reference genome, or segment thereof, into the assembled structure.

[0572] In such a case, due to the complexity of joining sequences of reads that are not completely in agreement, a graph structure may be employed, such as where overlapping reads may agree on a single sequence in one segment, but branch into multiple sequences in an adjacent segment, as explained above. Such an assembly graph, therefore, may be a sequence graph, where each edge or node represents a nucleotide or sequence of nucleotides that is considered to be joined contiguously to the sequences at connected edges or nodes. In particular cases, such an assembly graph may be a k-mer graph, where each node represents a k-mer, or sequence of nucleotides of (typically) fixed length k, and where connected nodes are considered to overlap with each other in longer observed sequences, typically overlapping by k-1 nucleotides. In various methods, there may be one or more transformations performed between one or more sequence graphs and k-mer graphs.

[0573] Although assembly graphs are employed in haplotype-based variant calling, and some of the graph processing methods employed are similar, there are important differences. De novo assembly graphs are generally much larger and employ longer k-mers. While variant calling assembly graphs are constrained to be reasonably structured and relatively simple, such as having no cycles and flowing from source to target along a reference sequence backbone, de novo assembly graphs tend to be more complex and structured, with cycles, dangling paths, and other anomalies not only allowed, but subject to special analysis. De novo assembly graph coloring is sometimes employed, assigning "colors" to nodes and edges, signifying, for example, the biological sample from which they originate or match a reference sequence. Therefore, a wider variety of graph analysis and processing functions need to be employed for de novo assembly graphs, often in an iterative or recursive fashion, and specifically due to the size and complexity of de novo assembly graphs, processing functions tend to be extremely computationally intensive.

[0574] Thus, as discussed above, an ideal model for such graph processing on a tightly integrated CPU/GPU/QPU+FPGA platform is to maintain such graphs in cache-coherent shared memory to alternate processing between CPU/GPU/QPU software functions and FPGA hardware. In such a case, a software chain processing a given graph can interactively command multiple compute-intensive graph processing steps to be performed by a hardware engine, and then inspect the results to determine the next steps to be performed by the hardware, such as by making hardware calls. As above, this processing model greatly benefits from implementation on a tightly integrated CPU+FPGA platform with cache-coherent shared memory and high-bandwidth/low-latency CPU/FPGA interconnect.

[0575] Additionally, as described below herein, tertiary analysis includes genomic processing that may follow graph assembly and/or variant calling, which in clinical applications may include variant annotation, phenotype prediction, disease testing, and/or treatment response prediction as described herein. The reasons why it is beneficial to perform tertiary analysis on such a tightly integrated CPU/GPU/QPU+FPGA platform are that such a platform configuration enables effective acceleration of primary and/or secondary processing, which is computationally very intensive, and is ideal for continuing with tertiary analysis on the same platform for convenience and reduced turnaround time, and minimizing transmission and copying of large genomic data files. Therefore, a loosely or tightly coupled CPU/GPU/QPU+FPGA platform is a good choice, but a tightly coupled platform may include additional benefits due to the fact that tertiary analysis steps and methods vary widely from one application to another, and in any case where compute-intensive steps slow down tertiary analysis, custom FPGA acceleration of these steps can be implemented in an optimized way.

[0576] For example, a particular benefit to tertiary analysis on a tightly integrated CPU/GPU/QPU and/or FPGA platform is the ability to iteratively reanalyze genomic data, leveraging CPU/GPU/QPU and/or FPGA secondary processing acceleration, in response to partial or intermediate tertiary results, which may further benefit from the tightly integrated configuration. For example, after tertiary analysis detects a possible phenotype or disease, but with limited confidence as to whether the detection is true or false, focused secondary reanalysis can be performed with extremely high effort on the particular reference reads and regions that impact detection, thereby improving the accuracy and confidence of relevant variant calling, and in turn improving confidence in detection calling. Additionally, if the tertiary analysis determines structural variant genotype information from the ancestry of the analyzed individual, the secondary analysis can be repeated using a different or modified reference genome that is more suitable for the specific individual, thereby increasing the accuracy of variant calling and improving the accuracy of other tertiary analysis steps.

[0577] However, if tertiary analysis is performed on a CPU-only platform after primary and secondary processing (possibly accelerated on a separate platform), then reanalysis with secondary processing tools is likely to be too slow to be useful on the tertiary analysis platform itself, and the alternative is offloading to a faster platform, which is also prohibitively slow. Thus, in the absence of any form of hardware or quantum acceleration on the tertiary analysis platform, primary or secondary processing must generally be completed before tertiary analysis can begin, without the possibility of easy reanalysis or iterative secondary analysis and/or segmentation of analytical functions. But on an FPGA and/or quantum-accelerated platform, and especially a tightly integrated CPU and/or GPU and/or QPU and/or FPGA platform where secondary processing is maximally efficient, iterative analysis becomes practical and useful.

[0578] Accordingly, as indicated above, the modules disclosed herein may be implemented in the hardware of the chip, such as by being wired thereto, and in such cases their implementation may be such that their operation may occur at a faster speed, with greater accuracy, compared to when implemented in software, such as when there are minimal instructions to be fetched, read, and/or executed. Additionally, in various cases, the functions to be performed by one or more such modules may be distributed such that several of the functions may be configured to be implemented by the host CPU and/or GPU and/or QPU software, while in other cases, several other functions may be performed by the hardware of an associated FPGA, such as where the two or more devices perform their respective functions with each other, such as in a seamless manner. For such purposes, the CPU, GPU, QPU and/or FPGA or ASIC or structured ASIC may be tightly coupled, such as through a low-latency, high-bandwidth interconnect such as a QPI, CCVI, CAPI, and the like. Consequently, in some cases, the highly compute-intensive functions to be performed by one or more of these modules may be performed by a quantum processor implemented by one or more quantum circuits.

[0579] Thus, given the implementation of hardware and/or single quantum processing, the modules of the disclosure may operate directly according to their operating parameters, such as without needing to fetch, read and/or execute instructions, such as when implemented in CPU software only. Additionally, memory requirements and processing times may be further reduced, such as where communications with the chip are via files, e.g., stored locally in the FPGA/CPU/GPU/QPU cache, such as in a cache-coherent manner, rather than via extensive access to external memory. Of course, in some cases, the chip and/or card may be scaled to include more memory, such as more on-board memory, in order to enhance parallel processing capabilities, thereby resulting in even faster processing speeds. For example, in certain embodiments, a chip of the disclosure may include an embedded DRAM such that the chip does not have to rely on external memory, which would therefore result in a further increase in processing speed, such as when a Burrows-Wheeler or De Brujin Graph algorithm may be employed instead of a hash table and hash function, which may in many cases rely on external, e.g., host, memory. In such cases, the operation of a portion or all of the segmentation may be performed in 6 or 10 or 12 or 15 or 20 minutes or less, such as from start to finish.

[0580] As indicated above, there are a number of different points at which any given module may be positioned on the hardware or be positioned remotely from it, such as on a cloud-accessible server. When a given module is positioned on the chip, e.g., wired to the chip, its function may be performed by the hardware, however, when desired, the module may be positioned remotely from the chip, at which point the platform may include the necessary instrumentation to send the relevant data to a remote location, such as a server, e.g., quantum server, accessible via the cloud, so that the particular module functionality may be engaged for further processing of the data in accordance with desired user-selected protocols. Accordingly, part of the platform may include a web-based interface for performing one or more tasks relating to the operation of one or more of the modules disclosed herein. For example, when mapping, alignment, and/or classification are all modules that can occur on-chip, in many cases, one or more of local realignment, duplicate marking, base quality core recalibration, and/or variant calling can occur in the cloud.

[0581] Particularly, once the genetic data has been generated and/or processed, e.g., in one or more primary and/or secondary processing protocols, such as by being mapped, aligned, and/or classified, such as to produce one or more variant call files, e.g., to determine how a subject's genetic sequence data is different from one or more reference sequences, a further aspect of the disclosure may be directed to performing one or more other analytical functions on the generated and/or processed genetic data, such as for further, e.g., tertiary, processing, as depicted in Figures 40 . For example, the system may be configured for further processing of the secondarily processed and/or generated data, such as by running it through one or more microarray analysis runs 700, such as one or more of a microarray analysis run, a genome, e.g., whole genome, analysis run, a genotyping analysis run, a genome ... exome, epigenome analysis segmentation, metagenome analysis segmentation, microbiome analysis segmentation, genotyping analysis segmentation including joint genotyping, variant analysis segmentation including structural variant segmentations, somatic variant segmentations, and GATK and/or MuTect2 segmentations, as well as R sequencing segmentations and other genetic analysis segmentations.

[0582] Additionally, in various instances, an additional layer of processing 800 may be provided, such as for disease diagnosis, therapeutic treatment, and/or prophylactic prevention, such as including NIPT, NICU, Cancer, LDT, AgBio, and other such disease diagnosis, prophylaxis, and/or treatments that employ data generated by one or more of the present primary and/or secondary and/or tertiary pipelines. For example, particular bioanalytical segmentations include genome segmentations, epigenome segmentations, metagenome segmentations, genotyping segmentations, variants, e.g., GATK/MuTect2 segmentations, and other such segmentations. Thus, the devices and methods disclosed herein can be used to generate genetic sequence data, which data can then be used to generate one or more variant call files and/or other associated data that can be further subjected to other tertiary processing pipelines in accordance with the devices and methods disclosed herein, such as for diagnosis of particular and/or general disease, as well as for prophylactic and/or therapeutic treatment and/or development modalities. See, for example, Figures 41 B, C and 43.

[0583] As described above, the methods and/or systems presented herein may include generating and/or otherwise acquiring genetic sequence data. Such data may be generated or otherwise acquired from any suitable source, such as by an NGS or "sequencer on a chip" technology. Once generated and/or acquired, the methods and systems herein may include subjecting the data to further processing, such as by one or more secondary processing protocols 600. The secondary processing protocols may include one or more of mapping, aligning, and classifying the generated genetic sequence data, such as to produce one or more variant call files, for example, in order to determine how a subject's genetic sequence data differs from one or more reference sequences or genomes. An additional aspect of the disclosure may be directed to performing one or more other analytical functions on the generated and/or processed genetic data, e.g., secondary result data, such as for further processing, e.g., tertiary processing 700/800, wherein the processing may be performed on or in association with the same chip or chip array as that housing the aforementioned sequencer technology.

[0584] Accordingly, in a first case, such as with respect to the generation, acquisition, and/or transmission of genetic sequence data as shown in Figures 37 to 41, such data may be generated locally or remotely and/or the results thereof may then be directly processed, such as by means of a local computing resource 100, or may be transmitted to a remote location, such as a remote computing resource 300, for further processing, e.g., for secondary and/or tertiary processing, see Figures 42, For example, the generated genetic sequence data may be processed locally and directly, such as when the sequencing and secondary processing functionalities are housed on the same chip set and/or within the same device at site 10. Likewise, the generated genetic sequence data may be processed locally and indirectly, such as when the sequencing and secondary processing functionalities occur separately by distinct apparatus that share the same facility or location, but may be separated by a space despite being connected in a manner that is not directly related to the same device. communicable manner, such as via a local area network 10. In a further case, genetic sequence data may be derived remotely, such as by a remote NGS, and the resulting data may be transmitted via a cloud-based network 30/50 to a remote off-site location 300, such as geographically separated from the sequencer.

[0585] Specifically, as illustrated in Figure 40A, in various embodiments, a data generation apparatus, e.g., nucleotide sequencer 110, may be provided on-site, such as when the sequencer is a “sequencer on a chip” or an NGS, wherein the sequencer is associated with a local computing resource 100 directly or indirectly, such as via a local network connection 10/30. The local computing resource 100 may include or be otherwise associated with one or more data generation engines 110 and/or data acquisition engines 120. Such engines may be any engine configured to generate and/or otherwise acquire data, such as analog, digital, and/or electromagnetic data related to one or more genetic sequences of a subject or group of subjects, such as when genetic sequence data is in a BCL or FASTQ file format.

[0586] For example, such a data generation engine 110 may be a primary processor, such as a sequencer, such as an NGS, a sequencer on a chip, or other similar engine for generating genetic sequence information. Additionally, such data acquisition engines 120 may be any engine configured to receive data, such as generated genetic sequence information, and/or in conjunction with the data generator 110 and/or the computing resource 100 capable of subjecting the same to one or more secondary processing protocols, such as a secondary processing segmentation apparatus configured to operate a mapper, an aligner, a classifier, and/or a variant caller protocol on the generated and/or acquired sequence data, as described herein. In various instances, data generation apparatus 110 and/or data acquisition apparatus 120 may be networked together, such as through a local area network 10, such as for local storage 200, or may be networked together through a cloud-based and/or local area network 30, such as to transmit and/or receive data, such as digital data related to primary and/or secondary processing of genetic sequence information, such as to or from a remote location, such as for processing 300 and/or remote storage 400. In various embodiments, one or more of these components may be communicably coupled together by a hybrid network, as described herein.

[0587] The local computing resource 100 may also include or be otherwise associated with a compiler 130 and/or a processor 140, such as a compiler 130 configured to compile the generated and/or acquired data and/or data associated therewith, and a processor 140 configured to process the generated and/or acquired and/or compiled data and/or control the system 1 and its components, as described herein, such as to perform primary, secondary, and/or tertiary processing. For example, any suitable compiler may be employed, however, in certain cases, further efficiencies may be achieved not only by implementing a tightly coupled configuration, as discussed above, for efficient and coherent transfer of data between system components, but may additionally be achieved by implementing a just-in-time (JIT) computer language compiler configuration. Furthermore, in certain instances, processor 140 may include a workflow management system to control the operation of the various system components with respect to data generated, received, and/or to be processed through the various stages of the platform segmentations.

[0588] Specifically, as used herein, just-in-time (JIT) refers to a device, system, and/or method for converting acquired and/or generated file formats from one another. In a broad usage framework, the JIT system disclosed herein may include a compiler 130, or other computing architecture, e.g., a processing program, that may be implemented in a manner to convert various codes from one form to another. For example, in one implementation, a JIT compiler may function to convert byte code or other program code that contains instructions that need to be interpreted, into instructions that can be sent directly to an associated processor 140 for immediate execution, such as without the need for interpretation of the instructions by the particular machine language. In particular, after a coding program, e.g. a Java program, has been written, the source language statements may be compiled by a compiler, e.g. a Java compiler, into byte code, rather than compiled into code containing instructions that correspond to any given processing language of the particular hardware platform. This byte code compilation action is therefore platform-independent code that can be shipped to any platform and executed on that platform independently of its underlying processing. Thus, a suitable compiler may be a compiler that is configured to compile the byte code into platform-specific executable code that can then be executed immediately. In this case, the JIT compiler may function to convert one file format to another as soon as possible, such as "on the fly".

[0589] Therefore, a properly configured compiler as described herein has the ability to overcome several deficiencies in the art. Specifically, previously compiled programs that were written in a specific language had to be recompiled and/or rewritten on each specific computer platform on which they were written. In the present compilation system, the compiler may be configured so that it only has to write and compile a program once, and once written in a particular form, it can be converted to one or more other forms almost immediately. More specifically, the compiler 130 may be a JIT, or may be in another similar dynamic translation compiler format, that has the ability to write instructions in a platform-unknown language that do not have to be recompiled and/or rewritten dependent on the specific computer platform on which it is deployed. For example, in a particular usage model, the compiler may be configured to interpret compiled byte code and/or other encoded instructions into instructions that are understandable by a given particular processor for converting one file format to another, regardless of the computing platform. In particular, the JIT system herein has the capability of receiving a genetic file, such as representing genetic code, for example, where the file is a BCL or FASTQ file, for example, generated from a genetic sequencer, and rapidly converting the same to another form, such as a SAM, BAM, and/or CRAM file, such as using the methods disclosed herein.

[0590] Particularly, in various instances, the system disclosed herein may include a first and/or a second compiler 130a and 130b, such as a virtual compilation machine, that handles one or a plurality of byte code instruction conversions at a time. For example, use of a Java-type just-in-time compiler, or other suitably configured second compiler, within the present system platform will allow for the compilation of instructions into byte code that can then be converted into the particular system code, e.g., as if the program had been initially compiled on that platform. Accordingly, once the code has been compiled and/or (recompiled, such as by the JIT compiler(s) 130), it will run quickly on the computer processor 140. Therefore, in various embodiments, just-in-time (JIT) compilation, or other dynamic translation compilation, may be configured to be performed during the execution of a given program, e.g., at runtime, rather than prior to execution. In such a case, this may include the step (or steps) of translation to machine code or translation to another format, which can then be performed directly, thus allowing one or more of interpretation and/or ahead-of-time (AOT) compilation.

[0591] More particularly, as implemented within the present system, a typical genome sequencing data stream produces data in one or more file formats, derived from one or more computing platforms, such as in a BCL, FASTQ, SAM, BAM, CRAM, and/or VCF file format or their equivalents. For example, a typical DNA sequencer 110, e.g., an NGS, produces raw signals representing called bases that are designated herein as reads, such as in a BCL and/or FASTQ file, which may optionally be further processed, e.g., enhanced image processing, and/or compressed 150. Likewise, the reads from the generated BCL/FASTQ files may then be further processed within the system, as described herein, to produce mapping and/or alignment data, such data produced, e.g., from the mapped and aligned reads, may be in a SAM or BAM file format or, alternatively, a CRAM file format. Furthermore, the SAM or BAM file may then be processed, such as through a variant calling procedure, to produce a variant calling file, such as a VCF file or a gVCF file. Consequently, all of these produced BCL, FASTQ, SAM, BAM, CRAM, and/or VCF files, once produced, are (extremely) large files that all need to be stored, such as in the local system memory architecture 200 or remotely 400. Storing any of these files is expensive. Storing all of these file formats is extremely expensive.

[0592] As indicated, just-in-time (JIT) or other dynamic translation compilation or dual compilation analysis may be configured and implemented herein so as to reduce such high storage costs. For example, a JIT analysis scheme may be implemented herein so as to store data in only one format (e.g., a compressed FASTQ or BAM file format, etc.), while providing access to one or more file formats (e.g., BCL, FASTQ, SAM, BAM, CRAM, and/or VCF, etc.). Such a fast file conversion process can be accomplished by rapidly processing the genomic data using the respective hardware and/or quantum acceleration platforms disclosed herein, e.g., such as for mapping, alignment, classification, and/or variant calling (or component functions thereof, such as deduplication, HMM and Smith-Waterman, compression and decompression, and the like), in hardware engines on an integrated circuit, such as an FPGA, or by a quantum processor. Therefore, by implementing JIT or other similar analysis together with such acceleration, the genomic data can be processed to generate desired file formats on the fly, at speeds comparable to normal file access. Thus, considerable storage savings can be realized by JIT-type processing with little or no loss of access speed.

[0593] In particular, two general options are useful for the underlying storage of the genomic data produced herein so as to be accessible for JIT-type processing, these include the storage of unaligned reads (e.g., which may include compressed FASTQ, or unaligned compressed SAM, BAM, or CRAM files), and the storage of aligned reads (e.g., which may include compressed BAM or CRAM files). However, since the accelerated processing disclosed herein allows any of the file formats referred to to be rapidly derived, e.g., on the fly, the underlying file format for storage may be selected so as to achieve the smallest compressed file size, thereby decreasing the storage overhead. Therefore, because of the comparatively smaller file size for unprocessed, e.g., unaligned raw read data, there is an advantage to storing unaligned reads so that data fields are minimized. Likewise, there is an advantage of storing the processed and compressed data, such as in a CRAM file format.

[0594] More particularly, in view of the rapid processing speeds achievable by the devices, systems, and methods of their use disclosed herein, in many cases there may be no need to store mapping and/or alignment information for each and every read, because such information can be rapidly derived upon need, such as immediately. Furthermore, although a compressed FASTQ file format (e.g., FASTQ.gz) is commonly used for storage of genetic sequence data, such unaligned read data may be stored in more advanced compressed formats as well, such as after mapping and/or alignment in SAM, BAM, or CRAM files, which may further reduce file size, such as by the use of compact binary representation and/or more targeted compression methods. Therefore, such file formats may be compressed prior to storage, decompressed after storage, and processed rapidly, such as immediately, to convert one file format from another.

[0595] One advantage of storing aligned reads is that much or all of the sequence content of each read can be omitted. Specifically, system efficiency can be increased and storage space saved by only storing the differences between the read sequences and the selected reference genome, such as at indicated variant alignment positions of the read. More specifically, since differences from the reference are usually sparse, the aligned position and list of differences can often be more compactly stored than the original read sequence. Therefore, in many cases, storing an aligned read format, e.g., when storing data related to the differences of aligned reads, may be preferable to storing unaligned read data. In such a case, if an aligned read and/or variant call format is used as the underlying storage format, such as in a JIT procedure, other formats, such as a SAM, BAM, and/or CRAM compressed file formats, may also be used.

[0596] Along with the aligned and/or unaligned read file data to be stored, a wide variety of other data, such as metadata derived from the various computations set forth herein, may also be stored. Such computed data may include mapped, alignment, and/or subsequent read processing data, such as alignment scores, mapping confidence, edit distance from reference, etc. In certain cases, such metadata and/or other extra information need not be retained in the underlying storage for JIT analysis, such as in those cases where it needs to be reproduced immediately, such as by the accelerated data processing described herein.

[0597] Regarding metadata, this data may be a small file that instructs the system how to go back and forth from one file format in conversion to another file format. Thus, the metadata file allows the system to create a bit-compatible version of any other file type. For example, moving forward from a source data file, the system only needs to access and implement the metadata instructions. Along with fast file format conversion, JIT also enables fast compression and/or decompression and/or storage, such as in a genomics dropbox memory cache.

[0598] As discussed in greater detail below, once sequence data is generated 110, it may be stored locally 200 and/or may be made accessible for storage remotely, such as in a cloud-accessible dropbox-type memory cache 400. For example, once in the genomic dropbox, the data may appear as accessible in the cloud 50 and may then be further processed, e.g., substantially immediately. This is particularly useful when there are a plurality of mapping/alignment/classification/variant calling systems 100/300, such as with one on each side of the cloud interface 50 facilitating automatic upload and processing of the data, which may be further processed, such as using the JIT technology described herein.

[0599] For example, an underlying storage format for JIT processing and/or compilation may contain only minimal data fields, such as read name, base quality scores, alignment position and/or orientation in the reference, and a list of differences from the reference, such as where each field can be compressed in a manner optimal for its data type. Various other metadata may be included and/or otherwise associated with the storage file. In such a case, the underlying storage for JIT analysis may be on a local file system 200, such as on hard disk drives and solid state drives, or a network storage resource, such as a NAS or object or Dropbox-type storage system 400. Particularly, when multiple file formats, such as BCL, FASTQ, SAM, BAM, CRAM, VCF, etc., have been produced for a genomic data set, which may be subjected to JIT processing and/or storage, the JIT or other similar compilation and/or analysis system may be configured to convert the data into a single underlying storage format for storage. Additional data, such as metadata and/or other information (which may be small) necessary to reproduce all other formats desired by accelerated genomic data processing may also be associated with the field and stored. Such additional information may include one or more of: a list of file formats to be reproduced, data processing commands to reproduce each format, unique ID (e.g., URL or MD5/SHA hash) of reference genome, various parameter settings, such as for mapping, alignment, classification, variant calling, and/or any other processing as described herein, randomization seeds for processing steps, e.g., using pseudorandomization, to deterministically reproduce the same results, User interface, and the like.

[0600] In various cases, the data to be stored and/or retrieved in a JIT or similar dynamic translation processing and/or analysis system may be presented to the user or other applications in a variety of ways. For example, one option is to have the JIT analysis store a standard or custom "JIT object" file format, such as for storage and/or retrieval as a SAM, BAM, CRAM, or other custom file format, and provide user tools to quickly convert the JIT object to the desired format (e.g., in a local temporary store 200) using the accelerated processing disclosed herein. Another option is to present the appearance of multiple file formats, such as BCL, FASTQ, SAM, BAM, CRAM, VCF, etc., to the user and user applications, so that file system access to multiple file formats utilizes one JIT procedure, so that only one file type needs to be saved, and from that file type, all other files can be generated immediately. An additional option is to produce user tools that otherwise accept specific file formats (BCL, FASTQ, SAM, BAM, CRAM, VCF, etc.) that have the ability to be presented as a JIT object instead and can automatically call JIT analysis to obtain the data in the desired data format, e.g. BCL, FASTQ, SAM, BAM, CRAM, VCF, etc. automatically when called.

[0601] Accordingly, JIT procedures are useful for providing access to multiple file formats, e.g., BCL, FASTQ, SAM, BAM, CRAM, VCF, and the like, from a single file format by quickly processing the underlying stored compressed file format. Additionally, JIT remains useful even if only a single file format is to be accessed, due to the fact that compression is still achieved relative to direct storage of the accessed format. In such a case, the underlying file storage format may be different from the accessed file format and/or may contain less metadata and/or may be compressed more efficiently than the accessed format. Furthermore, in such a case, as discussed above, the file is compressed prior to storage and decompressed upon retrieval, e.g., automatically.

[0602] In various instances, the JIT analysis methods as provided herein may also be used for transmission of genomic data, over the Internet or other network, to minimize transmission time and decrease consumed network bandwidth. In particular, in a storage application, a single compressed underlying file format may be stored and/or one or more formats may be accessed via decompression and/or accelerated genomic data processing. Similarly, in the transmission application, only a single compressed underlying file format need be transmitted, e.g., from a source network node to a destination network node, such as where the underlying format may be chosen primarily for the smallest compressed file size, and/or where all desired file formats may be generated at the destination node by or for processing genomic data, such as immediately. In this way, only one compressed data file format needs to be used for storage and/or transfer, from such a file format the other various file formats can be derived.

[0603] Accordingly, in view of Figure 40A, hardware and/or quantum accelerated genomic data processing as described herein may be utilized at (or by) the source network node to generate and/or compress the underlying format for transmission, and the destination network node to decompress and/or generate other desired file formats by accelerated genomic data processing. However, JIT or other dynamic translation analysis remains useful in the transmission application even if only one of the source node and the destination node utilizes hardware and/or quantum accelerated genomic data processing. For example, a data server sending large amounts of genomic data may utilize hardware and/or quantum accelerated genomic data processing to generate the compressed underlying format for transmission to multiple destinations. In such cases, each destination may use slower software genomic data processing to generate other desired data formats. Therefore, although the speed advantage of JIT analysis is diminished at the destination node, transmission time and network utilization are still usefully reduced, and the source node has the ability to provide many transmissions efficiently due to its corresponding hardware and/or quantum accelerated genomic data processing apparatus.

[0604] Furthermore, in another example, a data server receiving uploaded transmissions of large amounts of genomic data, e.g., from multiple sources, may utilize hardware and/or quantum accelerated genomic data storage and/or processing, while the multiple source nodes may utilize slower software running on a CPU/GPU to generate the compressed underlying file format for transmission. Alternatively, hardware and/or quantum accelerated genomic data processing may be utilized by one or more intermediate network nodes, such as a gateway server, between the source and destination nodes to transmit and/or receive genomic data in a compressed underlying file format, in accordance with JIT or another dynamic translation analysis method, thereby gaining the benefits of reduced transmission time and network utilization without burdening said intermediate network nodes with excessive software processing.

[0605] Thus, as can be seen with respect to Figure 40A, in certain instances, the local computing resource 100 may include a compiler 130, such as a JIT compiler, and may further include a compressor unit 150 that is configured to compress data, such as primary generated and/or acquired and/or processed secondary data (or tertiary data), such data may be compressed, such as prior to transfer over a local network 10 and/or cloud 30 and/or hybrid cloud-based network 50, such as in a JIT analysis procedure, and may be decompressed subsequent to transfer and/or prior to use.

[0606] As described above, in various instances, the system may include a first integrated and/or quantum circuit 100 such as for performing a mapping, alignment, classification, and/or variant calling operation to generate one or more of the mapped, aligned, classified, deduplicated, and/or variant called data. Additionally, the system may include an additional integrated and/or quantum circuit 300 such as for employing the result data in performing one or more genomic and/or bioinformatics segmentation analyses, such as for tertiary processing. For example, the result data generated by the first integrated and/or quantum circuit 100 may be used, for example, by the first or a second integrated and/or quantum circuit 300 in performing an additional genomic and/or bioinformatics segmentation processing procedure. Specifically, secondary processing of genomic data may be performed by a first hardware accelerated and/or quantum processor 100 to produce results data, and tertiary processing may be performed on such results data, such as wherein the additional processing is performed by a CPU and/or GPU and/or QPU 300 that is operatively coupled to the first integrated circuit. In such a case, the second circuit 300 may be configured to perform tertiary processing of the genomic variation data produced by the first circuit 100. Accordingly, the results data derived from the first integrated server acts as an analysis engine driving the additional processing steps described herein with reference to tertiary processing, such as by the second integrated and/or quantum processing circuit 300.

[0607] However, the data generated in each of these primary and/or secondary and/or tertiary processing steps may be immense, requiring very high resource and/or memory costs, such as for storage, either locally 200 or remotely 400. For example, in a first primary processing step, generated nucleic acid sequence data 110, such as in a BCL and/or FASTQ file format, may be received 120, such as from an NGS 110. Regardless of the file format of this sequence data, the data may be employed in a secondary processing protocol as described herein. The ability to receive and process primary sequence data directly from an NGS, such as in a BCL and/or FASTQ file format, is quite useful. In particular, instead of converting the NGS sequence data file, e.g., BCL, to a FASTQ file, the file may be directly received from NGS, e.g., as a BCL file, and may be processed, such as by being received and converted by the JIT system, e.g., immediately, to a FASTQ file which may then be processed, as described herein, such as to produce mapped, aligned, classified, deduplicated, and/or variant-called result data which may then be compressed, such as in a SAM, BAM, and/or CRAM file, and/or may be subjected to further processing, such as by one or more of the disclosed genomic tertiary processing segments.

[0608] Consequently, such data once produced needs to be stored in some manner. However, such storage is not only resource intensive, it is also expensive. Specifically, in a typical genomics protocol, the sequenced data, once generated, is stored as a large FASTQ file. Then, once processed, such as by being subjected to a mapping and/or alignment protocol, a BAM file is created, such a file is also typically stored, increasing the expense of genomic data storage, such as having to store both a FASTQ and a BAM file. Furthermore, once the BAM file is processed, such as by being subjected to the variant calling protocol, a VCF file is produced, such a VCF also typically needs to be stored. In such a case, in order to adequately provide and make use of the generated genetic data, all three of the FASTQ, BAM and VCF files may need to be stored, either locally 200 or remotely 400. Additionally, the original BCL file may also be stored. Such storage is inefficient as well as being expensive and memory resource intensive.

[0609] However, the computational power of the hardware and/or quantum processing architectures deployed herein, coupled with JIT compilation, compression, and storage, significantly mitigates these inefficiencies, resource costs, and expenses. For example, in view of the methods deployed and the processing speeds achieved by the present accelerated integrated circuits, such as for converting a BCL file to a FASTQ file, and then converting a FASTQ file to a SAM or BAM file, and then converting a BAM file to a CRAM and/or VCF file, and back again, the present system significantly reduces the number of computing resources and/or file sizes required for efficient processing and/or storage of such data. The benefits of these systems and methods are further enhanced by the fact that only one file format, e.g., a BCL, FASTQ, SAM, BAM, CRAM, and/or VCF, needs to be stored, from which all other file formats can be derived and processed. In particular, only one file format needs to be saved, and from such a file, any of the other file formats can be generated quickly, e.g., immediately, in accordance with the methods disclosed herein, such as in a just-in-time, or JIT, compilation format.

[0610] For example, in accordance with the typical prior methods, a large amount of computing resources, e.g., server farms and large memory banks, are required for processing and storing FASTQ files being generated by an NGS sequencer. In particular, in a typical case, once the NGS produces the large FASTQ file, the server farm would then be employed to receive and convert the FASTQ file into a BAM and/or CRAM file, such processing may take up to a day or more. However, once produced, the BAM file itself must then be stored, requiring additional time and resources. Likewise, the BAM or CRAM file may be processed to generate a VCF, which may also take up to another day or more, and such a file will also need to be stored, thereby incurring additional costs and expenses. More particularly, in a typical case, the FASTQ file for a human genome consumes about 90 GB of storage per file. Likewise, a typical human genome BAM file may consume about 160 GB. VCF files may also need to be stored, although such files are much smaller than FASTQ and/or BAM files. SAM and CRAM files may also be generated through secondary processing procedures, and these may also need to be stored.

[0611] Prior to the technologies provided herein, it was computationally intensive to go from one step to another, e.g., from one file format to another, and therefore all of the data for those file formats would typically have to be stored. This is in part due to the fact that if a user wanted to go back and regenerate one or more of the files, it would require a large amount of computing resources and time to redo the processes involved in regenerating the various files, thus incurring a high monetary expense. Furthermore, when such files are compressed prior to storage, such compression can take from about 2 to about 5 to about 10 or more hours, with about the same amount of time required for decompression prior to reuse. Because of this high expense, typical users would not compress such files prior to storage and would also typically store all two, three or more file formats, e.g., BCL, FASTQ, BAM, VCF, incurring increased costs through the increased time.

[0612] Accordingly, the JIT protocols employed herein make use of the accelerated processing speeds achieved by the present hardware and/or quantum accelerators in order to achieve increased efficiency, reduced time and costs for both processing and storage. Instead of storing 2, 3 or more copies of the same general data in different file formats, only one file format needs to be stored, and immediately, any of the other file types can be generated, such as using the accelerated processing platforms discussed herein. In particular, from the storage of a FASTQ file, the present devices and systems make it easy to go back to a BCL file, or forward to a BAM file and then further to a VCF, such as in less than 30 minutes, such as within 20 minutes, or about 15 or 10 minutes or less.

[0613] Therefore, using the segmentations and processing speed offered by the wired/quantum processing engines disclosed herein, only a single file format need be stored, while other file formats can be easily and quickly generated from it. Then, instead of needing to store all three file formats, a single file format need be stored from which any other file format can be regenerated, such as immediately, at the exact time for additional processing steps desired by the user. Accordingly, the system can be configured for ease of use so that if a user simply interacts with a graphical user interface, such as presented on an associated display of the device, e.g., the user clicks on the FASTQ, BAM, VCF, etc. button presented on the GUI, the desired file format can be presented, while in the background, one or more of the system's processing engines can be performing the accelerated processing steps necessary to regenerate the requested file in the requested file format from the stored file.

[0614] Typically, one or more of a compressed version of a BCL, FASTQ, SAM, BAM, CRAM and/or VCF file will be saved, along with a small metafile that includes all the settings of how the system was run to create the compressed and/or stored file. Such metafile data details how the particular file format, e.g., FASTQ and/or BAM file, was generated and/or what steps would be required to go backwards or forwards in order to generate any of the other file formats. This process is described in more detail in this document below. In this manner, the process can proceed forwards or be reversed by going backwards using the settings stored in the metafile. This can be a reduction of about 80% or more in economic and storage cost if the computing function is bundled with the storage functions.

[0615] Accordingly, in view of the foregoing and as can be seen with respect to Figure 40A, a cloud-based server system for data analysis and storage is provided. For example, using a cloud-accessible server system as disclosed herein, a user may connect to a storage device, such as for storing input data. For example, a remote user may access the system in order to input genomic and/or bioinformatics data into the system, such as for storing and/or processing the same. In particular, a remote user of the system, e.g. using local computing resource 100, may access the system 1 in order to upload genomic data, e.g., such as one or more sequenced genomes of one or more individuals. As described in detail below, the system may include a user interface, e.g., accessing a suitably configured API, that will allow a user to access the BioIT platform in order to upload data to be processed, control processing parameters, and/or download output, e.g., results data, from the platform.

[0616] Specifically, the system may include an API, e.g., an S3 or "S3-like" object that allows access to one or more system memories for storing 400 and/or receiving stored files. For example, a cloud-accessible API object may be present, such as wherein the API is configurable to store data files in the cloud 50, such as in one or more storage buckets 500, e.g., an S3 bucket. Accordingly, the system may be configured to allow a user to access remotely stored files, e.g., via an S3 or S3-like API, such as by accessing the API via a cloud-based interface on a personal computing device.

[0617] Such an API may therefore be configured to allow access to the cloud 50 to thereby connect the user to one or more of the cloud-based servers 300 disclosed herein, such as uploading and/or downloading a particular stored file, for example, so as to make the files accessible between the cloud server 300 and the local hard drive 100. This may be useful, for example, to allow a remote user to provide, access data and/or download data on or from the server 300 and further perform one or more applications and/or computations on that data, either locally 100 or on the server 300, and then call the API to send the transformed data back to or from the cloud 50, for example, for storage 200 and/or further processing. This is particularly useful for the retrieval, analysis and storage of genomics data.

[0618] However, typical cloud-based data storage, e.g., "S3" storage, is expensive. This expense is compounded when storing large amounts of data associated with the fields of genomics and bioinformatics, where such costs become prohibitive. Additionally, the time required to record, upload, and/or download the data for use, e.g., either locally 100 or remotely 300, and/or for storage 400 also makes such cloud-based storage solutions less attractive. The present solutions disclosed herein overcome these and other such needs.

[0619] In particular, rather than going through a typical "S3" or other cloud-based object API, an alternative S3-compatible API is disclosed herein that can be implemented to reduce the transmission speed and/or cost of storing the data. In such an instance, when a user wants to store a file, rather than going through a typical cloud-based API, e.g., S3, the alternative service API system, e.g., the proprietary S3-compatible API disclosed herein, will launch a compute instance, e.g., a CPU and/or FPGA instance of the system, that will work to compress the file, generate a metadata index regarding what the data is and/or how the file was generated, etc. and will store the compressed file via an S3-compatible storage-type bucket 400. Accordingly, disclosed herein is a cloud-based service 50 that employs a compute instance 300, which may be initiated via an alternative API, in order to compress data prior to storage 400 and/or decompress data upon retrieval. In such an instance, what is therefore stored is not the actual file, but rather what is stored is a compressed version of the original file.

[0620] Specifically, in such an instance, the initial file may be in a first format, which may be uploaded to the system via the proprietary S3-compatible API, which receives the file, e.g., an F1 file, and may then perform a compute function on the file and/or compress the file, such as via a suitably configured CPU/GPU/QPU/FPGA processing engine 300, which then prepares the compressed file for storage as a compressed file, e.g., a compressed F1 file. However, when the compressed and stored file needs to be retrieved, it may then be decompressed, and such decompressed file may then be returned to the user. The advantage of such an accelerated compression and decompression system is that storage 400 of the compressed file results in significant savings in storage costs, such advantage being made possible by the compute and/or compression functionalities achieved by the systems disclosed herein.

[0621] Therefore, due to the fast and efficient computation and/or compression functionalities achieved by the systems, the user need not even know that the file is being compressed prior to storage and subsequently decompressed after storage and presented in the user interface. In particular, the system operates so quickly and efficiently that the user need not be aware of the multitude of compression, computation and/or decompression steps that occur when storing and/or retrieving the requested data; to the user, this all appears seamless and fast. However, the fact that the present storage system will cost less and be more efficient than previous storage systems will be evident.

[0622] Accordingly, in view of the above, object-based storage services are provided herein, wherein storage services can be offered at lower costs by combining a compute and/or compression instance together with storage functionality. In such an instance, typical storage costs can be replaced by compute costs, which are offered at a much lower level, due to the fact that, as disclosed herein, compute costs can be implemented in an accelerated manner, such as by a quantum computing platform and/or FPGA 300, as described herein. Therefore, the accelerated platforms disclosed herein can be configured as a fast and efficient storage and retrieval system that enables compressed storage of data that can be both compressed and stored and rapidly decompressed and retrieved at much lower costs and with greater efficiency and speed. This is particularly useful with respect to the storage of genomic data 400 and is compatible with the Just-In-Time processing functionalities disclosed herein, above. Therefore, in accordance with the devices, systems, and methods disclosed herein, a storage service is an objective that may be provided, wherein the storage service implements fast compression functionality, such as genomics-specific compression, in order to store genomic processing result data.

[0623] More particularly, as can be seen with respect to Figure 40A, in an exemplary implementation, the BioIT systems provided herein may be configured such that a segmentation server system 300, e.g., a portion thereof, receives the request on the API, e.g., S3-compatible API, which is operatively connected to a database 400 that is adapted to associate the initial file (F1) with the compressed version of the file (CF1), e.g., based on coupled metadata. Similarly, once the original CF1 files are decompressed and processed, the resulting results data files (F2) may then be compressed and stored as a CF2 file. Accordingly, when retrieval of the file is desired from the database 400, the server 300 has an API that has already associated the original file with the compressed file via appropriately configured metadata, so when retrieval is requested, a workflow management controller (WMS) of the system will start the compute instance 300, which will start the appropriate compute instance so as to perform any necessary computations and/or decompress the file for further processing, transmission, and/or presentation to the requesting user 100.

[0624] Thus, in various embodiments, an exemplary method may include one or more steps, in any logical order: 1) The request comes in via the API, e.g., S3-compatible API, 2) the API communicates with the WMS, 3) the WMS populates the database and starts the compute instance (or instances), 4) the compute instance (or instances) performs the required compression on the file F1 and generates the characteristic metadata and/or other relevant file associations (X), e.g., to produce a CF1 file X1, 4) thereby preparing the data for storage 400. This process may then be repeated for files F2, F3, Fn, e.g., other processed information, so that the WMS knows how the compressed file was generated, as well as where and how it was stored. It should be noted that a unique feature of this system is that multiple different users 100 may have access to the stored data 400 substantially simultaneously. For example, the compression systems and methods disclosed herein are useful in conjunction with the BioT platforms disclosed herein, through which, at any time during processing, result data can be compressed and stored in accordance with the methods herein, and accessible to others, with the correct permissions.

[0625] In connection with performing genomic analysis, a user 100 may access system 300 herein, for example, via a genomic analysis API, such as an S3 or S3-compatible API, upload genomic data, such as in a BCL and/or FASTQ file or other file format, and thereby request the performance of one or more genomic operations, such as mapping, alignment, classification, deduplication, variant calling, and/or other operations. System 300 receives the request on a workflow manager API, the workflow manager system evaluates the incoming requests, indexes the jobs, forms a queue, allocates resources, e.g., instance allocation, and generates the segmentation workflow. Accordingly, when a request comes in and is pre-processed and queued, an instance allocator, e.g., API, will then process the various job-specific instances, described in more detail in this document below, in accordance with the job designs. Therefore, once the jobs are indexed, queued, and/or stored in an appropriate database 400, the workflow manager will then pull the data from storage 400, e.g., S3 or S3-compatible storage, spin up an appropriate instance, which retrieves the file, and performs the appropriate processes on the data to perform one or more of the requested jobs.

[0626] Additionally, when a plurality of jobs are requested to be performed on the data, requiring the performance of a plurality of instances, then once the first instance has performed its operations, the result data may be compressed and stored, such as in an appropriate memory instance, e.g., a first database, such as an elastic or flexible storage device, so as to wait while the additional segmentation instance (or instances) are processed and retrieve the result data for further processing, such as in accordance with the systems and methods disclosed herein above. Furthermore, as new requests come in and/or current jobs begin to execute, the workflow management system will be constantly updating the queue so as to allocate jobs to the appropriate instances, via an instance allocator API, so as to keep data flowing through the system and the system processes running efficiently.

[0627] Similarly, the system 300 may be constantly taking the results data and storing the data 200/400, for example, in a first or second database, prior to further processing and/or transmission, such as transmission back to the original requester 100 or a designated party. In certain cases, the results data may be compressed, as disclosed herein, prior to storage 400 and/or transmission. Furthermore, as noted above, the results data files generated when compressed may include appropriate metadata and/or other associated data, wherein the results data may be differently named as it flows through the system, such as going from an F1 file to an F1C file to an F2 file to an F2C file, and so on, as the data is processed and moves through the platform segmentation, for example, as directed by a file associations API.

[0628] Accordingly, because of the proprietary dedicated APIs as disclosed herein, the system may have a common backbone to which other services may be coupled and/or additional resources, e.g., instances, may be brought online so as to ensure that all segmentation operations run smoothly and efficiently. Likewise, when desired, the compressed and stored result data files may be called, whereby the workflow manager will process the appropriate compute and/or decompress database instance to decompress result data for presentation to the requester. It is noted that in various cases, the specified compute and compression instance, as well as the specified compute and decompress instance, may be a single instance or multiple instances and may be deployed as a CPU, FPGA, or tightly coupled CPU/FPGA, tightly coupled CPU/CPU, or tightly coupled FPGA/FPGA. In certain cases, one or more of these or other instances disclosed herein may be deployed as a quantum processing unit.

[0629] Accordingly, in view of the disclosures herein, in one aspect, a device for performing one or more of a plurality of functions in performing genomic sequence analysis operations is provided. For example, once data has been received, e.g., by a remote user 100, and/or stored 400 within the cloud-based system, the input data may be accessed by the WMS and may be prepared for further processing, e.g., for secondary analysis, the results thereof may then be transmitted back to the local user 100, e.g., after being compressed, stored 400, and/or subjected to further processing, e.g., tertiary processing, by the server system 300.

[0630] In certain cases, the secondary processing steps disclosed herein, in particular implementations, may be performed by a local computing resource 100 and may be implemented by software and/or hardware, such as being performed by a suitable computing resource 200, wherein the computing resource 200 includes a CPU core, such as from about 4 to about 14 to about 24 or more CPU cores, and may further include one or more FPGAs. The suitable local computing resource 100 may be configured to access a large storage block 200, such as 120 GBs of RAM, which access may be direct, such as being directly coupled thereto, or indirectly, such as being communicably coupled thereto via a local cloud-based network 30.

[0631] Specifically, within a local system, data may be transmitted to and from memory 200 via suitably configured SSD drives that are adapted to write processing job data to, e.g., genomics jobs to be processed and read processed result data from memory 200. In various embodiments, local computing resource 100 may be communicably coupled to a sequencer 110 from which a BCL and/or FASTQ file may be obtained, e.g., from the sequencer and written to the SSD drives, directly, such as via a suitably configured interconnect. Local computing resource 100 may then perform one or more secondary processing operations on the data. For example, in one embodiment, local computing resource is a LINUX® server that has 24 CPUs, such CPUs may be coupled to a suitably configurable FPGA that is adapted to perform one or more of the secondary processing operations disclosed herein.

[0632] Therefore, in particular cases, the local computing device 100 may be a "workbench" computing solution having a BioIT chipset that is configured to perform one or more of the secondary and/or tertiary processing on the genetic data. For example, as disclosed herein, the computing resource 100 may be associated with a PCIe card that is inserted into the computing device so as to be thereby associated with one or more internal CPU cores, GPUs, QPUs, and/or associated memories. Particularly, the components of the computing device 100 including the processing units, associated memories, and/or associated PCIe card(s), having one or more FPGA/ASIC chipsets therein, may be in communication with each other, all of which may be provided within a housing, such as in a compiled manner that is typical within the art. More particularly, the build may be configured for workbench use, or in various cases, it may be configured and provided and/or usable within a remotely accessible server rack. In other embodiments, the CPU/FPGA/Memory chip sets and/or associated interconnected express card(s) may be associated with a Next Generation sequencing device to form a unit therewith.

[0633] Accordingly, in a particular case, a desktop build may include a plurality of CPUs/GPUs/QPUs coupled to one or more FPGAs, such as 4 CPUs/GPUs, or 8, or 12, 16, 20, 22, or 24 CPUs, or more, which may be coupled to 1, or 2, or 3, or more FPGAs, such as within a single housing. Specifically, in a particular case, a build compute resource is provided wherein the compute resource includes 24 CPU cores, a reconfigurable FPGA, a database, e.g., 128x8 RAM, one or more SSDs, such as wherein the FPGA is adapted to be at least partially reconfigurable between operations, such as between performing mapping and alignment. Therefore, in such a case, the BCL and/or FASTQ files generated by the sequencing apparatus 110 may be read into the CPU and/or transferred to the FPGA for processing, and the result data thereof may be read back to the associated CPU via the SSD drives. Accordingly, in this embodiment, the local computing system 100 may be configured to offload various high-computing functionalities to an associated FPGA, thereby increasing the speed, accuracy, and efficiency of the bioinformatics processing. However, while a desktop compilation solution 100 is useful, for example, in a local installation, it may not be suitable for access by a plurality of users who may be located remotely from the compilation.

[0634] Particularly, in various instances, a cloud-based server solution 50 may be provided, such as when the server 300 may be remotely accessible. Accordingly, in particular instances, one or more of the integrated circuits (CPU, FPGA, QPU) disclosed herein may be provided and configured to be accessed via a cloud-based interface 50. Therefore, in particular instances, a workbench build computing resource as described above may be provided wherein the build configuration is adapted to be portable to the cloud and remotely accessible. However, such a configuration may not be sufficient to handle a large amount of traffic from remote users. Accordingly, in other instances, one or more of the integrated circuits disclosed herein may be configurable as a server-based solution 300 configurable as part of a server rack, such as when the server-accessible system is configured to be remotely accessed, such as via the cloud 50.

[0635] For example, in one embodiment, a local computing resource or server 100 having one or more, e.g., a plurality, of CPU cores and/or GPU cores and/or QPU and associated memories may be provided in conjunction with one or more of the FPGAs/ASICs disclosed herein. Particularly, as noted above, in one deployment, a desktop build may be provided, wherein the build includes a build of 18 to 20 to 24 or more CPU/GPU cores having SSDs, 128 x 8 RAM, and one or more BioIT FPGA/ASIC circuitry, and further includes a suitably configured communications module having transmitters, receivers, antennas, as well as WIFI, Bluetooth, and/or cellular communications capabilities that are adapted to allow the build to be remotely accessible. In this implementation, such as when a single FPGA is provided, the FPGA (or FPGAs) can be adapted to be reconfigured, such as partially reconfigured, between one or more of the various stages of genomic analysis segmentation.

[0636] However, in other instances, a server system is provided and may include up to about 20-24-30-34-36 or more CPU/GPU cores and about 972 GB of RAM, or more, which may be associated with one or more, such as about two or four or about six or about eight or more FPGAs, such FPGAs being configurable as described herein. For example, in one implementation, the one or more FPGAs may be adapted to be reconfigured, such as partially reconfigured, between one or more of the various steps of the genomic analysis segmentation. However, in various other implementations, a set of dedicated FPGAs may be provided, such as where each FPGA is dedicated to performing a specific BioIT operation, such as mapping, alignment, variant calling, etc., thereby making the reconfiguration step obvious.

[0637] Accordingly, in various cases, one or more FPGAs may be provided, such as where the FPGA(s) is adapted to be reconfigurable between various slicing operations. However, in other cases, one or more of the FPGAs may be configured to be dedicated to performing one or more functions without the need to be partially or fully configured. For example, the FPGAs provided herein may be configured to be dedicated to performing one or more computationally intensive operations in BioIT slicing, such as where one FPGA is provided and dedicated to performing a mapping operation and another FPGA is provided and configured to perform an alignment operation, although in some cases a single FPGA may be provided and configured to be at least partially reconfigured between performing both a mapping and an alignment operation.

[0638] Additionally, other operations in the segmentation that may also be performed by reconfigurable or dedicated FPGAs may include performing a BCL conversion/transposition operation, a Smith-Waterman operation, an HMM operation, a local realignment operation, and/or various other variant calling operations. Likewise, various of the segmentation operations may be configured to be performed by one or more of the associated CPUs/GPUs/QPUs of the system. Such operations may be one or more less computationally intensive operations of the segmentation, such as to perform a sort, a deduplication, and other variant calling operations. Thus, the comprehensive system may be configured to perform a combination of operations partly via the CPU/GPU/QPU and partly via hardware, such as via an FPGA/ASIC of the system.

[0639] Accordingly, as can be seen with respect to Figure 40B, in various deployments of the cloud-based system 50, the system may include a plurality of computing resources, including a plurality of instances and/or tiers of instances, such as wherein the instances and/or tiers of instances are configured to perform one or more of the BioIT segmentation operations disclosed herein. For example, in various deployments of the system, multiple CPU/GPU/QPU and/or wired integrated circuit instances may be provided to perform dedicated genomic segmentation analysis functions provided herein. For example, multiple FPGA instances may be provided to perform dedicated genomic analysis operations, such as one FPGA instance to perform mapping, another to perform alignment, another to perform local realignment and/or other Smith-Waterman operations, another to perform HMM operations, and the like.

[0640] Similarly, multiple CPU/GPU/QPU instances may be provided to perform dedicated genomic analysis operations, such as a CPU/GPU/QPU instance to perform signal processing, classification, deduplication, compression, various variant calling operations, and the like. In such cases, an associated memory or memories may be provided, such as between the various segmentation computation steps, to receive result data as it is computed, compiled, and processed through the system, such as between the multiple CPU and/or FPGA instances and/or layers thereof. Additionally, it should be noted that the size of the multiple CPU and/or FPGA instances may vary depending on the computational needs of the cloud-based system and may range from small to medium to large to very large, and the number of CPU/GPU/QPU and FPGA/ASIC instances may vary accordingly.

[0641] Additionally, as can be seen with respect to Figure 40B, the system may further include a workflow manager that is configured to schedule and direct the movement of data through the system and from one instance to another and/or from one memory to another. In some cases, the memory may be a plurality of memories that are dedicated memories that are instance-specific, and in other cases, the memory may be one or more memories that are configured to be elastic and thus have the ability to be switched from one instance to another, such as a switchable elastic block storage memory. In still other cases, the memory may be non-instance-specific and thus have the ability to be communicably coupled to a plurality of instances, such as for elastic file storage.

[0642] Additionally, the workflow manager may be a dedicated instance itself, such as a CPU/GPU/QPU core that is dedicated and/or configured to determine what jobs need to be performed and when and what resources will be used to perform those jobs, as well as to queue the jobs and direct them from resource to resource, e.g., instance to instance. The workflow manager may include or may be otherwise configured as a load estimator and/or form an elastic control node that is a dedicated instance that may be operated by a processor, e.g., a CPU/GPU/QPU core. In many cases, the workflow manager may have a database connected to it, which may be configured to manage all jobs that need to be, are being, or have been processed. Thus, the WMS manager may be configured to detect and manage how data flows through the system, determining how to allocate system resources and when to bring more resources online.

[0643] As indicated above, in certain instances, a server and/or workbench-based solution may be provided wherein the computing device includes a plurality of X CPU core servers having a size of Y that may be configured to power one or more FPGAs having a size of Z, wherein X, Y, and Z are numbers that may vary depending on the processing needs of the system but should be selected and/or otherwise configured to be optimized, e.g., 10, 14, 18, 20, 24, 30, etc. For example, typical system configurations are optimized to perform the BioIT operations of the system described herein. Specifically, certain system configurations have been optimized so as to maximize data flow from multiple CPU/GPU/QPU instances to multiple integrated circuits, such as FPGAs, of the system, wherein the CPU and/or FPGA sizes may vary relative to each other based on the processing needs of the system. For example, one or more of the CPU and/or FPGA may have a size that is relatively small, medium, large, extra large, or extra-extra large. More specifically, the system architecture may be configured such that the CPU/FPGA hardware is sized and configured to operate in an optimally efficient manner so as to keep both instance platforms busy at all times, such as where the CPUs outnumber the FPGA(s) 4 to 1, 8 to 1, 16 to 1, 32 to 1, 64 to 2, etc.

[0644] Thus, while it is generally good to have large FPGA capabilities, however, it may not be efficient to have a high-capacity FPGA to process data if there is sufficient data that needs to be processed being fed into the system. In such a case, only a single or partial FPGA may be deployed. In particular, in an ideal arrangement, the workflow management system directs the data flow to identified CPUs and/or FPGAs that are configured to keep the system and its components computing full-time. For example, in an exemplary configuration, one or more, e.g., 2, 3, or 4 or more CPU/GPU/QPU cores may be configured to feed data to a small, medium, large, extra-large FPGA, or a portion thereof. Specifically, in one embodiment, a specific CPU instance may be provided, such as for performing one or more of the BiolT processing operations disclosed herein, such as wherein the CPU instance is accessible in the cloud and includes up to 4, 8, 16, 24, 30, 36 CPU cores, such cores may or may not be configured to be operationally coupled to a portion of one or more FPGAs.

[0645] For example, a cloud accessible server rack 300 may be provided wherein the server includes a CPU core instance having from about 4 CPU cores to about 16 to about 24 CPU cores that are operatively connected to an FPGA instance. For example, an FPGA instance may be provided, such as wherein an average size of an FPGA is X, and the included FPGA may be of a size of about 1/8X, X, 2.5X up to 8X, or even about 16X, or more. In various cases, additional CPU/GPU/QPU cores and/or FPGAs may be included and/or provided as a combined instance, such as wherein there is a large amount of data to process, and wherein the number of CPU cores is selected so as to keep the FPGA(s) busy at all times. Therefore, the ratio of CPUs to FPGAs (or FPGAs) can be provided by being matched in a way that optimizes data flow, and thus the system can be configured to be elastically scaled up or down as needed, for example, to minimize overhead while optimizing utilization based on workflow.

[0646] However, when the CPU(s) does not generate enough work to keep the FPGA busy and/or fully utilized, the configuration will be less than ideal. Provided herein, therefore, is a flexible architecture of one or more instances, which may be directly coupled to each other, or may have the ability to be coupled to each other, in a manner that is tailored so that the CPU/FPGA software/hardware runs efficiently so as to ensure that the present CPUs/GPUs/QPUs optimally feed the available FPGA(s), and/or a portion thereof, so as to keep both instance platforms busy at all times of execution. Then, enabling such a system to be accessible from the cloud will ensure that a plurality of data is provided to the system so that it can be queued by the workflow manager and directed to the specific CPU/FPGA resources that are configured and have the capacity to receive and process the data in an optimally efficient manner.

[0647] For example, in some configurations, cloud-accessible instances may include a plurality of numbers and sizes of CPUs/GPUs/QPUs, and additionally, there may be cloud-accessible instances that include a plurality of numbers and sizes of FPGAs (or ASICs) and/or QPUs. There may even be instances that have a combination of such instances. However, in various iterations, the provided CPU/GPU/QPU and/or FPGA/QPU and/or mixed instances may have far more of one instance and/or far less of the other instance to efficiently execute the present BioIT segmentation processing platforms disclosed herein. Accordingly, systems and architectures, flexible combinations thereof, and/or methods for deploying the same for efficient formation and use of a bioinformatics and/or genomics segmentation processing platform as made accessible via the cloud 50 are disclosed herein.

[0648] In such systems, the number and configurations of selected CPU(s)/GPUs/QPUs may be selected and configured to process the least computationally intensive operations, and the number and configurations of FPGA(s) and/or QPUs may be adapted to handle computationally intensive tasks, such as where data is seamlessly passed back and forth between CPU/GPU/QPU and FPGA/QPU instances. Additionally, one or more memories may be provided for storing data, e.g., results data, between the various steps of the procedures and/or between the various different instance types, thereby avoiding a substantial period of instance latency. Specifically, during mapping and alignment, very little of the CPU/GPU is utilized due to the intensive nature of the computations that these tasks are configured to be performed by the hardware implementations. Likewise, during variant calling, tasks can be divided in such a way as to be approximately evenly distributed among the CPU/FPGA instances in their tasks, such as when Smith-Waterman and HMM operations can be performed by hardware and various other operations can be performed by software running on one or more CPU/GPU/QPU instances.

[0649] Accordingly, the architectural parameters presented herein are not necessarily limited to the architecture of a cluster, but rather the system is configured to have more flexibility in organizing its deployments and relying on the workflow manager to determine which instances are active when, how, and for how long, and direct which computations are performed on which instances. For example, the number of CPUs and/or FPGAs to be brought online, and operationally coupled to each other, should be selected and configured in such a manner that the activated CPUs and FPGAs, as well as their attendant software/hardware, are kept optimally busy. In particular, the number of CPUs, and their operation, should be configured so as to keep the number of FPGAs, or a portion thereof, busy at all times, such that the CPUs are optimally and efficiently feeding the FPGA(s) so as to keep both instances and their component parts running proficiently.

[0650] Thus, in this manner, the workflow management controller of the system may be configured to evaluate the workflow and organize and divide the workflow such that the tasks that can most optimally be performed by the CPUs/GPUs/QPUs are directed to the number of CPUs required to optimally perform those operations, and the tasks that can most optimally be performed by the FPGA(s)/ASICs/QPUs are directed to the number of FPGAs required to optimally perform those operations. An elastic and/or efficient memory may be additionally included to efficiently transmit the results of these operations from one instance to another. In this manner, a combination of machines and memories may be configured and combined to optimally scale based on the extent of the work to be performed, and the optimal configuration and use of the instances to best perform that work efficiently and cost effectively.

[0651] Specifically, the cloud-based architectures presented herein show that several known shortcomings in prior architectural offerings can cause inefficiencies that can be overcome by flexibly allowing more CPU/GPU/QPU core instances to access multiple different hardware instances, e.g., FPGAs, or portions thereof, that have been organized in a more intentional manner so as to be optimized for deployment in that format. For example, the system may be configured such that there is a higher proportion of available CPU/GPU instances that can be remotely accessible so as to be busy full-time producing results data that can be optimally fed to the available FPGA/QPU instance(s) so as to keep the selected FPGA instance(s) busy full-time. Therefore, it is desirable to provide a structured architecture that is as efficient as possible and busy full-time. It should be noted that configurations where very few CPUs feed into very many FPGAs, such that one or more of the FPGAs are being underutilized, are not efficient and should be avoided.

[0652] In a deployment, as can be seen with respect to Figure 40B, the architecture can be configured to include virtually several different layers or tiers, such as a first tier that has a first number of X CPU cores, e.g., from 4 to about 30 CPU cores, and a second tier that has from 1 to 12 or more FPGA instances, where the size of the FPGAs can be in the range of small to medium to large, etc. A third tier of CPU cores and/or a fourth tier of additional FPGAs and so on can also be included. Thus, there are many instances available on the cloud-based server 300, such instances that include simply CPUs or GPUs and/or instances that include FPGAs and/or combinations thereof, such as in one or more tiers described herein. Consequently, in such a way, the architecture can be organized in a flexible and elastic way so that specific more intensive computing functions are performed by the hardware instances or QPUs and those functions that can be operated by the CPUs are directed to the appropriate CPU/QPU at the appropriate level for general processing purposes and, when necessary, the number of CPU/FPGA instances can be increased or decreased within the system as needed.

[0653] For example, an architecture may be elastically scaled to minimize system overhead while maximizing optimal utilization. Specifically, the architecture may be configured to maximize efficiency and reduce latency by combining the multiple instances into multiple different virtual tiers. In particular, a plurality, e.g., a significant portion and/or all, of the Tier 1 CPU/GPU instances may be configured to feed into the multiple Tier 2 FPGA instances, which have been specifically configured to perform specific functions, such as a mapping FPGA and an alignment FPGA. At a further tier, one or more additional CPUs (or the same as Tier 1) may be provided, such as to perform a sort and/or deduplication operations and/or various variant calling operations. Still further, one or more additional tiers of FPGAs may be configured to perform a Needleman-Wunsch, Smith-Waterman, an HMM, variant calling operation, and the like. Therefore, the first-tier CPUs may be engaged to form an initial level of a genomics analysis, such as to perform general processing steps, including queuing and preparing data for further segmentation analysis, such data, once processed by one or a plurality of CPUs, may be fed to one or more additional levels of dedicated FPGA instances, such as where the FPGA instance is configured to perform compute-intensive functions.

[0654] Thus, in a particular implementation, the CPU/GPU instances in the slicer route data, once prepared, to one or two Tier 2 FPGA mapping and alignment instances. Once mapping has been performed, the result data may be stored in memory and/or otherwise fed to an alignment instance, where alignment may be performed, for example, by at least one dedicated Tier 2 FPGA instance. Likewise, the processed mapped and aligned data may then be stored in memory and/or directed to a Tier 3 CPU instance for further processing, which may be the same Tier 1 instance or a different instance, such as to perform a less compute-intensive genomic analysis function, such as to perform a classification function. Additionally, once the Level 3 CPUs have performed their processing, the resulting data may then be routed back to the other Level 2 FPGA instances or to a Level 4 FPGA instance, such as for additional genomic processing intensive functions, such as to perform a Needleman-Wunsch (NW), SmithWaterman (SW) processing function, for example, on a dedicated NW or SW FPGA instance. Likewise, once SW analysis has been performed, such as via a dedicated SW FPGA, then the processed data may be sent to one or more associated memories and/or further down the processing thread, such as to another, e.g., Level 4 or 5, or back to the Level 1 or 3 CPU and/or FPGA instance, such as to perform HMM and/or Variant Call analysis, such as on a dedicated FPGA and/or additional layer of CPU processing core.

[0655] In such a manner, latency and efficiency issues may be overcome by combining the various different instances, at one or more different levels, in order to provide a segmentation platform for genomic processing. Such a configuration may involve more than one scaling and/or combining instance, where the instances may be configured such that they specialize in performing dedicated functions. In such a case, the Mapping FPGA instance performs only mapping, and similarly, the Alignment FPGA instance performs only alignment, and so on, rather than a single instance performing end-to-end segmentation processing. However, in other configurations, one or more of the FPGAs may be at least partially reconfigured, such as between performing segmentation tasks. For example, in certain embodiments, as the genomic analyses to be performed herein are a multi-step process, the code of an FPGA may be configured to be changed midway through the processing process, such as when the FPGA completes the mapping operation, it may be reconfigured to perform one or more of alignment, variant calling, Smith-Waterman, HMM, and the like.

[0656] Thus, the segmentation manager, e.g., workflow management system, may function to manage the queue of genomic processing requests that are formulated by Level I CPU instances to be broken into distinct jobs, aggregated, and routed to the appropriate job-specific CPU and then to job-specific FPGA instances for further processing, such as for mapping and/or alignment, e.g., at Level 2, wherein the mapped and aligned data, once processed, may be sent backward or forward to the next level of CPU/FPGA processing of the result data, such as for performance of various steps in the variant calling module.

[0657] For example, the variant calling function may be split into a plurality of operations, which may be performed in software, then routed to Smith-Waterman and/or HMM processing on one or more FPGA hardware instances, and then may be dispatched to a CPU for continuous variant calling operations, such as when the entire platform is elastic and/or efficiently scaled and deployed to minimize the cost of expensive FPGA instances, while maximizing utilization, minimizing latency, and thus optimizing operations. Consequently, in this way, fewer hardware instances are required due to their pure processing capabilities and wired specificity, and thus the number of FPGAs to the number of CPUs may be minimized and their usage of, e.g., the FPGAs may be maximized, and thus the system optimized to keep all instances busy at all times. Such a configuration is ideally designed for genomic processing analysis, especially for mapping, alignment, and variant calling.

[0658] An additional structural element that may be included, e.g., as an attachment, to the slicing architecture disclosed herein is one or more elastic and/or efficient memory modules, which may be configured to function to provide block storage of the data, e.g., results data, as it is transitioned through the slicing. Accordingly, one or more Elastic Block Data Stores (EBDS) and/or one or more efficient (flexible) block data storage modules may be inserted between one or more of the processing levels, e.g., between different instances and/or instance levels. In such a case, the storage device may be configured such that as data is processed and results are obtained, the processed results may be directed to the storage device before being routed to the next processing level, such as by a dedicated FPGA processing module. The same storage device may be employed across all instances, or instance levels, or a plurality of storage devices may be employed across multiple instances and/or instance levels, such as for storing and/or compiling and/or queuing results data. Accordingly, one or more memories may be provided such that multiple instances of the system may be coupled to and/or have access to the same memory so as to have the ability to view and access the same or similar files. Accordingly, one or more elastic memories (memories with the ability to be coupled to a plurality of instances sequentially) and/or efficient memories (memories with the ability to be coupled to a plurality of instances simultaneously) may be present, whereby multiple instances of the system are configured to read from and write to the same or similar memory.

[0659] For example, in an exemplary embodiment with respect to configurations employing such elastic memories, prior to sending data directly from one instance and/or one processing tier to another, the data may be routed to an EBDS, or other memory device or structure, e.g., an effective memory block, for storage and thereafter routed to the appropriate wired processing module. Specifically, a block storage module may be attached to a node for in-memory storage where data may be written to the BSD for storage at one tier and the BSD may be flipped to another node to route the stored data to the next processing tier. In this manner, one or more, e.g., multiple, BDS modules may be included in the segmentation and configured to be flipped from one node to another in order to participate in the transition of data across the segmentation.

[0660] Additionally, as indicated above, a more flexible Archive Storage Device may be employed, such as a device that has the ability to be coupled to one or more instances concurrently, such as without having to be switched from one to another. In such a manner, the system may be elastically scaled at each level of the system, such that at each level there may be a different number of nodes to process the data at that level, and once processed, the result data may be written to one or more associated EBDS devices which may then be switched to the next level of the system so as to make the stored data available to the next level of processors for performance of their specific tasks at that level.

[0661] Accordingly, there are many steps in the processing segmentation, e.g., at its attendant nodes, once the data is prepared for processing, e.g., preprocessing, wherein the data, once prepared, is directed to an appropriate processing instance at a level where resulting data can be generated, then the result data can be stored, e.g., within an EDS device, queued and prepared for the next processing stage by being inverted to the next node of instances and routed to the next instance for processing by the next order of FPGA and/or CPU processing instances, wherein additional result data can be generated, and again, once generated, the result data can be directed back to the same level or forward to the next level of EDS for storage before being advanced to the next processing stage.

[0662] In particular, in a particular deployment, the flow through the segmentation might look like the following: CPU (e.g., a 4-CPU core, or C4 instance): data prepared (queued and/or stored); FPGA (e.g., a 2XL FPGA - 1/8 of a full server, or an F1 instance): Mapping, temporary storage; FPGA (e.g., a 2XL FPGA - 1/8 of a full server, or an F1 instance): alignment, temporary storage; CPU: sorting, temporary storage; CPU: deduplication, temporary storage; CPU: variant 1 call, temporary storage; FPGA (e.g., an F1 or 16XL or F2 instance): SmithWaterman, temporary storage; FPGA (e.g., F1 or F2 instance): HMM, temporary storage; CPU: variant 2 call, temporary storage; CPU: VCGF, temporary storage, and so on. Additionally, a workflow management system may be included to control and/or direct the flow of data through the system, such as where the WMS may be deployed on a CPU core, such as a 4-core CPU or C4 instance. It is noted that one or more of these steps may be performed in any logical order and may be implemented by any suitably configured resource, such as implemented in software and/or hardware, in a variety of different combinations. And it should be noted that any of these operations may be performed on one or more CPU instances and one or more FPGA instances at one or more theoretical processing levels, such as to form the BioIT processing described herein.

[0663] As indicated, a workflow manager may be included, such as where the WMS is deployed on one or more CPU cores. Therefore, in various cases, the WMS may have a database operationally coupled to it. In such a case, the database includes the various operations or jobs to be queued, pending jobs, as well as the history of all previous or currently being performed jobs. As such, the WMS monitors the system and the database to identify any new jobs to be performed. Accordingly, when a pending job is identified, the WMS initiates a new analysis protocol on the data and delegates the same to the appropriate instance node(s). Accordingly, the workflow manager keeps track of where all the input files are, stored, being processed, or to be stored, and therefore directs and instructs the instances of the various processing nodes to access the respective files at each location, begin reading the files, begin implementing the processing instructions, and where to write the result data. And so the WMS directs the systems to pass the result data to idle line processing nodes. The WMS also determines when new instances need to be started and brought online so that each step or level of processing can be dynamically scaled. So the WMS identifies, organizes, and directs the distinct jobs that need to be performed at each level, and further directs the result data that is written to memory for storage, and once the job is completed, another node starts up, reads the next job, and performs the next iterative operation.

[0664] In such a manner, input jobs may be spread across many different instances, such instances may be scaled, for example, independently or collectively, by including fewer or more instances. Such instances may be employed to construct nodes in order to more efficiently balance resource usage, such instances may comprise a partial or full instance. The workflow manager may also direct and/or control the usage of one or more memories, such as between the processing steps disclosed herein. The multiple instances may also include complementary programming to allow them to communicate with each other and/or with the multiple memories, in order to virtualize the server. The WMS may also include a load estimator in order to elastically control the usage of the nodes.

[0665] Additionally, with respect to the use of memory, one or more EBDS or other suitably configured data and/or file storage devices may be attached to one or more of the plurality of nodes, e.g., across multiple instance levels, such as for temporary storage between multiple different processing steps. Thus, the storage device may be a single storage device configured to be attached to all of the plurality of instances, e.g., an efficient memory block such as elastic file storage, or it may be multiple storage devices, such as a storage device per instance or instance type that is switchable between instances, e.g., elastic block storage device. Accordingly, in such a manner, each tier of processing instances and/or memory may be elastically scaled on an as-needed basis, such as across each of the different nodes or node levels, such as to process one or multiple genomes.

[0666] In view of the architecture herein, one or a plurality of genomes may be introduced into the system for processing, such as from one or more lanes of a flow cell of a Next Generation Sequencer, as indicated in Figure 1. Specifically, providing a cloud-based server system 300 as described herein will allow a plurality of jobs to be stacked and/or queued for processing, such jobs may be processed by the various different instances of the system simultaneously or sequentially. Accordingly, the segmentation may be configured to support a plurality of jobs being processed by a virtual array of processors that are coupled to appropriately configured memory devices so as to facilitate efficient processing and data from one instance to another. Furthermore, as indicated, a single memory device may be provided, wherein the memory device is configured to be coupled to a plurality of different instances, e.g., at the same time. In other instances, the memory device may be an elastic-type memory device that may be configured to be coupled to a first instance, for example, at a single time, and then being reconfigured and/or otherwise decoupled from the first instance and switched to a second instance.

[0667] As such, in a deployment, one or more elastic block storage devices may be included, and the system may be configured to include a switching control mechanism. For example, a switch controller may be included and configured to control the operation of such memory devices as they switch from one instance to another. Such a configuration may be arranged to allow data transfer by segmenting dedicated processors, thereby increasing system efficiency, e.g., across all instances, such as by streaming data through the system, allowing each tier to scale independently and bringing processors online as needed to scale efficiently.

[0668] Additionally, the workflow management system algorithm may be configured to determine the number of jobs, the number of resources to process those jobs, the order of processing, and direct the flow of data from one node to another by flipping or switching one or more flexible switching devices, and when necessary, may bring additional resources online to handle an increase in workflow. It should be noted that this configuration may be adapted to avoid copying data from one instance to the next, which is inefficient and time-consuming. Instead, flipping elastic storage from one set of instances to another, e.g., pulling it from one node and pinning it to a second node, may greatly increase the efficiency of the system. Furthermore, in many cases, instead of employing EBSD, one or more elastic file storage devices, e.g., single memory devices with the ability to be coupled to a plurality of instances without needing to be flipped from one to another, may be employed, so as to increase data throughput between instances, making the system even more efficient. Additionally, it should be noted, as indicated earlier in this document, in another configuration, the CPUs of the architecture may be directly coupled to each other. Likewise, the multiple FPGAs may be directly coupled to each other. And, as indicated above, the CPUs may be directly coupled to the FPGAs, such as where such coupling is via a tight coupling interface as described above.

[0669] Accordingly, with respect to user storage and access of the generated results data, from a system wide perspective, all of the generated results data need not be stored. For example, the generated results data will typically be in a particular file format, e.g., a BCL, FASTQ, SAM, BAM, CRAM, VCF file. However, each of these files is extensive and storing all of them would consume a lot of memory and thus incur a lot of expense. However, an advantage of the present devices, systems and methods herein is that all of these files need not be stored. Rather, given the fast processing speeds and/or fast compression and decompression rates achievable by the system's components and methods, only a single file format, e.g., a compressed file format, need be stored, such as in the cloud-based database 400. Specifically, only a single data file format need be stored, from which file format, by deploying the system's devices and methods, all other file formats can be derived. And, because of the high compression and decompression rates achievable by the system, the same is typically a compressed file, e.g., a CRAM file.

[0670] In particular, as can be seen with respect to Figure 40A, in one implementation, a user of a local computing resource 100 may upload data, such as genomic data, e.g., a BCL and/or FASTQ file, to the system via the cloud 50 for receipt by the cloud-based computing resource, e.g., the server 300. The server 300 will then either temporarily store the data 400 or begin processing the data in accordance with the job request by the user 100. Upon processing the input data, the computing resource 300 will thereby generate result data, such as in a SAM or BAM and/or VCF file. The system may then store one or more of these files, or may compress one or more of these files and store those. However, in order to reduce cost and make more efficient use of resources, the system may store a single file, e.g., compressed, from which all other file formats may be generated, such as using the devices and methods disclosed herein. Accordingly, the system is configured to generate data files, e.g., results data, which may be stored in a database 400 associated with server 300 that is accessible via cloud 50, in a cost-effective manner.

[0671] Accordingly, using a local computing resource 100, a user of the system may log on and access the cloud-based server 300 , may upload data to the server 300 or database 400, and may request that one or more jobs be performed on that data. The system 300 will then perform the requested jobs and store the results data in the database 400. As noted, in particular cases, the system 300 will store the generated results data in a single file format, such as a CRAM file. Furthermore, with the click of a button, the user may access the stored file, and with another click of a button, all other file formats may then be made accessible. For example, in accordance with the methods disclosed herein, given the fast processing capabilities of the systems, they would then be processed and generated in the background, e.g., immediately, thereby decreasing processing time and storage load as well as costs, such as where computing and storage functions are bundled together.

[0672] Particularly, there are two parts of this fast storage process that are enabled by the speed of performing the accelerated operations disclosed herein. More particularly, because multiple mapping, alignment, classification, deduplication, and/or variant calling processing operations can be deployed in a wired and/or quantum processing configuration, the production of the results data, in one or more file formats, can be rapidly achieved. Additionally, because of the closely coupled architectures disclosed herein, a seamless compression and storage of the results data, e.g., in a FASTQ, SAM, BAM, CRAM, VCF file format, is further achieved.

[0673] Still further, because of the accelerated processing provided by the system devices and because of their seamless integration with associated storage devices, data resulting from system processing operations, such data to be stored, can be both efficiently compressed prior to storage and decompressed subsequent to storage. Such efficiencies thus reduce storage costs and/or penalties associated with decompressing files prior to use. Accordingly, because of these advantages, the system can be configured to allow for the uninterrupted compression and storage of only a single file type, with immediate regeneration of any of the other file types as needed or requested by the user. For example, a BAM file, or a compressed SAM or CRAM file associated therewith, can be stored, and from that file the others can be generated, for example, in a forward or reverse direction, such as to reproduce a VCF or FASTQ or BCL file, respectively.

[0674] For example, in one embodiment, a FASTQ file may be originally input to the system, or otherwise generated, and stored. In such a case, when going in the forward direction, a checksum of the file may be taken. Likewise, once the result data is produced, when going backwards, another checksum may be generated. These checksums may then be used to ensure that any additional file formats to be generated and/or recreated by the system, in the forward or reverse direction, match each other identically and/or to their compressed file formats. In such a manner, it may be ensured that all necessary data is stored, in as efficient a manner as possible, and the WMS knows exactly where the data is stored, what file format it is stored in, what the original file format was, and from this data, the system may regenerate any file format in an identical manner going forwards or backwards between file formats (once the template is originally generated).

[0675] Thus, the speed advantage of "just in time" compilation is enabled in part by hardware- and/or quantum-enabled generation of the relevant files, such as in generating a BAM file from a previously generated FASTQ file. In particular, compressed BAM files, including SAM and CRAM files, are not typically stored within a database because of the increased time it takes before processing to decompress the stored compressed file. However, the JIT system allows this to be done without substantial penalties. More particularly, by implementing the devices and processes disclosed herein, not only can generated sequence data be compressed and decompressed rapidly, e.g., nearly instantaneously, it can also be stored efficiently. Additionally, from the stored file, in whatever file format it is stored, any of the other file formats can be regenerated in just moments.

[0676] Thus, as can be seen with reference to Figure 40C, when hardware and/or quantum accelerated processing performs various secondary processing procedures, such as mapping and alignment, classification, deduplication, and variant calling, an additional compression step may also be performed, such as in an all-in-one process, prior to storage in compressed form. Then, when the user wishes to analyze or otherwise use the compressed data, the file may be retrieved, decompressed, and/or converted from one file format to another, and/or analyzed, such as by the JIT engine(s) being loaded onto the wired processor or configured within the quantum processor, and subjecting the compressed file to one or more JIT segmentation procedures.

[0677] Accordingly, in various cases, when the system includes an associated FPGA, the FPGA may be completely or partially reconfigured, and/or a quantum processing engine may be arranged, so as to perform a JIT procedure. In particular, the JIT module may be loaded into the system and/or configured as one or more engines, such engines may include one or more compression engines 150 that are configured to work in the background. Thus, when a given file format is called, the JIT system may perform the necessary operations on the requested data so as to produce a file in the requested format. Such operations may include compression and/or decompression as well as conversion so as to derive the requested data in the identified file format.

[0678] For example, when genetic data is generated, it is usually produced in a raw data format, such as a BCL file, which can then be converted to a FASTQ file, for example, by the NGS generating the data. However, with the present system, raw data files, such as in BCL or another raw file format, can be continuously streamed or otherwise transmitted to the JIT module, which can then convert the data to a FASTQ file and/or another file format. For example, once a FASTQ file is generated, the FASTQ file can then be processed as disclosed herein and a corresponding BAM file can be generated. And likewise, from the BAM file, a corresponding VCF can be generated. Additionally, SAM and CRAM files can also be generated during the appropriate steps. Each of these steps can be performed very quickly, especially once the appropriate file format has been generated. Therefore, once the BCL file is received, for example, directly from the sequencer, the BCL can be converted to a FASTQ file or it can be directly converted to a SAM, BAM, CRAM and/or VCF file, such as a hardware and/or quantum implemented mapping/alignment/classification/variant calling procedure.

[0679] For example, in one usage model, on a typical sequencing instrument, a large number of different subject genomes may be loaded into the individual lanes of a single sequencing instrument to be run in parallel. Consequently, at the end of the run, a large number of diverse BCL files, derived from all the different lanes and representing the entire genomes of each of the different subjects, are generated in a multiplex complex. Accordingly, these multiplexed BCL files may then be demultiplexed, and respective FASTQ files may be generated representing the genetic code for each individual subject. For example, if, in a sequencing run, N BCL files are generated, these files will need to be demultiplexed, layered, and stitched together for each subject. This stitching is a complex process in which each subject's genetic material is converted into BCL files, which may then be converted into a FASTQ file or used directly for mapping, alignment, and/or classification, variant calling, and the like. This process can be automated to greatly speed up the various steps of the process.

[0680] Furthermore, as can be seen with respect to Figure 40A, once such data has been generated 110 and therefore needs to be stored, e.g., in whatever file format is selected, the data may be stored in a password protected and/or encrypted memory cache, such as in a dedicated genomics dropbox-type memory 400. Accordingly, as the generated and/or processed genetic data exits the sequencer, the data may be processed and/or stored and made available to other users on other systems, such as in a dropbox-type cache 400. In such a case, the automated bioinformatics analysis segmentation system may then access the data in the cache and begin processing the same automatically. For example, the system may include a management system, e.g., a workflow management system 151, that has a controller, such as a microprocessor or other intelligence, e.g., artificial intelligence, that manages the retrieval of BCL and/or FASTQ files, e.g., from memory cache, and then directs the processing of that information so as to generate a BAM, CRAM, SAM, and/or VCF, thereby automatically generating and outputting the various processing results and/or storing the same in memory dropbox 400.

[0681] A unique benefit of JIT processing, as implemented within this usage model, is that JIT allows the multiple genetic files produced to be compressed, for example, prior to data storage, and to be decompressed on the fly prior to use. Thus, JIT processing may compile and/or compress and/or store the data as it is exiting the sequencer, where such storage is in a secure genomic dropbox memory cache. Such genomic dropbox cache 400 may be a cloud-accessible memory cache 50 that is configured for storage of genomic data received from one or more automated sequencers 110, such as where the sequencer(s) is located remotely from the memory cache 400.

[0682] In particular, once sequence data has been generated 110, for example by a remote NGS, it may be compressed 150 for transmission and/or storage 400 so as to reduce the amount of data being uploaded and stored in the cloud 50. Such upload, transmission, and storage may be accomplished quickly because of the data compression 150 that occurs in the system, just as prior to transmission. Additionally, once uploaded and stored in the cloud-based memory cache 400, the data may then be retrieved, locally 100 or remotely 300, to be processed in accordance with the BiolT segmentation devices, systems, and methods disclosed herein to generate a mapping, alignment, classification, and/or variant calling file, such as a SAM, BAM, and/or CRAM file, which may then be stored along with a metafile that presents information as to how the generated file, e.g., SAM, BAM, CRAM, etc. file, was produced.

[0683] Therefore, when taken together with the metadata, the compressed SAM, BAM and/or CRAM file can then be processed to produce any of the other file formats, such as FASTQ and/or VCF files. Consequently, as discussed above, JIT can immediately be used to regenerate the FASTQ or VCF file from the compressed BAM file and vice versa. The BCL file can also be regenerated in a similar manner. It should be noted that SAM and CRAM files can be similarly compressed and/or stored and can be used to produce one or more of the other file formats. For example, a CRAM file, which can be deconverted from CRAM, can be used to produce a variant call file and likewise for the SAM file. Therefore, only the SAM, BAM and/or CRAM file need be saved and from these files the other file formats, e.g. VCF, FASTQ, BCL files, can be reproduced.

[0684] Accordingly, as can be seen with respect to Figure 40A, a mapping and/or alignment and/or classification and/or variant calling instrument 110, e.g., a workbench computer, may be at the site 100 and/or another corresponding second instrument 300 may be remotely located and made accessible in the cloud 50. Such a configuration, together with the devices and methods disclosed herein, is adapted to enable a user to rapidly perform a BiolT analysis "in the cloud" as disclosed herein so as to produce results data. The results data may then be processed so as to be compressed, and once compressed, the data may be configured for transmission, e.g., back to the local computing resource 100, or may be stored in the cloud 400 and made accessible via a cloud-based interface by the local computing resource 100. In such a case, the compressed data may be a SAM, BAM, CRAM, and/or VCF file.

[0685] Specifically, the second computing resource 300 may be another workbench solution, or it may be a server-configured resource, such as wherein the computing resource is accessible via the cloud 50 and is configured to represent a mapping and/or alignment and/or classification and/or variant calling instrument. In such a case, a user may request that the cloud-based server 300 perform one or more BioIT jobs on the uploaded data, e.g., BCL and/or FASTQ data. In such a case, the server 300 will then access the stored and/or compressed file(s) and may process the data so as to rapidly process such data and generate one or more results data, which data may then be compressed and/or stored. Additionally, from the results data file, one or more BCL, FASTQ, SAM, BAM, VCF, or other file formats may be generated, e.g., on the fly, using JIT processing. This configuration thus alleviates the typical transfer speed bottleneck.

[0686] Therefore, in various embodiments, the system 1 may include, a first mapping and/or alignment and/or classification and/or variant calling instrument 100, which may be positioned locally 100, such as for local data production, compression 150 and/or storage 200; and a second instrument 300 may be positioned remotely and associated in the cloud 50, whereby, the second instrument 300 is configured to receive the generated and compressed data and store the same, for example, via an associated storage device 400. Once stored, the data may be accessed by decompressing and converting the stored files into one or more of the other file formats.

[0687] Therefore, in an implementation of the system, data, e.g., raw sequence data, such as in a BCL or FASTQ file format, that is generated by a data generating apparatus, e.g., a sequencer 110, may be uploaded to and stored in the cloud 50, such as in an associated genomics dropbox-type memory cache 400. This data may then be accessed directly by the first mapping and/or alignment and/or classification and/or variant calling instrument 100, as described herein, or may be accessed indirectly by the server resource 300, which may then process the sequence data to produce mapped, aligned, classified, and/or variant result data.

[0688] Accordingly, in various embodiments, one or more of the storage devices disclosed herein may be configured to be accessible, with appropriate permissions, via the cloud. For example, a plurality of the system's results data may be compressed and/or stored in a memory, or other suitably configured database, wherein the database is configured as a genomics dropbox cache 400, such as wherein a plurality of results data may be stored in a SAM, BAM, CRAM, and/or VCF file, which may be accessible remotely. Specifically, it should be noted that with respect to Figure 40A , a local instrument 100 may be provided, wherein the local instrument may be associated with the sequencing instrument 110 itself or may be remote therefrom but associated with the sequencing instrument 110 via a local cloud 30, and the local instrument 100 may further be associated with a local storage facility 200 or remote memory cache 400, such as wherein the remote memory cache is configured as the genomics dropbox. Furthermore, in various instances, a second mapping and/or alignment and/or classification and/or variant calling instrument 300, e.g., a cloud-based instrument, with appropriate authorities, may also be connected to the genomics dropbox 400 so as to access files, e.g., compressed files, stored by the local computing resource 100, and may then decompress such files to make the results available for further, e.g., secondary or tertiary, processing.

[0689] Accordingly, in various instances, the system may be continuously streamed such that as data is generated and exits sequencer 110, such as in raw data format, it may be immediately uploaded to cloud 50 and stored in a genomics dropbox 400, or it may be streamed to a BioIT processing system 300 for further processing and/or compression prior to being uploaded and stored 400. Once stored within memory cache 400, the system may then immediately queue the data for retrieval, compression, decompression, and/or for further processing, such as by another associated BioIT processing apparatus 300, which, when processed into results data, may then be compressed and/or stored 400 for later use. At this point, a tertiary processing segment may be initiated whereby the stored result data from secondary processing may be decompressed and used, such as for tertiary analysis, in accordance with the methods disclosed herein.

[0690] Therefore, in various embodiments, the system may be segmented such that all data output from sequencer 110 may either be compressed, for example, by a local computing resource 100, prior to transfer and/or storage 200, or the data may be transferred directly to the genomics dropbox folder for storage 400. Once so received, the stored data may then be substantially immediately queued for retrieval and compression and/or decompression, such as by a remote computing resource 300. After being decompressed, the data may be substantially immediately available for processing, such as for mapping, alignment, classification, and/or variant calling, to produce secondarily results data that may then be recompressed for storage. The compressed secondary result data can then be accessed, for example, in the genomic dropbox 400, decompressed, and/or used in one or more tertiary processing procedures. Because the data can be compressed when stored and decompressed substantially immediately when retrieved, it is available for use by many different systems and in many different bioanalytical protocols at different times simply by accessing the storage cache of dropbox 400.

[0691] Therefore, in such ways, the BioIT platform segmentations presented herein can be configured to offer incredible flexibility of data generation and/or analysis and are adapted to handle input of particular forms of genetic data in multiple formats in order to process the data and produce output formats that are compatible with various downstream analyses. Accordingly, as can be seen with respect to Figure 40C, presented herein are systems and methods for performing genetic sequencing analysis, which can include one or more of the following steps: First, a file input is received, the input can be in one or more of a FASTQ or BCL or other form of genetic sequence file format, such as in a compressed file format, such file can then be decompressed and/or processed through various steps disclosed herein in order to generate a VCF/gVCF, such file can then be compressed and/or stored and/or transmitted. Such compression and/or decompression may occur at any suitable stage throughout the process.

[0692] For example, once a BCL file is received, it may undergo segmentation analysis, such as in a sequential manner, as disclosed herein. For example, once received, the BCL file may be converted and/or demultiplexed, such as into a FASTQ and/or FASTQgz file format, and such file may be sent to a mapping and/or alignment module, e.g., of a server 300, to be mapped and/or aligned in accordance with the apparatuses and methods of use thereof described herein. Additionally, in various cases, the mapped and aligned data, such as into a SAM or BAM file format, may be sorted by position and/or any duplicates may be marked and removed. The files may then be compressed, such as to produce a CRAM file, for example, for transmission and/or storage, or they may be forwarded to a variant calling module, for example, HMM, to be processed to produce a variant calling file, VCF or gVCF.

[0693] More specifically, as can be seen with respect to FIGS. 40C and 40D, in certain instances, the file to be received by the system may be continuously streamed or otherwise transferred to the system directly from the sequencing apparatus, e.g., NGS 110, and as such, the transferred file may be in a BCL file format. When the received file is in a BCL file format, it may be converted and/or otherwise demultiplexed into a FASTQ file for processing by the system, or the BCL file may be processed directly. For example, platform segmentation processors may be configured to receive BCL data that is continuously streamed directly from the sequencer, as described with respect to FIGS. 1, or they may receive data in a FASTQ file format. However, receiving sequence data directly as it is continuously streamed from the sequencer is useful because it allows data to go directly from raw sequencing data to be directly processed, for example, in one or more of SAM, BAM and/or VCF/gVCF for output.

[0694] Accordingly, once the BCL and/or FASTQ file is received, for example, by a computing resource 100 and/or 300, it may be mapped and/or aligned by the computing resource, such mapping and/or alignment being performed into single-end or paired-end reads. For example, once received, the sequence data may be compiled into reads, for analysis, such as with read lengths that may be in the range of about 10, or about 20, such as 26, or 50, or 100, or 150 base pairs or less up to about 1 K, or about 2.5 K, or about 5 K, even about 10 K base pairs or more. Likewise, once mapped and/or aligned, the sequence can then be classified, such as by position classification, such as through reference range discretization and/or by classification of discretizations by reference position. Furthermore, the sequence data can be processed by means of duplicate marking, such as based on starting position and CIGAR chain, in order to generate a high quality duplicate report, and any marked duplicates can be removed at that point. Consequently, a mapped and aligned SAM file can be generated, which can be compressed to form a BAM/CRAM file, such as for storage and/or further processing. Furthermore, once the BAM/CRAM file has been retrieved, the mapped and/or aligned sequence data can be forwarded to a system variant calling module, such as a haplotype variant caller with reassembly, which in some cases may employ one or more of a Smith-Waterman alignment and/or Hidden Markov Model that may be implemented in a combination of software and/or hardware to generate a VCF.

[0695] Therefore, as seen in Figure 40D, the system and/or one or more of its components may be configured to have the capability to convert BCL data into FASTQ or SAM/BAM/CRAM data formats, which may then be sent through the system for further processing and/or data reconstruction. For example, once the BCL data is received and/or converted into a FASTQ file and demultiplexed and/or deduplicated, the data may then be forwarded to one or more of the segmentation modules disclosed herein, such as for mapping and/or alignment, which, depending on the number of samples being processed, will result in the production of one or more, e.g., multiple, SAM/BAM files. These files may then be classified, deduplicated, and forwarded by a variant calling module to produce one or more VCF files. These steps may be repeated for greater context and accuracy. For example, once the sequence data is mapped or aligned, e.g. to produce a SAM file, the SAM file may then be compressed into one or more BAM files, which may then be passed to a VCF engine for conversion by system processing to a VCF/gVCF, which may then be compressed into a CRAM file. Consequently, the files to be output through the system may be a Gzip and/or a CRAM file.

[0696] In particular, as can be seen with respect to Figures 40C and 40D, one or more of the files, once generated, may be compressed and/or transferred from one system component to another, e.g., from a local resource 100 to a remote resource 300, and once received, may then be decompressed, e.g., if previously compressed, or converted/demultiplexed. More particularly, once a BCL file is received, either by a local resource 100 or a remote resource 300, it may be converted to a FASTQ file which may then be processed by the system's integrated circuitry so as to be mapped and/or aligned, or may be transmitted to a remote resource 300 for such processing. Once mapped and/or aligned, the resulting sequence data, e.g. in a SAM file format, may be further processed, such as by being compressed one or more times, e.g. in a BAM/CRAM file, such data may then be processed by position sorting, duplicate marking, and/or variant calling, the results of which, e.g. in a VCF format, may then be compressed once more and/or stored and/or transmitted, such as from a remote resource 300 to a local resource 100.

[0697] More particularly, the system may be adapted to process BCL data directly, thereby eliminating a FASTQ file conversion step. Likewise, BCL data may be fed to segmentation to produce a unique output VCF file per sample. Intermediate SAM/BAM/CRAM files may also be generated on demand. The system, therefore, may be configured to receive and/or transmit one or more data files, such as a BCL or FASTQ data file containing sequence information, and process the same to produce a data file that has been compressed, such as a SAM/BAM/CRAM data file.

[0698] Accordingly, as can be seen with respect to Figure 41 A, a user may wish to access the compressed file and convert it to a native version of the generated BCL 111c and/or FASTQ 111d file, such as to subject the data to additional, e.g., more advanced signal processing 111b, such as for error correction. Alternatively, the user may access the raw sequence data, e.g., in a BCL or FASTQ file format 111, and subject such data to additional processing, such as for mapping 112 and/or alignment 113 and/or other related functions 114/115. For example, the results data from these procedures may then be compressed and/or stored and/or subjected to further processing 114, such as for classification 114a, deduplication 114b, recalibration 114c, local realignment 114d, and/or compression/decompression 114e. The same or another user may then wish to access the compressed form of the mapped and/or aligned results data and then perform further analysis on the data, such as to produce one or more variant calls 115, e.g., via HMM, Smith-Waterman, Conversion, etc., which may then be compressed and/or stored. An additional user of the system may then access the compressed VCF file 116, decompress it, and subject the data to one or more tertiary processing protocols.

[0699] Furthermore, a user may wish to perform a segmentation comparison. Mapping/alignment/classification/variant calling is useful for performing various genomic analyses. For example, if additional DNA or RNA analysis, or some other type of analysis, is desired later, a user may wish to run the data through another segmentation, and therefore having access to the original regenerated data file is very useful. Likewise, this process may be useful such as where it may be desired to create or recreate a different SAM/BAM/CRAM file, such as where there is a new or different reference genome generated, and therefore it may be desired to re-do the mapping and alignment to the new reference genome.

[0700] Storing compressed SAM/BAM/CRAM files is additionally useful in that it allows a user of system 1 to benefit from the fact that a reference genome forms the main chain of the result data. In such a case, it is not the data that agrees with the reference that is important, but rather how the data disagrees with the reference. Therefore, only those data that disagree with the reference are essential for storage. Consequently, system 1 can benefit from this fact by storing only what is important and/or useful to users of the system. Thus, the entire genome file (showing agreement and disagreement with the reference), or a subportion thereof (showing only agreement and disagreement with the reference), can be configured to be compressed and stored. It can therefore be seen that since only the differences and/or variations between the reference and the genome being examined are the most useful to examine, in many embodiments only these differences need to be stored, as anything that is the same as the reference does not need to be reviewed again. Consequently, since any given genome differs only slightly from a reference, e.g. 99% of human genomes are typically identical, once the BAM file is created, only the variations between the reference genome need to be reviewed and/or saved.

[0701] Additionally, as can be seen with respect to Figure 41B, another useful component of a cloud accessible system 1 provided herein is a workflow management controller 151 that can be used to automate system workflow. Such system animation can include utilizing the various system components to access data, either locally 100 or remotely 300, as and/or where it becomes available, and then substantially automatically subjecting the data to additional processing steps, relative to the BioIT segmentations disclosed herein. Accordingly, the workflow management controller 151 is a core automation technology for directing the various segmentations of the system, e.g., 111, 112, 113, 114, and/or 115, and in various cases can employ an artificial intelligence component 121a.

[0702] For example, the system 1 may include an artificial intelligence (A/I) module that is configured to analyze the various data from the system and, in response thereto, communicate its findings to the workflow management system 151. Particularly, in various instances, the A/I module may be configured to analyze the various genomic data presented to the system, as well as the result data that is generated by processing such data, so as to identify and determine various relationships between such data and/or with any other data that may be input to the system. More particularly, the A/I module may be configured to analyze various genomic data in correspondence with a plurality of other factors so as to determine any relationships, e.g., effect-based relationships, between the various factors, e.g., data points, that may be informative as to the effects of the considered factors on the determined genomic data, e.g., variance data, and vice versa.

[0703] Specifically, as described in greater detail below, the A/I module may be configured to correlate a subject's genomic data generated by the system with any electronic medical records, for that subject or others, so as to determine any relationships between the same and/or any other relevant factors and/or data. Accordingly, such other data that may be used by the system in determining any relevant effects and/or relationships that such factors may have on a subject and/or his/her genomic data and/or health include: NIPT data, NICU data, Cancer-related data, LDT data, Environmental and/or Ag Bio data, and/or such other data. For example, additional data to be analyzed may be derived by such other factors as environmental data, clade data, microbiome data, methylation data, structural data, e.g., read-through or chimeric data, germline variant data, allele data, RNA data, and other such data related to a subject's genetic material. Therefore, the A/I module can be used to link various related data flowing through the system to the determined variants in the genome of one or more subjects together with one or more other possible related effect-based factors.

[0704] In particular, the AI engine may be configured to run on a CPU/GPU/QPU, and/or it may be configured to run as an accelerated AI engine that may be deployed on an FPGA and/or Quantum Processing Unit. Specifically, the AI engine may be associated with one or more, e.g., all, of the various databases of the system, so as to allow the AI engine to explore and process the various data flowing through the system. Additionally, when a subject whose genome is being processed provides the appropriate authorization to access genomic and patient registry data, the system is then configured to correlate the various data sets with each other and may further mine the data to determine various meaningful matches, associations, and/or relationships.

[0705] More specifically, the A/I module may be configured to implement a machine learning protocol with respect to the input data. For example, the genomic data of a plurality of subjects that are generated from the analyses being performed herein may be stored in a database. Likewise, with appropriate authorizations and authentications, the Electronic Medical/Health Records (EMR) for the subject whose genomic DNA was processed may be obtained and may likewise be stored in the database. As described in greater detail below, the processing engine (or engines) may be configured to analyze the subjects' genomic data, as well as their EMR data, in order to determine any correlations between the two. These correlations will then be explored, the observed relationships reinforced, and the results thereof may be used to more effectively and efficiently perform the various functions of the system.

[0706] For example, the AI processing engine may access the subject's genomic data, in correlation with the known diseases or conditions of those subjects, and from such analysis, the AI module may learn to perform predictive correlations based on such data, so as to become increasingly capable of predicting the presence of a disease and/or other similar conditions in other individuals. In particular, by determining such correlations between the genomes of others with their EMR, e.g., with respect to the presence of disease markers, the AI module may learn to identify such correlations, e.g., system-determined disease markers, in the genomes of others, thereby having the ability to predict the possibility of a disease or other identifiable conditions. More particularly, by analyzing a subject's genome against known or determined genetic markers of disease and/or determining the variance of the subject's genome and/or further determining a potential relationship between the genomic data and the subject's health condition, e.g., EMR, the A/I module may have the ability to draw conclusions not only for the subject being sampled, but for others who may be sampled in the future. This may be done, for example, in a systematic manner, on a subject-by-subject basis, or it may be done within populations and/or within geographically distinct locations.

[0707] More particularly, with respect to the present systems, a backlog of reads is produced. The backlog may overlap regions known to have a higher probability of significant variance. Accordingly, the system will, on the one hand, analyze the backlog to determine the presence of variance, while at the same time, based on its previous findings, it will already know the probability that a variance should or should not exist, e.g., it will have an initial prediction as to what the response should be. The likelihood of the expected variance existing or not will be informative when analyzing that region of the genomes of others. For example, this may be one data point in a sum of data points being used by the system to make better variant calls, and/or better associate such variants with one or more disease states or other health conditions.

[0708] For example, in an exemplary learning protocol, A/I analysis may include taking an electronic image of an array of one or more regions in a genome, such as for those regions suspected of encoding one or more health conditions, and associating that image with known variance calls from other arrays, such as where that variance may or may not be known to be related to disease states. This may be done repeatedly with the system learning to process the information, make appropriate associations, and make correct calls more and more quickly and with greater accuracy. Once this has been accomplished for several, e.g., all, of the regions of the genome suspected of causing a disease, the same may be repeated for the rest of the genome, e.g., until the entire genome has been reviewed. Likewise, this may be repeated multiple times for a plurality of sample genomes, without stopping, so as to train the system, e.g. the variant caller, to make more accurate calls, earlier, and with greater efficiency, and/or to enable the tertiary processing module to better identify unhealthy conditions.

[0709] Accordingly, the system receives many inputs with known answers, performs the analysis and computes the answer, and thus learns from the process, e.g., rendering an image of a cluster relative to a genome, and then learning to make a call based on another genome, earlier and earlier as it is more readily determined that future clusters resemble previously captured images that are known to be related to unhealthy conditions. Thus, the system may be configured to learn to make predictions as to the presence of variants, e.g., based on pattern recognition, and/or by predicting the relationship between the presence of such a variance with one or more medical conditions.

[0710] More specifically, the more the system performs partial or whole genome analysis and determines the relationship between variations and various conditions, e.g., on a plurality of samples, the better at making predictions, e.g., based on partial or whole genome images of the clusters, the system becomes. This is useful when predicting disease states based on the cluster images and/or other read analysis and may include constructing a correlation between one or more of the EMR (including phenotypic data), the cluster image and/or known variants (genotypic data) and/or disease states or conditions, e.g., from which predictions may be made. In various cases, the system may include a transcription function, such as to have the ability to transcribe any physical notes that may be a part of the subject's medical record, so as to include such data with associations.

[0711] In one use model, a subject may have a mobile tracker and/or sensor, such as a mobile phone or other computing device, which may be configured to track the subject's location as well as to sense the user's environmental and/or physiological conditions at that location. Other sensed data may also be collected. For example, the mobile computing device may include a GPS tracker, and/or its location may be determined by triangulation by cell towers, and may be further configured to transmit its collected data, for example, via cellular, WIFI, Bluetooth, or other suitably configured communications protocol. Thus, the mobile device may track and categorize environmental data pertaining to geographic locations, environmental conditions, physiological status, and other sensed data that the subject owning the mobile computer encounters in his or her daily life. The collected location, environmental, physiological, health, and/or other associated data, e.g., ZNA data, may then be transmitted, e.g., regularly and periodically, to one or more of the database systems herein, wherein the collected ZNA data may be correlated to the subject's patient history, e.g., EMR records, and/or their genomic data, as determined by the system herein.

[0712] Likewise, in various instances, one or more of these data may be forwarded from the ZNA collection and analysis platform to a central repository, e.g., at a government facility, for analysis on a larger scale, e.g., nationwide, such as in accordance with the Artificial Intelligence disclosed herein. For example, the database, e.g., a government-controlled database, may have recorded environmental data to which the subject's environmental data may be compared. For example, in an exemplary case, a NICU test may be performed on a mother, a father, and their child, and then, throughout the lives of all three, their environmental and genomic and medical record data may be continually collected and correlated with each other and/or one or more models, such as over the lifetime of the individuals, especially with respect to the onset of mutations, such as due to environmentally impactful factors. This data collection can be carried out throughout the individual's life and can be carried out on a family as a whole basis, in order to better build a data collection database and better predict the effects of such factors on genetic variation, and vice versa.

[0713] Accordingly, the workflow management controller 151 allows the system 1 to receive inputs from one or more sources, such as one or multiple sequencing instruments, e.g., 110a, 110b, 110c, etc., and multiple inputs from a single sequencing instrument 110, wherein the data being received represents the genomes of multiple subjects. In such cases, the workflow management controller 151 not only keeps track of all incoming data but also efficiently organizes and facilitates secondary and/or tertiary processing of the received data. Accordingly, the workflow management controller 151 allows the system 1 to seamlessly connect to large and small sequencing centers, wherein all types of genetic material may be coming through one or more sequencing instruments 110 at the same time, all of which may be transferred to the system 1, such as via the cloud 50.

[0714] More specifically, as can be seen with respect to Figure 41A, in various instances, one or a plurality of samples may be received into system 1, and therefore, system 1 may be configured to efficiently receive and process the samples, either sequentially or in parallel, such as in a multiple sample processing regime. Accordingly, to continuously stream and/or automate the processing of multiple samples, the system may be controlled by a comprehensive Workflow Management System (WMS) or LIMS (laboratory information management system) 151. WMS 151 enables users to easily schedule multiple workflow runs for any given segmentation, as well as to fine-tune or speed up NGS analysis algorithms, platform segmentations, and their attendant applications.

[0715] In such a case, each run sequence may have a barcode on it indicating what type of sequence it is, what file format it is in, and/or what processing steps were performed and what processing steps need to be performed. For example, the barcode may include a manifest indicating "this is a genome run, from subject X, in file format Y, so this data needs to go through segmentation Z" or it may equally indicate "this is result data from A that needs to go into this reporting system". Accordingly, as data is received, processed, and transmitted through the system, the barcodes and results will be uploaded to the workflow management system 151, such as LIMS (laboratory information management system). LIMS, in this case, may be a standard tool that is employed for laboratory management or it may be a specifically designed tool used to manage process flow.

[0716] In any case, the workflow management controller 151 tracks a barcoded sample from when it arrives at a given site, e.g., for storage and/or processing, until the results are sent to the user. In particular, the workflow management controller 151 is configured to track all data as it flows through the system from end to end. More particularly, as the sample arrives, the barcode associated with the sample is read, and based on that reading, the system determines which workflows are requested and prepares the sample for processing. Such processing may be simple, such as being performed across a single genome segmentation, or it may be more complex, such as being performed across multiple, e.g., five segmentations, that need to be stitched together. In a particular model, the generated or received data may be run through the system to produce processed data, the processed data may then be run through an equivalent GATK module, the results may be compared, and then the sample may be passed to another segmentation for further processing, e.g., tertiary 700. See Figure 41B.

[0717] Thus, the system as a whole may be executed in accordance with several different processing threads. Indeed, many system processes may be interconnected, wherein the workflow manager 151 is notified or otherwise determines that a new job is pending, quantifies the job matrices, identifies the resources available to perform the required analyses, loads the job into the system, receives incoming data, for example from the sequencer 110, loads the same, and then processes the same. In particular, once the workflow is established, it may be saved, and then a modified barcode is assigned to that workflow, and the automated process occurs in accordance with the workflow guidelines.

[0718] Prior to the present automated workflow management system 151, it would take a long period of time for many bioinformatics professionals to configure and set up the system and its component parts, and it would then require more time to actually perform the analysis. To make matters more complicated, the system would have to be reconfigured before receiving the next sample to analyze, requiring even more time to reconfigure the system to analyze the new set of samples. With the technology disclosed herein, the system can be fully automated. In particular, the present system is configured to automatically receive multiple samples, map them to multiple different workflows and segmentations, and run them on the same system card or on multiple different system cards.

[0719] Accordingly, the workflow management system 151 reads the job requirements from the barcodes, allocates resources to perform the jobs, e.g., regardless of location, updates the sample barcode, and directs the samples to the allocated resources, e.g., processing units, for processing. Accordingly, the workflow manager 151 determines the secondary and/or tertiary analysis protocols 700 that will be performed on the received samples. These processing units are resources that are available to outline and perform the operations allocated to each data set. In particular, the workflow controller 151 controls the various operations associated with receiving and reading the sample, determining jobs, allocating resources for the performance of those jobs, e.g., secondary processing, connecting all system components, and advancing the sample set through the system from component to component. The controller 151 therefore acts to manage the overall system from beginning to end, e.g., from sample receipt to VCF generation, and/or through tertiary processing, see Figure 41B.

[0720] In additional instances, as can be seen with respect to Figure 41C, the system 1 may include an additional layer of processing modules 800, as configured to provide additional processing, for example, of secondary and/or tertiary processing result data, such as for diagnosis, disease discovery, and/or therapeutics and/or prophylaxis thereof. For example, in various instances, an additional layer of processing 800 may be provided, such as for disease diagnosis, therapeutic treatment, and/or prophylactic prevention 70, such as including NIPT 123a, NICU 123b, Cancer 123c, LDT 123d, AgBio123e, and other such disease diagnosis, prophylaxis, and/or treatments employing data generated by one or more of the present primary and/or secondary and/or tertiary pipelines.

[0721] Accordingly, herein is disclosed a system 1 for producing and using a local 30 and/or global 50 hybrid cloud network. For example, currently, the local cloud 30 is primarily used for private storage, such as at a remote storage location 400. In such a case, data computation is performed locally 100 by a local computing resource 140, and when storage needs are extensive, the local cloud 30 may be accessed in order to store data generated by the local computing resource 140, such as through the use of a private remote storage resource 400. Therefore, the generated data is typically managed entirely on-site locally 100. In other embodiments, the data may be generated, computed, and managed entirely off-site by securely connecting to a remote computing resource 300 via a private cloud interface 30.

[0722] In particular, in a general deployment of a bioinformatics analysis platform, the local compute 140 and/or storage 200 functions are maintained locally at site 100. However, when storage needs exceed local storage capacity, the data may be updated via a local cloud access 30 so as to be stored securely off-site at site 400. Furthermore, when there is a need for stored data 400 to be made available to other remote users, such data may be transferred and made available via a global cloud interface 50 to remote storage 400 as well, but for global access. In such a case, when the computing resources 140 required for performance of the computing functions are minimal but the storage requirements are extensive, the computing function 140 may be maintained locally 100 while the storage function 400 may be maintained remotely, for example, for private or global access, wherein the fully processed data is transferred back and forth between the local processing function 140, such as for local processing only, and the storage function 400, such as for remote storage 400 of the processed data, such as by employing JIT protocols disclosed above herein.

[0723] For example, this may be exemplified with respect to sequencing function 110, such as with a typical NGS, wherein computing and/or data generation resource 100 is configured to perform functions required for sequencing genetic material to produce genetic sequenced data, e.g., reads, wherein data is generated at site 100 and/or transferred locally to site 30. Such reads, once generated, such as by the on-site NGS, may then be transferred, e.g., as a BCL or FASTQ file, over cloud network 30, such as to storage 400 at a remote location 300 in a manner that they are recalled from cloud 30 when needed, such as for further processing. For example, once sequence data has been generated and stored, e.g., 400, the data may then be recalled, e.g., for local use, such as for the performance of one or more of secondary 600 and/or tertiary 700 processing functions, i.e., at a location remote from storage facility 400, e.g., locally 100. In such a case, local storage facility 200 merely serves as a storage cache, in which data is placed while awaiting transfer to or from the cloud 30/50, such as to or from remote storage facility 400.

[0724] Likewise, when the computing function is extensive, such as requiring one or more remote computing servers or computing cluster cores 300 to process the data, and when the storage demands for storing the processed data 200 are relatively minimal compared to the computing resources 300 required to process the data, the data to be processed may be sent, such as via the cloud 30, to be processed by a remote computing resource 300, wherein the resource may include one or more computing cores or clusters of resources, e.g., one or more super computing resources. In such a case, once data has been processed by the cloud-based computer core 300, the processed data may then be transferred via the cloud network 30 to be stored locally 200 and made readily available for use by the local computing resource 140, such as for local analysis and/or diagnostics. Obviously, remotely generated data 300 can also be stored remotely 400.

[0725] This may be further exemplified with respect to a typical secondary processing function 600, such as when pre-processed sequenced data, e.g., read data, that is stored locally 200 and is accessed, such as by local computing resource 100, and transmitted via a cloud internet 30 to a remote computing facility 300 to be further processed therein, e.g., in a secondary 600 or tertiary processing function 700, to obtain processed result data that may then be sent back to local facility 100 for storage 200 therein. This may be the case when a local practitioner generates sequenced read data using a local data generation facility 100, e.g., automated sequencer, to produce a BCL or FASTQ file, and then sends that data over the network 30 to a remote computing facility 300, which then operates one or more functions on that data, such as a Burrows-Wheeler or Needlemen-Wunsch transform and/or Smith-Waterman alignment function on that sequence data to generate results data, e.g., in a SAM file format, which may then be compressed and transmitted over the Internet 30/50, e.g., as a BAM file, to the local computing facility 100 for examination by it in one or more local managed processing protocols, such as to produce a VCF that may then be stored locally 200. In various cases, the data may also be stored remotely 400.

[0726] However, what is needed is a seamless integration between the coupling between local 100 and remote 300 computer processing, as well as between local 200 and remote 400 storage, such as in the hybrid cloud-based system 50 disclosed herein. In such a case, the system may be configured such that local 100 and remote 300 computing resources are configured to operate seamlessly together, so that data to be processed thereby may be allocated in real time to either the local 200 or remote 300 computing resource without paying an extensive penalty due to throughput and/or operational efficiency. This may be the case, for example, when the software and/or hardware and/or quantum processing to be installed or otherwise operated by computing resources 100 and 300 are configured to correspond to each other and/or are the same or functionally similar, e.g., the hardware and/or software is configured in the same way in order to operate the same algorithms in the same way on the generated and/or received data.

[0727] For example, as can be seen with respect to Figure 41A, a local computing resource 100 can be configured to generate or to receive generated data and therefore can include a data generation engine 110, such as for generating and/or analyzing primary data 500, for example, in order to produce a FASTQ and/or BCL sequence file. This data generation engine 110 can be or can be associated with a local computer 100, as described throughout this document, that has a processor 140 that can be configured to operate one or more software applications and/or can be wired to perform one or more algorithms, such as in a wired configuration, on the generated and/or acquired data. For example, the data generation engine 110 may be configured to one or more of generate data, such as sequence data 111. In various embodiments, the generated data may be sensed data 111a, such as data that is detectable as a change in voltage, ionic concentration, electromagnetic radiation, and the like; and/or the data generation engine 110 may be configured to generate and/or process signal, e.g., analog or digital signal data, such as data representing one or more nucleotide identities in a sequence or chain of associated nucleotides. In such a case, the data generation engine 110, e.g., sequencer 111, may be further configured to perform preliminary processing of the generated data such as for signal processing 111b or to perform one or more base calling operations 111c, such as on the data, in order to produce sequence identity data, e.g., a FASTQ and/or BCL file 111d.

[0728] It should be noted that in this case, the produced data 111 may be generated locally and directly, such as by a local data generation and/or computation resource 110 , e.g., an NGS or sequencer-on-a-chip. Alternatively, the data may be produced locally and indirectly, e.g., by a remote generation and/or computation resource, such as a remote NGS. The data 111, for example, in BCL and/or FASTQ file format, once produced may then be transferred indirectly via the local cloud 30 to the local computing resource 100 for such secondary processing 140 and/or storage therein at a local storage resource 200, such as while awaiting further local processing 140. In such a case, when the data generating resource is remote from the local processing resources 100 and/or storage 200, the corresponding resources may be configured such that the remote and/or local storage, remote and local processing, and/or communication protocols employed by each resource may be adapted to integrate smoothly and/or seamlessly with each other, for example, by operating the same hardware, similar and/or equivalent hardware, and/or having the same hardware configurations, similar and/or equivalent hardware configurations, and/or employing the same communications and/or transfer protocols, which in some cases may have been deployed at the time of the data generation process. of manufacture or later.

[0729] Specifically, in one implementation, these functions may be deployed in a wired configuration such as where the sequencing function and the secondary processing function are maintained on the same chip or associated chip or set of chips, for example, such as where the sequencer and secondary processor are directly interconnected on one chip as described herein. In other implementations, these functions may be deployed on two or more separate devices via software, for example, on a quantum processor, CPU, or GPU that has been optimized to allow the two remote devices to communicate continuously with each other. In other implementations, a combination of hardware and software implementations optimized to perform the recited functions may also be employed.

[0730] More specifically, the same configurations may be deployed with respect to the performance of mapping, alignment, classification, variant calling, and/or other functions that may be employed by local 100 and/or remote 300 computing resources. For example, local and/or remote computing resources 300 may include software and/or hardware configured to perform one or more secondary layers 600 of processing functions 112-115, and/or tertiary layers 700/800 of processing functions, on locally and/or remotely generated data, such as genetic sequence data, in a manner that the processing and results thereof may be seamlessly shared with and/or stored by one another. In particular, the local computing function 100 and/or the remote computing function 300 may be configured to generate and/or receive primary data, such as genetic sequence data, for example in a BCL and/or FASTQ file format, and operate one or more secondary processing protocols 600 and/or tertiary processing protocols 700 on such generated and/or acquired data. In such a case, one or more of these protocols may be implemented in a software, hardware, or combinatorial format, such as executed on a quantum processor, a CPU, and/or a GPU. For example, the data generation 110 and/or the local 100 and/or remote processing facility 300 may be configured to perform one or more of a mapping operation 112, an alignment operation 113, variant calling 115, or other related function 114 on the acquired or generated data in software and/or hardware.

[0731] Accordingly, in various embodiments, the data generation resource, such as the sequencer 111, e.g., NGS or sequencer-on-a-chip, whether implemented in software and/or hardware or a combination thereof, may be further configured to include an initial layer of processors 500, such as a programmer, various analytical components, comparators, graphers, scorers, and the like, in order to assist the data generator 111, e.g., sequencer, in converting biological information into raw read data, such as in a BCL or FASTQ format file 111d. Furthermore, the local computing resource 100, whether implemented in software and/or hardware, or a combination thereof, may be further configured to include an additional platform of processors 600 such as may include a mapping engine 112, or may otherwise include programming to execute a mapping algorithm on the genetic sequence data, such as to perform Burrows-Wheeler transform and/or other algorithms to construct a hash table and/or execute a hash function 112a on said data, such as for hash seed mapping, so as to generate mapped sequence data. Additionally, the local computing resource 100, whether implemented in software and/or hardware, or a combination thereof, may also include an alignment engine 113 as described herein, or may otherwise include programming to operate an alignment algorithm on genetic sequence data, e.g., mapped sequenced data, such as to perform a gapped and/or gapless Smith-Waterman and/or Needleman-Wunsch alignment, or other similar scoring algorithm 113a on said data, in order to generate aligned sequence data.

[0732] The local computing 100 and/or data generation resource 110 may also be configured to include one or more other modules 114, whether implemented in software and/or hardware, or a combination thereof, that may be adapted to perform one or more other processing functions on genetic sequence data, such as mapped and/or aligned sequence data. Thus, the one or more other modules may include a suitably configured engine 114, or otherwise include programming, to perform the one or more other processing functions, such as a classification function (or functions) 114a, deduplication 114b, recalibration 114c, local realignment 114d, duplicate marking 114f, Base Quality Score Recalibration 114g, and/or a compression function (such as to produce a SAM, reduced BAM, and/or CRAM compression and/or decompression file) 114e, in accordance with the methods described herein. In various instances, one or more of these processing functions may be configured as one or more segmentations of the system 1.

[0733] Similarly, system 1 may be configured to include a module 115, whether implemented in software and/or hardware, or a combination thereof, which may be adapted to process data, e.g., sequenced, mapped, aligned, and/or classified data, in a manner that produces a variant call file 116. In particular, system 1 may include a variant call module 115 for operating one or more variant call functions, such as a Hidden Markov Model (HMM) and/or a GATK function 115a, such as in a wired configuration and/or via one or more software applications, e.g., locally or remotely, and/or a converter 115b therefor. In various instances, such a module may be configured as one or more segmentations of system 1.

[0734] In particular embodiments, as shown in Figure 41B, system 1 may include a local computing function 100 that may be configured to employ a computer processing resource 150 to perform one or more additional computer processing functions on data, e.g., BCL and/or FASTQ data, generated by system data generator 110 or acquired by system acquisition engine 120 (as described herein), such as by being transferred thereto by, e.g., a third party 121, such as via a cloud network 30 or hybrid cloud 50. For example, a third party analyzer 121 may deploy a remote computing resource 300 in order to generate relevant data in need of further processing, such as genetic sequence data or the like, wherein the data may be communicated to system 1 via network 30/50 in order to be further processed. This may be useful, for example, when the remote computing resource 300 is an NGS, configured to take raw biological data and convert it into a digital representation thereof, such as in the form of one or more FASTQ files containing reads of genetic sequence data, and when further processing is desired, such as to determine how the generated sequence of an individual is different from that of one or more reference sequences, as described herein, and/or it is desired to subject the results thereof to further, e.g., tertiary, processing.

[0735] In such a case, the system 1 may be adapted to allow one or more parties, e.g., a primary and/or secondary and/or third party user, to access the associated local processing resources 100 and/or a suitably configured remote processing resource 300 associated therewith, in a manner so as to allow the user to perform one or more quantitative and/or qualitative processing functions 152 on the generated and/or acquired data. For example, in one embodiment, the system 1 may include, e.g., in addition to the primary and/or secondary processing pipelines 600 and/or secondary processing pipelines 600, a third layer of processing modules 700/800, wherein the processing modules may be configured to perform one or more processing functions on the primary and/or secondary processed generated and/or acquired data.

[0736] Particularly, in one embodiment, system 1 may be configured to generate and/or receive processed genetic sequence data 111 that has been mapped 112, aligned 113, classified 114a, and/or further processed remotely or locally 114 in order to generate a variant call file 116, wherein the variant call file may then be subjected to further processing, such as within system 1, such as in response to second-party and/or third-party analysis requests 121. More particularly, system 1 may be configured to receive processing requests from third parties 121 and further be configured to perform such requested secondary 600 and/or tertiary 700/800 processing on the generated and/or acquired data. Specifically, system 1 may be configured to produce and/or acquire genetic sequence data 111, may be configured to take such genetic sequence data and map 112, align 113, and/or classify 114a the same, and process the same to produce one or more variant call files (VCFs) 116, and additionally, system 1 may be configured to perform a tertiary processing function 700/800 on the data, e.g., with respect to the one or more VCFs generated or received by system 1.

[0737] In particular, the system 1 may be configured to perform any form of tertiary processing 700 on the generated and/or acquired data, such as subjecting the same to one or more segmentation processing functions 700 such as to generate genome data 122a, e.g., whole genome, epigenome data 122b, metagenome data 122c, and the like, including genotyping data 122d, e.g., joint genotyping, variant analysis data, including GATK 122e and/or MuTect2 analysis data 122f, among other potential data analytic segmentations, such as a microarray analysis segmentation, exome analysis segmentation, microbiome analysis segmentation, RNA sequencing segmentations, and other genetic analysis segmentations. Additionally, the system 1 may be configured to perform an additional layer of processing 800 on the generated and/or processed data, such as including one or more of non-invasive prenatal testing (NIPT) 123a, ICU N/P 123b, cancer-related diagnostic and/or therapeutic modalities 123c, various laboratory developed tests (LDT) 123d, agricultural biological (Ag Bio) applications 123e, or other such healthcare-related processing function 123f. See Figure 41C.

[0738] Thus, in various embodiments, where a primary user may access and/or configure system 1 and its various components directly, such as through direct access thereto, such as through local computing resource 100 as disclosed herein, system 1 may also be adapted to be accessed by a secondary party, such as one connected to system 1 via a local network or an intranet connection 10 in order to configure and operate system 1 within the local environment. Additionally, in certain embodiments, the system may be adapted to be accessed and/or configured by third parties 121, such as via an associated hybrid cloud network 50 that connects the third parties 121 to system 1, such as via an application program interface (API) accessible via one or more graphical user interface (GUI) components. Such GUI may be configured to allow the third party user to access the system 1 and, using the API, configure the various system components, modules, associated pipelines and other associated data generation and/or processing functionality in order to operate only those system components necessary and/or useful to the third party and/or requested or desired to be operated by the third party.

[0739] Accordingly, in various instances, system 1 as disclosed herein may be adapted to be configurable by a primary, secondary, or tertiary user of the system. In such an instance, system 1 may be adapted to allow the user to configure system 1 and thereby arrange its components in such a manner as to distribute any, all, or a selection of the analytical system capabilities, e.g., 152, to be operated on data that is generated, acquired, or otherwise transferred to the system, e.g., by the primary, secondary, or third-party user, such that system 1 operates only those portions of the system necessary or useful to operate the analyses requested by the user to obtain the desired results therefrom. For example, for these and other such purposes, an API may be included in system 1, wherein the API is configured to include or be otherwise operatively associated with a graphical user interface (GUI) that includes an operable menu and/or related list of system function calls from which the user may select and/or otherwise make in order to configure and operate the system and its components as desired.

[0740] In such a case, the GUI and/or system menu function calls may direct user-selectable operations of one or more of a first stratum of operations 600, including: sequencing 111, mapping 112, alignment 113, classification 114a, variant calling 115, and/or other associated functions 114 in accordance with the teachings herein, such as with respect to the primary and/or secondary processing functions described herein. Furthermore, when the desired GUI and/or system menu function calls may be direct, operations of one or more of a second layer of operations 700 including: a genome, e.g., whole gemome, analysis segment 122a, epigenome segment 122b, metagenome segment 122c, a genotyping, e.g., joint, genotyping segment 122d, variant segments, e.g., GATK 122e and/or MuTect2 122f analysis segments, including structural variant segments, as well as other tertiary analysis segments, such as a microarray analysis segment, exome analysis segment, microbiome analysis segment, RNA sequencing segments, and other genetic analysis segments. Additionally, when desired, the GUI menu and system function calls may direct user-selectable operations of one or more of a third layer of operations 800, including: non-invasive prenatal testing (NIPT) 123a, ICU N/P 123b, cancer-related diagnostic and/or therapeutic modalities 123c, various laboratory developed tests (LDT) 123d, agricultural biological (Ag Bio) applications 123e, or other such healthcare-related processing functions 123f.

[0741] Accordingly, system and menu function calls may include one or more primary, secondary, and/or tertiary processing functions in order to allow the system and/or its component parts to be configured, such as with respect to performing one or more data analysis pipelines as selected and configured by the user. In such a case, the local computing resource 100 may be configured to correspond to and/or mirror the remote computing resource 300 and/or, likewise, the local storage resource 200 may be configured to correspond to and/or mirror the remote storage resource 400 so that the various components of the system may be operated and/or the data generated thereby may be stored locally or remotely in a seamless distributed manner as chosen by the use of the system 1. Additionally, in particular embodiments, the system 1 may be made accessible to third parties to operate proprietary analysis protocols 121a on the generated and/or processed data, such as by operating through an artificial intelligence interface designed to find correlations therebetween.

[0742] The system 1 may be configured to perform any form of tertiary processing on the generated and/or acquired data. Thus, in various embodiments, a primary, secondary, or tertiary user may access and/or configure any level of the system 1 and its various components directly, such as through direct access to the computing resource 100, indirectly, such as through a local network connection 30, or through an associated hybrid cloud network 50 that connects the party to the system 1, such as through an appropriately configured API that has appropriate permissions. In such a case, the system components may be presented as a menu, such as a GUI-selectable menu, wherein the user may select from all of the various processing and storage options desired to be operated on the data presented by the user. Additionally, in various cases, the user may upload his or her own system protocols to be adopted and operated by the system to process various data in a manner designed and selected by the user. In such a case, the GUI and associated API will allow the user to access system 1 and, using the API, add and configure the various system components, modules, associated pipelines and other associated data generation and/or processing functionality in order to operate only those system components that are necessary and/or useful to the party and/or requested or desired to be operated by the party.

[0743] Referring to Figure 41C, one or more of the modules demarcated above, and their respective associated functions and/or features, may be configured to be performed remotely, such as via a remote computing resource 300, and be further adapted to be transmitted to system 1, such as in a seamless transfer protocol via a global cloud-based internet connection 50, such as via a suitably configured data acquisition engine 120. Accordingly, in such a case, a local computing resource 100 may include a data acquisition engine 120, as configured to transmit and/or receive such acquired data and/or associated information.

[0744] For example, system 1 may include data acquisition engine 120 that is configured in a manner that allows continued processing and/or storage of data to occur in an uninterrupted and stable manner, such as via a cloud-based network 50, wherein processing functions are distributed both locally 100 and remotely 300. Similarly, when one or more of the results of such processing may be stored locally 200 and/or remotely 400, such that the system seamlessly allocates to which local or remote resource a given job should be sent for processing and/or storage regardless of where the resource is physically located. Such distributed processing, transfer, and acquisition may include one or more of sequencing functions 111, mapping 112, alignment 113, classification 114a, duplicate marking 114c, deduplication, recalibration 114d, local realignment 114e, Base Quality Score Recalibration 114f, and/or a compression function 114g, as well as a variant calling function 116, as described herein. When stored locally 200 or remotely 400, the processed data, in whatever state it is in the process, may be made available to local 100 or remote 300 processing resources, such as for further processing prior to retransmission and/or re-storage.

[0745] Specifically, system 1 may be configured to produce and/or acquire genetic sequence data 111, may be configured to take such genetic sequence data and process it locally 140 or transfer the data via a suitably configured cloud network 30 or hybrid cloud 50, such as to a remote processing facility for remote processing 300. Additionally, once processed, system 1 may be configured to store the processed data remotely 400 or transfer it back to local storage 200. Accordingly, system 1 may be configured for local or remote generation and/or processing of data, such as where the generation and/or processing steps may be of a first tier of primary and/or secondary processing functions 600, wherein the tier may include one or more of: sequencing 111, mapping 112, alignment 113, and/or classification 114a in order to produce one or more variant call files (VCFs) 116.

[0746] Furthermore, the system 1 may be configured for local or remote data generation and/or processing, such as wherein the generation and/or processing steps may be from a second platform of tertiary processing functions 700, wherein the platform may include one or more of data management and/or acquisition according to a genome segmentation 122a, an epigenome segmentation 122b, a metagenome segmentation 122c, a genotyping segmentation 122d, variants of, e.g., GATK 122e and/or MuTect2, analysis segmentation 122f, as well as other tertiary analysis segmentations, such as a microarray analysis segmentation, a microbiome analysis segmentation, an exome analysis segmentation, as well as RNA sequencing segmentations and other genetic analysis segmentations. Additionally, system 1 may be configured for local or remote generation and/or processing of data, such as wherein the generation and/or processing steps may be a third layer of tertiary processing functions 800, wherein the layer may include one or more of generating and/or acquiring data related to and including: non-invasive prenatal testing (NIPT) 123a, ICU N/P 123b, cancer-related diagnostic and/or therapeutic modalities 123c, various laboratory developed tests (LDT) 123d, agricultural biological applications (Ag Bio) 123e, or other healthcare-related processing functions 123f.

[0747] In particular embodiments, as shown in Figure 41C, system 1 may be further configured to allow one or more parties to access the system and transfer information to or from associated local processing 100 and/or remote processing resources 300, as well as to store information locally 200 or remotely 400 in a manner that allows the user to choose what information is processed and/or stored and where in system 1. In such a case, a user may decide not only what primary, secondary, and/or tertiary processing functions are performed on generated and/or acquired data, but also how those resources are distributed and/or where the results of such processing are stored. For example, in one configuration, the user may select whether the data is generated locally or remotely, or a combination thereof, whether the same is subjected to secondary processing and, if so, to which secondary processing modules the same is subjected and/or which resource operates which of these processes, and may further determine whether the data so generated or acquired is further subjected to tertiary processing and, if so, to which modules and/or which tertiary processing layers the same is subjected and/or which resource operates which of these processes, and likewise where the results of these processes are stored for each step of the operations.

[0748] Particularly, in one embodiment, the user may configure the system 1 of Figure 41A such that the generation of genetic sequence data 111 occurs remotely, such as by an NGS, but the secondary processing 600 of the data occurs locally 100. In such a case, the user may then determine which of the secondary processing functions occur locally 100, such as by selecting processing functions, such as mapping 112, alignment 113, classification 111, and/or producing a VCF 116, from a menu of available processing options. The user may then select whether the locally processed data is subjected to tertiary processing and, if so, which modules are activated in order to further process the data and whether such tertiary processing occurs locally 100 or remotely 300. Likewise, the user may select various options for the various layers of tertiary processing options and where any generated and/or acquired data is to be stored, locally 200 or remotely 400, at any given step or time of operation.

[0749] More particularly, a primary user may configure the system to receive processing requests from third parties, wherein the third parties may configure the system to perform such requested primary, secondary, and/or tertiary processing on the generated and/or acquired data. Specifically, the user or second and/or third party may configure the system 1 to produce and/or acquire genetic sequence data, locally 100 or remotely 200. Additionally, the user may configure the system 1 to take such genetic sequence data and map, align, and/or classify the same, locally or remotely, so as to produce one or more variant call files (VCFs). Additionally, the user may configure the system to perform a tertiary processing function on the data, e.g., with respect to the one or more VCFs, locally or remotely.

[0750] Even more particularly, the user or other party may configure the system 1 in order to perform any form of tertiary processing on the generated and/or acquired data and where such processing is to occur on the system. Thus, in various embodiments, the primary, secondary and/or third party user 121 may access and/or configure the system 1 and its various components directly, such as by directly accessing the local computing function 100, through a local network connection 30 or through an associated hybrid cloud network 50 that connects the party 121 to the system 1, such as through an application program interface (API) accessible through one or more graphical user interface (GUI) components. In such a case, the third party user may access System 1 and use the API to configure the various system components, modules, associated pipelines and other associated data generating and/or processing functionalities in order to operate only those system components that are necessary and/or useful to the third party and/or requested or desired to be operated by the third party, and further allocate which computing resources will provide the requested processing and where the result data will be stored.

[0751] Accordingly, in various instances, system 1 may be configurable by a primary, secondary, or tertiary user of the system, who may configure system 1 to arrange its components in such a manner as to install any, all, or a selection of the analytical system capabilities to be operated on data that the user directly generates, causes to be generated by system 1, or causes to be transferred to system 1, such as through a network associated therewith, such as via data acquisition engine 120. In such a manner, system 1 is configurable to operate only those portions of the system necessary or useful for the analyses desired and/or requested by the requesting party. For example, for these and other such purposes, an API may be included, wherein the API is configured to include an operable GUI menu and/or a related list of system function calls from which the user may select in order to configure and operate the system as desired.

[0752] Additionally, in particular embodiments, system 1 may be made accessible to a primary user and/or third parties, such as government regulators, such as the Federal Drug Administration (FDA) 70b, or allow primary users and/or third parties to collect, compile, and/or access a database of genetic information derived or acquired and/or otherwise compiled by system 1, in order to form an electronic medical record (EMR) database 70a and/or allow government access and/or oversight of the system, such as the FDA for Drug Development Evaluation. System 1 may also be defined to conglomerate, compile, and/or annotate data 70c and/or allow other high-level users to access the same.

[0753] Accordingly, the system 1 and/or its components may be configured to be accessed by a remote user, such as a primary user or third party, and therefore one or more of the computer resources 100 and/or 300 may include a user interface and/or may further include a display device having a graphical user interface to allow a potential user of the system to access the system in order to transmit sample data for input into one or more of the BioIT segmentations disclosed herein and/or to receive result data therefrom. The GUI or other interface may be configured to allow the user to manage the system components, for example, through a suitably configured web portal, and to track sample processing progress, regardless of whether the computing resources to be engaged are available locally 100 or remotely 300. Accordingly, the GUI may list a set of jobs that may be performed, for example, mapping 112, alignment 113, etc. and/or a set of resources to perform the jobs, and the user can self-select which jobs he wants to perform and by which resources. Thus, in a case such as this, each individual user can build on his own unique analysis workflow, or can use a predetermined one, such as by clicking, dragging, or otherwise selecting the particular work projects he wants to perform.

[0754] For example, in one usage model, a dashboard is presented with a GUI interface that may include a plurality of icons representing the various processes that may be deployed and executed on the system. In such an instance, a user may click or drag selected work process icons onto a workflow interface to construct a desired work flow process, which, once constructed, may be saved and used to establish control instructions for the sample set creator. Once the desired work designs have been selected, the workflow management controller 151 may configure the desired work flow processes (e.g., secondary analysis) and then identify and select the resources to perform the selected analysis.

[0755] Once the workflow analysis process begins, the dashboard can be viewed to track progress through the system. For example, a dashboard can indicate how much data is running through the system, what processes are being performed on the data, how much has been accomplished, how much processing remains, which workflows have been completed and what still needs to be accessed, the multiple projects to be executed, and which rounds have been completed. Essentially, complete access to everything running in the system, or a subportion thereof, can be provided to the workspace.

[0756] Furthermore, in various instances, the workspace may include several different user interfaces that may be accessible through one or more tabs. For example, a tab for accessing system controls may be a “local resources tab 100,” which, when selected, allows a user to select control functions that may be deployed locally. Another tab may be configured to access “cloud resources 300,” which, when selected, allows a user to select other control functions that may be deployed remotely. Accordingly, in interaction with the dashboard, a user may select which resources should perform which tasks, and as such may increase or decrease resource usage as required to meet design requirements.

[0757] Thus, as computational complexity increases, and/or increased speed is desired, the user (or the system itself, e.g., WMS 151) can bring more and more resources online as needed, such as by the mere click of a button, instructing the workflow manager to bring local 100 and/or cloud-based resources 300 online as needed to complete the task on the desired schedule. In this manner, although the system is automated and/or controlled by the workflow manager controller 151, a user of the system can still set the control parameters and, when needed, can bring cloud-based resources 300 online. Accordingly, the controller 151 can expand to the cloud 50/300 as needed to bring additional processing and/or storage resources 400 online.

[0758] In various instances, the desktop interface may be configured as a mobile application or "app" that is accessible via a mobile device and/or desktop computer. Accordingly, in one aspect, a genomic marketplace, or cohort, may be provided that allows a plurality of users to collaborate on one or more research projects, so as to form an electronic cohort marketplace that is accessible via the dashboard app, e.g., a web-based browser interface. As such, the system may provide an online forum for conducting collaborative research and/or a marketplace for developing various analytical tools for analyzing genetic data, which system may be accessible directly via the system interface, or via the app, to allow remote control of the system by a user.

[0759] Accordingly, in various embodiments, as can be seen with respect to Figure 42A, a hybrid cloud 50 is provided, wherein the hybrid cloud is configured to connect a local computing resource 100 and/or storage 200 with a remote computing resource 100 and/or storage 200, such as when the local and remote resources are separated from each other distally, spatially, geographically, and the like. In such a case, the local and distal resources may be configured to communicate with each other in a manner so as to share information, such as digital data, seamlessly between the two. In particular, the local resources may be configured to perform one or more types of processing on the data, such as prior to transmission over the hybrid network 50, and the remote resources may be configured to perform one or more types of further processing on the data.

[0760] For example, in a particular embodiment, system 1 may be configured such that a generation and/or analysis function 152 is configured to be performed locally 100 by a local computing resource, such as for the purpose of performing a primary and/or secondary processing function in order to generate and/or process genetic sequence data as described herein. Additionally, in various embodiments, the local resources may be configured to perform one or more tertiary processing functions on the data, such as one or more of a genome, exome, and/or epigenome analysis or a cancer, microbiome, and/or other DNA/RNA processing analysis. Additionally, when such processed data is intended to be transferred, such as to a remote computing 300 and/or storage resource 400, the data may be transformed, such as by means of a suitably configured transformer, wherein the transformer may be configured for indexing, converting, compressing, and/or encrypting the data, such as prior to transfer over the hybrid network 50.

[0761] In particular cases, such as when generated and processed data is transferred to a remote computing resource, e.g., server 300, for further processing, such processing may be global in nature and may include receiving data from a plurality of local computing resources 100, collating such pluralities of data, annotating the data and comparing the same, such as to interpret the data, determine trends therein, analyze the same for various biomarkers, and assist in the development of diagnostic, therapeutic and/or prophylactic treatments. Accordingly, in various instances, remote computing resource 300 may be configured as a data processing hub, such as where data from a variety of sources may be transferred, processed, and/or stored while waiting for the same to be transformed and/or transferred, such as by being accessed by local computing resource 100. More particularly, remote processing hub 300 may be configured to receive data from a plurality of resources 100, process the same, and distribute the processed data back to the plurality of local resources 100 in order to enable collaboration among researchers and/or resources 100. Such collaboration may include various data sharing protocols and may further include preparing the data for transfer, such as by allowing a user of system 1 to select from among various security protocols and/or privacy settings in order to control how the data will be prepared for transfer.

[0762] In a particular case, as shown in Figure 42B, a local computing resource 100 and/or storage resource 200 is provided, such as on-site at a user location. The computing resource 100 and/or storage resource 200 may be coupled to a data generation resource 121, such as an NGS or sequencer-on-a-chip as described herein, such as via a direct connection or an intranet 10, wherein the sequencer 121 is configured to generate genetic sequencing data, such as BCL and/or FASTQ files. For example, sequencer 121 may be part of and/or housed in the same apparatus as that of computing resource 100 and/or storage unit 200, so as to have a direct communicable and/or operable connection thereto, or sequencer 121 and computing resource 100 and/or storage resource 200 may be part of apparatus separate from each other but housed in the same facility and thus connected via a wired connection or intranet 10. In some cases, sequencer 121 may be housed in a separate facility from that of computing resource 100 and/or storage 200 and thus may be connected via an internet 30 or hybrid cloud 50 connection.

[0763] In such cases, the genetic sequence data may be processed 100 and stored locally 200, prior to being transformed, via a suitably configured transformer, or the generated sequence data may be transmitted directly to one or more of the transformer and/or the analyzer 152, such as via a suitably configured local connection 10, intranet 30, or hybrid cloud connection 50, as described above, such as prior to being processed locally. In particular, as the data generation resource 121, the transformer 151 and/or the analyzer 152 may be part of and/or housed in the same apparatus as that of the storage resource 100 and/or the storage unit 200, so as to have a direct communicable and/or operable connection thereto, or the transformer and/or the analyzer 152 and the computing resource 100 and/or the storage resource 200 may be part of apparatus separate from each other but housed in the same facility and thus connected via a wired or intranet connection 10. In some cases, the transformer 151 and/or the analyzer 152 may be housed in a separate facility than that of the computing resource 100 and/or storage 200 and thus may be connected via an internet 30 or hybrid cloud connection 50.

[0764] For example, the transformer may be configured to prepare data to be transmitted prior to analysis or after analysis, such as via a suitably configured computing resource 100 and/or analyzer 152. For example, the analyzer 152 may perform a secondary and/or tertiary processing function on the data as described herein, such as to analyze the generated sequence data with respect to determining its genomic and/or exomic characteristics 152a, its epigenomic features 152b, any various DNA and/or RNA markers of interest and/or cancer indicators 152c and their relationships to one or more microbiomes 152d, as well as one or more other secondary and/or tertiary processes as described herein.

[0765] As indicated, the generated and/or processed data may be transformed, such as by means of a suitably configured transformer, such as prior to transmission through the system 1 from one component thereof to another, such as through a direct, local connection 10, over the internet 30 or in a hybrid cloud 50. Such transformation may include one or more of conversion 151d, such as when data is converted from one form to another; comprehension 151c, including conversion into code, decoding and/or taking data from an otherwise incomprehensible form and transforming it into a comprehensible form, or from one comprehensible form to another; indexing 151b, such as including compiling and/or pasting the generated data from one or more resources and making it findable and/or searchable, such as by means of a generated index; and/or encryption 151a, such as creating a lockable and unlockable password-protected data set, such as prior to transmission over an internet 30 and/or hybrid cloud 50.

[0766] Thus, as can be seen with respect to Figure 42C, in these and/or other such cases, the hybrid cloud 50 may be configured to allow uninterrupted and protected transmission of data by all components of the system, such as when the hybrid cloud 50 is adapted to allow the various users of the system to configure its component parts and/or the system itself in order to meet the user's research and/or development, diagnostic, therapeutic and/or prophylactic testing needs. In particular, the hybrid cloud 50 and/or the various components of the system 1 may be operationally connected to compatible and/or corresponding API interfaces that are adapted to allow a user to remotely configure the various components of the system 1 in order to install the desired resources in the desired manner and to do so additionally locally, remotely or in a combination thereof, as based on the demands of the system and the particulars of the analyses being performed, all the while being able to communicate in a secure, encryptable environment.

[0767] In particular cases, system 1 may include a processing architecture 310, such as an interpreter, that is configured to perform an interpretation function 310. Interpreter 310 may perform one or a series of analytical functions on generated data, such as annotation 311, interpretation 312, diagnosis 313, and/or detection and/or an analysis function to determine the presence of one or more biomarkers, such as in genetic data. Interpreter 313 may be part of or separate from local computing resource 100, such as when interpreter 310 is coupled to computing resource 100 via a cloud interface, such as a hybrid cloud 50.

[0768] Additionally, an additional processing architecture 320 may be included, such as when architecture 320 is configured as a collaborator. Collaborator 320 may be configured to perform one or more functions directed at ensuring the security and/or privacy of data to be transmitted. For example, the collaborator may be configured to secure the data sharing process 321, to ensure transmission privacy 322, to set control parameters 323, and/or to initiate a security protocol 324. The collaborator 313 is configured to enable data sharing, such as to facilitate processing collaboration, as the collaborator 320 may be part of or separate from the local computing resource 100, such as when the collaborator 320 is coupled to the computing resource 100 via a cloud interface, such as a hybrid cloud 50. The interpreter 310, collaborator 320, and/or the local computing resource 100 may be additionally coupled to a remote computing resource 300, such as to enhance system efficiency by offloading computing functions 300 and/or storage 400 to the cloud 50. In various instances, the system 1 may be configured to enable secure third-party analysis 121 occurs, such as when third parties can connect to and engage the system such as through a properly configured API.

[0769] As can be seen in relation to Figure 43, the system 1 can be a multi-layer and/or multiplexed bioanalytical processing platform that includes data generating layers and/or data processing units that each have one or more processing segments that can be implemented in a systematic and concurrent or sequential manner so as to process genetic information from its primary processing stage to a tertiary and/or tertiary processing stage. Particularly, presented herein are devices configured to perform bioanalysis in one or more of hardware and/or software and/or quantum processing implementations, as well as methods of their use, and systems that include the same. For example, in one embodiment, a genomics processing platform may be provided and configured as a plurality of integrated circuits, which integrated circuits may be adapted as, or otherwise included in, one or more of a central or graphics processing unit, such as a general purpose CPU and/or GPU, a wired implementation, and/or a quantum processing unit. Particularly, in various embodiments, one or more segmentations of the genomics processing platform may be configured by one or more integrated and/or quantum circuits of a quantum processing unit.

[0770] Accordingly, the platforms presented herein may be configured to exploit the robust power of software and/or hardware and/or quantum processing implementations optimized for the performance of the various genetic sequencing and/or secondary and/or tertiary processing functions disclosed herein, which may be executed on one or more integrated circuits. Such integrated circuits may be seamlessly coupled together and may be further seamlessly coupled to various other integrated circuits, e.g., CPUs and/or GPUs and/or QPUs, of the system that are configured to execute the various software and/or wire-based applications of tertiary bioanalytical functions.

[0771] Particularly, in various embodiments, these processes may be performed by optimized software running on a CPU, GPU, and/or QPU, and/or may be implemented as a firmware configured integrated circuit, e.g., an FPGA, which may be part of the same device or separate devices that may be positioned on the same motherboard, different PCIe cards in the same device, separate devices in the same facility, and/or located in different facilities. Accordingly, the one or more processing units and/or integrated circuits may be directly coupled together, e.g., tightly coupled, such as physically incorporated into the same motherboard, or separate motherboards positioned in the same housing and/or otherwise coupled together, or they may be positioned on separate motherboards or pCIE cards that are capable of communicating with each other remotely, such as wirelessly and/or via a network interface, such as via a local cloud 30, and in various embodiments the one or more processing units and/or integrated circuits may be positioned geographically remote from each other but communicable via a hybrid cloud 50. In particular cases, the integrated circuit(s) that form or are a part of the CPU, GPU, and/or QPU, which integrated circuit(s) may be arranged as and/or part of the secondary and/or tertiary analytics platform, may be configured to form one or more analytics segments in which the various data generated may be fed into and out of, back and forth between, the various processing units and/or integrated circuits, such as in an uninterruptible and/or continuous transmission manner, so as to allow rapid transmission of data between the plurality of integrated circuits, and more particularly to expedite the analyses herein.

[0772] For example, in some instances, the various devices for use in accordance with the methods disclosed herein may include, or otherwise be associated with, one or more sequencing devices, to perform a sequencing protocol, which sequencing protocol may be performed by software running on a remote sequencer, such as by a Next Gen sequencer, e.g., Illumina's HiSeq Ten, located at a core sequencing facility, as made accessible via a cloud-based interface. In other instances, sequencing may be performed in a wired configuration running on a sequencing chip, as deployed by Thermo Fisher's Ion Torrent, or other sequencer chip technologies, wherein sequencing is performed by use of a semiconductor technology that delivers benchtop next generation sequencing, and/or by an integrated circuit configured as, or to otherwise include, a field effect transistor employing a graphene channel layer. In such cases, where sequencing is performed by one or more integrated circuits configured as, or to include, a semiconductor sequencing microchip, the chip(s) may be positioned remotely from one or more other processing units and/or integrated circuits disclosed herein, which may be configured to perform secondary and/or tertiary analysis on the sequenced data. Alternatively, the chips and/or processing units may be positioned relatively close to each other so as to be directly coupled together, or at least within the same general proximity to each other, such as in the same facility. In this and other such cases, a sequencing and/or analytical BioIT array may be formed such that the raw sequencing data generated by the sequencer may be rapidly communicated, e.g., streamed in a continuous manner, to the other analytical components of the array for direct analysis, such as in a continuous manner.

[0773] Additionally, once raw sequencing data (e.g., BCL data) or read data (e.g., FASTQ data) are produced by the sequencing instrument, such data may be transmitted to, and received by, an integrated circuit configured to perform various bioanalytical functions on genetic and/or protein sequences, such as in connection with the analysis of generated and/or received DNA, RNA, and/or protein sequence data. Such sequence analysis may involve comparing a generated or received nucleic acid or protein sequence to one or more databases of known sequences, such as to perform secondary analysis on the received data, and/or in some cases, to perform disease diagnosis, such as when the database of known sequences to perform the comparison may be a database containing morphologically distinct and/or negligible sequence data, which is data from genetic samples that pertain to or are believed to pertain to one or more disease states.

[0774] Accordingly, in various instances, once isolated and sequenced, the genetic data, e.g., DNA and/or RNA, may be subjected to secondary analysis, which may be performed on the received data, such as to perform mapping, alignment, selection, variant calling, and/or the like, so as to generate mapped and/or aligned data that may then be used to derive one or more VCFs detailing the difference between the mapped and/or aligned genetic sequence and the reference sequence. In particular, once secondary processing has occurred, the genetic information may then be passed to one or more tertiary processing modules of the system, such as for further processing thereof, so as to derive therapeutically and/or prophylactically effective results. More particularly, after variant calling, the mapper/aligner/variant caller may output a standard VCF file that is ready for and can be communicated to an additional integrated circuit to perform tertiary analysis, such as genome-related analyses, e.g., whole genome, analysis, genotyping, e.g., joint genotyping, analysis, microarray analysis, exome analysis, microbiome analysis, an epigenome analysis, a metagenome analysis, a joint genotyping analysis, an analysis of variance, e.g., a GATK analysis, structural variant analysis, somatic variant analysis, and the like, as well as an RNA sequencing analysis or other genomic analysis.

[0775] Thus, the bioanalytical platform, e.g., BioIT, presented herein may include highly optimized algorithms for mapping, alignment, selection, duplicate tagging, haplotype variant calling, compression, and/or decompression, such as in a software, wired, and/or quantum processing configuration. For example, while one or more of these functions may be configured to be performed entirely or partially in a wired configuration, in particular cases, the secondary and/or tertiary processing platform may be configured to execute one or more software and/or quantum processing applications, such as one or more programs directed to performing one or more bioanalytical functions, such as one or more of the functions disclosed below herein. In particular, the sequenced and/or mapped and/or aligned and/or other processed data may then be further processed by one or more other highly optimized algorithms for one or more of whole genome analysis, genotyping analysis, microarray analysis, exome analysis, microbiome analysis, epigenome analysis, metagenome analysis, joint genotyping, and/or a variant, e.g., GATK analysis, as implemented by software running on a general purpose CPU and/or GPU and/or QPU, although in certain cases one or more of these functions may be at least partially implemented in hardware.

[0776] Accordingly, as can be seen with reference to Figure 43 , in various embodiments, the multiplexed bioanalytical processing platforms are configured to perform one or more of primary, secondary, and/or tertiary processing. For example, the primary processing stage produces genetic sequence data, such as in one or more BCL and/or FASTQ files for transfer into system 1. Once in system 1, the sequenced genetic data, including any associated metadata, may be advanced to a secondary processing stage 600 to produce one or more variant call files. Accordingly, the system may also be configured to take the one or more variant call files along with any associated metadata, and/or other associated processed data, and in one or more tertiary processing stages, may perform one or more other operations thereon, such as for the purposes of performing one or more diagnostic and/or prophylactic and/or therapeutic procedures thereon.

[0777] In particular, an analysis of the data may be initiated, for example, in response to a user request 120, e.g., made from a remote computing resource 100, and/or in response to data submitted by third parties 121, and/or data automatically retrieved from a local 200 and/or remote storage facility 400. Such further processing may include a first processing tier, wherein a plurality of segmentation execution protocols 700 are configured to perform analytics on the determined genetic data, e.g., variation, of one or more subjects. For example, a first tier of tertiary processing units may include a genomic processing platform that is configured to perform genome, epigenome, metagenome, genotyping, and/or various variant analysis, and/or other bioinformatics-based analysis. Additionally, in a second tertiary processing layer, various disease diagnostic, research, and/or analysis protocols 800 may be performed, which analysis may include one or more of NIPT, NICU, cancer, LDT, biological, AgBio applications, and the like.

[0778] System 1 may be further adapted to receive and/or transmit various data 900 related to the procedures and processes disclosed herein, such as related to electronic medical record (EMR) data, Federal Drug Administration test and/or structuring data, annotation-relevant data, and the like. Such data may be useful in enabling a user to make and/or allow access to generated medical, therapeutic, and/or prophylactic modalities developed through use of system 1 and/or made accessible thereby. Accordingly, in various instances, the devices, methods, and systems disclosed herein enable the secure performance of genetic and bioanalytical analysis, as well as the secure transfer of the results thereof, in a forum that may be readily usable for downstream processing. Additionally, in various instances, the devices, methods, and systems disclosed herein enable the secure transmission of data into the system, such as from one or more health monitoring and/or data storage facilities and/or a government agency, such as the FDA or NIH. For example, the system may be configured to securely receive EMR/PHR data, as may be transmitted from a health care and/or storage facility for use in accordance with the methods disclosed herein, such as for the performance of genetic and bioanalytical analysis, as well as for the secure transfer of the results thereof, in a forum that may be readily usable for downstream processing.

[0779] In particular, the first tertiary processing layer 700 may include one or more genomic processing platforms, such as for performing genetic analysis, such as on mapped and/or aligned data, e.g., in a SAM or BAM file format, and/or for processing variant data, such as in a VCF format. For example, the first tertiary processing platform may include one or more of a genome segmentation, an epigenome segmentation, a metagenome segmentation, a joint genotyping segmentation, as well as one or more variant analysis segmentations, including: a GATK segmentation, a structural variant segmentation, a somatic variant calling segmentation, and in some cases may include an RNA sequencing analysis segmentation. One or more other genomic analysis segmentations may also be included.

[0780] More specifically, with reference to Figure 43 , in various instances, the multi-tier and/or multiplexed bioanalytical processing platform includes an additional layer of data generation and/or processing units. For example, in certain instances, the bioanalytical processing platform incorporates one or more processing threads, in one or more of software and/or hardware implementations, that are directed to performing one or more tertiary processing protocols. For example, in particular instances, a tertiary processing thread platform 700 may include one or more of a genome thread, an epigenome thread, a metagenome thread, a joint genotyping thread, a variance thread, such as a GATK thread, and/or other threads, such as an RNA thread. Additionally, a second tier of the tertiary processing analysis platform may include multiple processing threads, such as one or more of a microarray analysis thread, a genome analysis thread, e.g., whole genome, genotyping analysis thread, exome analysis thread, epigenome analysis thread, metagenome analysis thread, microbiome analysis thread, genotyping analysis thread including joint genotyping, variant analysis thread including structural variant threads, somatic variant threads, and GATK and/or MuTect2 threads, as well as RNA sequencing threads and other genetic analysis threads.

[0781] Accordingly, in one embodiment, the multi-layer bioanalytical processing platform includes a metagenomic slicer. For example, a metagenomic slicer may be included, such as for performing one or more environmental genomic processes. Particularly, in various embodiments, the metagenomic analysis may be configured to determine whether a group of organisms evolved from a common ancestor, such as a species or other clade. More particularly, in various embodiments, an environmental sample containing a multiplicity of living and/or dead organisms therein may be obtained, from which the DNA/RNA present may be isolated, sequenced, and processed by one or more of the processing platforms herein, so as to identify the particular species present and/or one or more other genomic factors relevant thereto. Such "environmental" samples may include a multitude of human microbiomes (e.g., related to microorganisms that are found in association with both healthy and diseased humans, including microorganisms found in skin, blood, saliva, and feces samples) as well as external environmental agents.

[0782] There are a variety of methods for deriving sequenced genetic samples for mutagenomic processing. A primary method includes a targeted 16S ribosomal RNA cloning protocol and/or gene sequencing. For example, 16S ribosomal RNA is highly variable across species (or even strains of a species). Accordingly, such RNA can be isolated and sequenced to produce a genetic biodiversity profile that is derived from naturally occurring biological samples, which can be used to inform A/I or other system databases. However, a problem with such sequencing is that a large amount of microbial biodiversity may be missing simply because of the manner in which it was cultured.

[0783] Accordingly, a second method includes a targeted shotgun and/or PCR protocol that can be used to generate samples of a plurality, e.g., all genes of all biological agents in the sampled communities, which once sequenced can reveal the genetic diversity of microscopic life. Specifically, in the shotgun sequencing method, an aggregated reference sequence can be generated, e.g., from many (e.g., tens of thousands) of reference genomes from different species. However, the aggregate size of this number of reference genomes is large. Therefore, it is advantageous to select one or more distinct subsequences from each reference genome in order to construct the aggregated reference sequence.

[0784] For example, such a subsequence may be in the range of several hundred bases to several thousand bases in length, which are ideally unique sequences that do not occur in other species (or strains). These subsequences can then be aggregated to construct reference sequences. Consequently, once isolated, sequenced, mapped and aligned, these metagenomic sequences can be compared to partial or full reference genomes for many species, and genetic biodiversity can be determined.

[0785] Thus, metagenomics offers a powerful lens for viewing the microbial world that can revolutionize our understanding of the living world. Consequently, in any such case where there is a significant presence of an organism's DNA present in a sample, that species can be identified as being in that environment. Ideally, in this manner, species not common to other species generally present in that environment can be identified. Specifically, when the coverage of all species is normalized to the environmental samples obtained, the genetic diversity of all species present can be determined and can be compared to the entire coverage, such as by comparing a portion of a particular organism's DNA to that of the biologically diverse reference gene sequence generated.

[0786] The significance of these analyses can be determined by Bayesian methods, such as estimating the probability of observing sequenced reads from a particular organism, assuming that a given species is or is not present. Bayesian probability methods are directed at describing the probability of an event, based on the conditions that may be related to that event. For example, if one is interested in determining the presence of cancer in a subject, and if the subject's age is unknown, and if it is determined that cancer is an age-related disease, then, using Bayes' theorem, information about the subject's age can be used to more accurately assess the probability of cancer.

[0787] Specifically, with the Bayesian probability interpretation, the theorem expresses how a subjective degree of convection can rationally change to account for observed evidence. Bayes' theorem is stated mathematically as the following equation: P(A/B) = P(B/A) P(A) / P(B), where A and B are events and P(B) + 0. P(A) and P{B) are the probabilities of observing A and B unrelated to each other. P(A | B), a conditional probability, is the probability of observing event A given that B is true. P(B | A) is the probability of observing event B given that A is true.

[0788] Accordingly, one or more steps of performing a Bayesian probability analysis in this context may include one or more of: Presence calls may be made for clades at multiple taxonomic levels: kingdom, phylum, class, order, family, genus, species, and/or strain. However, this is complicated by the fact that DNA tends to be increasingly similar between organisms that share lower taxonomic levels. Additionally, often a sample may match a reference genome from multiple species at a higher taxonomic level (or multiple strains of a species), and so in many cases only a more general clade (such as a genus or family) may be unambiguously called present, rather than a specific species or strain. Regardless, the devices, systems, and methods for using the same disclosed herein may be employed to overcome these and other such difficulties.

[0789] Specifically, in one embodiment, a method for determining the presence of two or more species or clades of organisms from a sample is provided. For example, in a first step, genomic sequence data reads may be obtained from a sample, such as when the reads may be in a FASTQ or BCL format. Genomic sequence mapping may be performed in order to map the reads to multiple genomic reference sequences. In this case, the genomic reference sequences may be an entire genome, or may be a partial genome in order to reduce the amount of data required for each species, strain, or clade. However, using larger portions of a genome will increase detection sensitivity, and each reference sequence used should be selected to represent each species, strain, or clade that will be distinct from one another.

[0790] For this purpose, all or a portion of the genomic sequence from the 16S ribosome of each species or clade can be used. In this manner, two or more genomic reference sequences from species, strains, or clades of organisms suspected of being in the sample can be constructed in order to detect members of these groups in the sample. Once constructed, an index for each of the genomic reference sequences can also be constructed. The indexes can be a hash table or a tree index, with a prefix or suffix tree index. Once the index has been constructed, the sample genomic sequence reads can be compared to each of the two or more indices. It can then be determined whether the sample genomic sequence reads map to each of the indices.

[0791] Similarly, genomic sequence reads may also be aligned to the genomic reference sequence (or genomic reference sequences) to which they are mapped. This will generate an alignment score, in accordance with the methods herein, which may be used in analyzing the likelihood that a read indicates the presence or absence of a species or clade of organism in the sample. Specifically, mapping and/or alignment may be achieved by the present software and/or hardware modules as described herein. In some embodiments, the mapped and aligned data may then be communicated to the computing resource 100/300 for further analysis and processing.

[0792] For example, mapped and/or aligned genomic sequence reads can be analyzed to determine the propensity that an organism having the genomic reference sequence is present in the sample. Similarly, a list of species, strains, or clades that are determined to be present in the environmental sample can be reported. In certain embodiments, the list can be reported with a confidence metric (e.g., P-value) to indicate the statistical confidence of the assessment. The entire list of species, strains, or clades of organisms analyzed can also be reported, along with an indication of which species, strains, or clades were present, and a confidence metric. It should be noted that, although described in relation to microbiome analysis, several of the techniques and procedures disclosed herein can be employed in the analysis of all other tertiary processing protocols, where appropriate.

[0793] For example, Figure 43B presents an exemplary implementation of a method for performing environmental analysis, such as of microbiomes in an environmental sample. For example, in a first case, an environmental sample may be obtained, and the various genetic materials may be isolated therefrom. The various genetic materials may then be processed and sequenced, such as by means of a suitably configured NGS.

[0794] Accordingly, in a first step 1000, once several genetic materials have been sequenced, for example by an NGS, they can be transmitted to the system 1 disclosed herein. In step 1010, one, two or more genomic reference sequences of interest, for example to be detected in the sample, can be constructed. In step 1020, an index for each of the one, two or more genomic reference sequences can be constructed. Additionally, in step 1030, the sequenced reads obtained from the genomic sample can then be compared to the one, two or more indices, such as by means of a suitably configured mapping module. In step 1040, it can then be determined whether the genomic sample's sequenced reads map to each of the two or more indices.

[0795] At this point, if desired, in step 1050, the mapped reads can be aligned with the genomic reference sequences to generate an alignment and/or an alignment score. Accordingly, once the genetic materials obtained in the sample are mapped and/or aligned, in step 1060, the probability of a given organism having the reference sequence being present in the sample can be determined. Once processed, a list of species, strains, and/or clades that are present in the sample can be identified and/or reported.

[0796] The tertiary processing platform disclosed herein may also include epigenomic targeting. In particular, epigenetics studies the non-coding genetic effects on the DNA sequence of an organism. The term also refers to the changes themselves: the functionally relevant changes to the genome that do not involve a change in the nucleotide sequence. Independently, epigenetic changes are stably heritable genotypes that result from changes in a chromosome that do not alter the DNA sequence. These changes may or may not be heritable. In particular, epigenetic changes modify the activation of certain genes, but not the genetic coding sequence of DNA. The same is true of the microstructure (non-coding) of DNA itself or the associated chromatin proteins may be modified, causing activation or silencing.

[0797] The epigenome is involved in the regulation of gene expression, development, tissue differentiation, and suppression of transposable elements. Unlike the underlying genome, which is largely static within an individual, the epigenome can be dynamically altered by environmental conditions. The field is analogous to genomics and proteomics, which are the study of the genome and proteome of a cell. Additionally, epigenomics involves the study of the complete set of epigenetic modifications to a cell's genetic material, known as the epigenome, which consists of a record of chemical changes to an organism's DNA and histone proteins. These changes can be passed on to the organism's offspring through transgenerational epigenetic inheritance. Changes to the epigenome can result in changes to chromatin structure and changes to genome function.

[0798] This epigenetic mechanism enables differentiated cells as a multicellular organism to express only those genes that are necessary for their own activity. Epigenetic changes are preserved when cells divide. In particular, most epigenetic changes only occur over the course of an individual organism's lifetime. However, if gene inactivation occurs in a sperm or egg cell that results in fertilization, then some epigenetic changes may be transferred to the next generation. Several types of epigenetic inheritance systems may play a role in what has become known as cellular memory. For example, various covalent modifications of DNA (e.g., cytosine hydroxymethylation and methylation) or of histone proteins (e.g., lysine acetylation, lysine and arginine methylation, serine and threonine phosphorylation, and lysine ubiquitination and sumoylation) may play central roles in many types of epigenetic inheritance. Because the phenotype of a cell or individual is affected by which of its genes are transcribed, heritable transcriptional states can give rise to epigenetic effects. Such effects on cellular and physiological phenotypic traits can result from external or environmental factors that turn genes on and off and affect how cells express genes.

[0799] For example, DNA damage can cause epigenetic changes. DNA damage is very common. Such damage is largely repaired, but at the site of DNA repair, epigenetic changes may remain. In particular, a double-strand break in DNA can initiate unscheduled epigenetic gene silencing by either causing DNA methylation or by silencing types of histone modifications (chromatin remodeling). Other examples of mechanisms that produce such changes are DNA methylation and histone modification, each of which alters how genes are expressed without altering the underlying DNA sequence. Nucleosome remodeling has also been shown to cause epigenetic silencing of DNA repair. Additionally, chemicals that damage DNA can also cause substantial DNA hypomethylation, such as through activation of oxidative stress pathways. Additionally, gene expression can be controlled through the action of repressor proteins that bind to silenced regions of DNA.

[0800] These epigenetic changes can last through cell divisions for the duration of the cell's life, and can also last for multiple generations, even though they do not involve changes in the organism's underlying DNA sequence; rather, non-genetic factors cause the organism's genes to behave (or "express") differently. An example of an epigenetic change in eukaryotic biology is the process of cell differentiation. During morphogenesis, totipotent stem cells become the various pluripotent cell lineages of the embryo, which in turn become fully differentiated cells. In other words, as a single fertilized egg cell—the zygote—continues to divide, all of the resulting daughter cells change into the different cell types in an organism, including neurons, muscle cells, epithelium, blood vessel endothelium, etc., activating some genes while inhibiting the expression of others.

[0801] There are several layers of regulation of gene expression. One way in which genes are regulated is through chromatin remodeling. Chromatin is the complex of DNA and the histone proteins with which it associates. If the way that DNA is wrapped around histones changes, gene expression can also change. A first way is through translational modification of the amino acids that make up the histone proteins. Histone proteins are made of long chains of amino acids. If the amino acids in the chain are changed, the shape of the histone can be changed. DNA is not completely unfolded during replication. It is possible, then, that modified histones could be loaded into each new copy of DNA. Once there, these histones could act as templates, initiating the surrounding new histones to be folded in the new way. By changing the shape of the histones around them, these modified histones would ensure that a lineage-specific transcriptional program is maintained after cell division.

[0802] The second pathway is the addition of methyl groups to DNA, primarily at CpG sites, to convert cytosine to 5-methylcytosine. 5-Methylcytosine performs similarly to regular cytosine, pairing with a guanine in double-stranded DNA. However, some areas of the genome are more heavily methylated than others, and highly methylated areas tend to be less transcriptionally active, through a mechanism that is not fully understood. Cytosine methylation can also persist from the germline of one parent into the zygote, marking the chromosome as inherited from one parent or the other (genetic imprinting). Although histone modifications occur throughout the entire sequence, the N-terminal nonstructures of histones (called histone tails) are particularly highly modified. These modifications include acetylation, methylation, ubiquitylation, phosphorylation, sumoylation, ribosylation, and citrullination.

[0803] Accordingly, DNA methylation is the presence of methyl groups on some DNA nucleotides, especially 'C' bases followed by 'G' dinucleotides, or "CpG". Methylation in promoter regions tends to suppress gene expression. Methylation analysis is the process of detecting which 'C' bases are methylated in a given sample genome. Bisulfite sequencing (MethylC-seq) is the most common method for detecting methylation using whole-genome sequencing, in which unmethylated cytosine ('C') bases are chemically converted to uracil ('U') bases, which become thymine ('T') bases upon PCR amplification. Methylated 'C' bases resist conversion.

[0804] Accordingly, in accordance with the devices and methods disclosed herein, detection of modifications of DNA molecules, wherein the modifications do not affect the DNA sequence but do affect gene expression, are provided herein, as by performing one or more mapping and/or alignment operations on epigenetic genetic material. In such methods, the obtained reads may be mapped and aligned to the reference genome in a manner that allows the converted 'T' bases to align with the reference 'C' positions, and the 'C bases may be replaced by 'T's in the reference sequence, prior to mapping/alignment. This allows for the imprecise mapping and alignment of reads which have bisulfite converted C's (now T's), thereby revealing the non-bisulfite converted (methylated) C's in the genomic sequence reads. For reverse complemented alignments, complementary substitutions can be used, for example, 'G's can be replaced by 'A's.

[0805] Similarly, the reference index builder (e.g., hash table) and mapper/aligner can be modified to perform these replacements automatically for MethylC-seq use. Alternatively, the mapper/aligner can be modified to allow direct alignment of read 'T's to reference 'C's, and reverse complemented alignment of read 'A's to reference 'G's. The methods disclosed herein improve accuracy and prevent erroneous direct alignment of read 'C's to reference 'T's, or erroneous reverse complemented alignment of read 'G's to reference 'A's.

[0806] Additionally, provided herein are methods for determining the methylation status of cytosine bases in genomic sequence reads. For example, in a first step, genomic sequence reads from bisulfite-treated nucleotide samples may be obtained. In particular, one or more modified sequencing protocols may be employed in order to generate the reads for secondary processing, in these regards. Specifically, one or more of: whole genome bisulfate sequencing; reduced representation bisulfate sequencing; methylated DNA immunoprecipitation sequencing; and methylation-sensitive restriction enzyme sequencing may be used to identify DNA methylation across portions of the genome, at varying levels of resolution down to a base pair level. Additionally, chromatin accessibility may be assessed, for example, where DNase I hypersensitive site sequencing may be performed, such as when the DNase I enzyme may be used to find open or accessible regions in the genome. Additionally, RNA sequencing and expression arrays can be used to identify expression levels of protein-coding genes. In particular, smRNA sequencing can be used to identify the expression of small non-coding RNAs, primarily miRNAs.

[0807] Accordingly, once sequenced to produce reads, a genomic reference sequence can be constructed for comparison to the reads. CpG locations in the genomic reference sequence can then be marked. Additionally, the genomic reference sequence can be pre-processed by replacing C's in genomic with T's. An index to the genomic reference sequence can be constructed. And once the index has been constructed, sample genomic sequence reads can be compared to the index, and it can be determined whether sample epigenomic sequence reads map to the index.

[0808] Additionally, mapped reads can be aligned to the genomic reference sequence to generate an alignment score. In certain embodiments, base substitutions can be made in the read sequence, and the read can be re-compared and realigned to the index. In some embodiments, an alignment orientation constraint can be used during mapping and/or alignment of a read, such that only forward alignments can be allowed with C to T substitutions in the read and genomic sequence reference, and only reverse complement alignments are allowed with G to A substitutions in the read and genomic sequence reference.

[0809] These mapping and alignment procedures can be achieved by the various software and/or hardware modules described herein. In some embodiments, the mapped and aligned data can then be communicated to a CPU/GPU/QPU for further analysis and processing. For example, mapped and aligned reads can be sorted by their mapped reference position. In some embodiments, duplicate reads can be flagged and removed. Overlapping reads from a backlog of reads can be analyzed over each flagged reference CpG location. In such a case, a thymine (T) that has replaced a cytosine (C) indicates an unmethylated cytosine and is flagged as such. And a cytosine that remains in the read sequence can be flagged as a methylated cytosine. Reverse complemented alignments of CpG locations can also be flagged as methylated or unmethylated. For example, a guanine (G) that has replaced an adenine (A) is marked as the reverse complement of an unmethylated cytosine (C), while a guanine (G) that remains in the reading frame is marked as the reverse complement of a methylated cytosine (C). The likely methylation status of each CpG location on each nucleoside strand can be reported, and an associated confidence metric (e.g., p-value) in the methylation call can be made. In some embodiments, the methylation status of the marked CpG locations can also be indicated for each chromosome of a diploid pair of chromosomes.

[0810] With respect to histone modification, histone modification includes various naturally occurring chemical modifications of the histone proteins upon which DNA is folded, resulting in more or less tightly folded DNA. Loosely folded DNA, for example, is associated with higher rates of gene expression. Such histone modifications can be determined by Chromatin Immunoprecipitation Sequencing (ChlP-Seq), which can be used to identify genome-wide patterns of histone modifications, such as by using antibodies against the modifications. Additionally, ChlP-seq is a method that can be employed to isolate and sequence DNA that is tightly bound to histones (or other selected proteins). After ChlP-seq has been performed, the sample can be prepared, DNA isolated and sequenced, and the sequenced DNA can then be mapped/aligned to a reference genome as disclosed herein, and the mapped coverage can be used to infer the level of histone binding at various loci in the genome. Further provided herein are methods for analyzing ChlP-derived nucleotide sequences, which are similar to the methods described below for analyzing structural variants.

[0811] It is noteworthy that epigenetics is useful in cancer research and diagnosis. For example, human tumors undergo a major disruption of DNA methylation and histone modification patterns. Indeed, the aberrant epigenetic landscape of the cancer cell is characterized by global genomic hypomethylation, CpG island promoter hypermethylation of tumor suppressor genes, an altered histone code for critical genes, and a global loss of monoacetylated and trimethylated histone H4. Accordingly, the methods disclosed herein can be used for cancer research and/or diagnostic purposes.

[0812] Additionally, the methods disclosed herein may be useful for generating one or more epigenomic and/or genome reference databases. For example, the methods disclosed herein, e.g., employing an A/I learning protocol of the system, may be useful for generating a human reference of epigenomes, such as from normal healthy individuals across a wide variety of cell lines, primary cells, and/or primary tissues. Such data produced may then be used to enhance the mapping and/or alignment protocols disclosed herein. Additionally, once a database of epigenomic differences has been generated, the database may be mined, e.g., by the A/I module in order to better characterize and determine relevant factors occurring in various disease states, such as cancer, dementia, Alzheimer's disease, and other neurological conditions.

[0813] Accordingly, in various instances, an epigenomic analysis may be performed to identify one or more or the entire set of epigenetic modifications that have occurred in the genetic material of a cell. In particular, using the methods disclosed herein, the epigenome of an organism, and/or the cells thereof, may be determined to catalog and/or record the chemical changes to the DNA and histone proteins of the cells of the organism. For example, an exemplary epigenomic analysis is presented herein in Figure 43C.

[0814] For example, in a first step, a genomic sample may be obtained from an organism, and genetic material isolated therefrom and sequenced. Then, once sequenced, in step 1000, sequenced reads from the sample may be transmitted to and received by system 1. In this case, the reads may be derived from a bisulfate-treated nucleotide sample. Similarly, in step 1010, a genomic reference sequence, e.g., for the organism, may be constructed as to perform a comparison of the epigenomic sample reads. In step 1012, any multiple CpG locations in the genomic reference sequence (or genomic reference sequences) may be identified.

[0815] Once identified, at 1014, the "C's" of the CpG locations in the reference may be replaced with "Ts", and at step 1020, an index for the modified genomic reference sequence may be generated. Once the index for the modified reference is generated, at step 1030, the sample genomic sequence reads may be compared to the index, and at step 1040 it may be determined whether the sample genomic sequence reads map to the index, as when mapped in accordance with the methods and apparatus disclosed herein. The mapped reads may then be aligned to the genomic reference sequence, and an alignment score may be generated, as by performing one or more alignment operations as discussed herein.

[0816] At this point, one of a couple of several analyses may be performed. For example, in step 1051, if greater context is desired, the base substitutions in the reads, as processed above, and/or the alignment orientation, and/or parameter constraints may be adjusted, and comparison steps 1030-1050 may be repeated. This process itself may be repeated as desired until a sufficient level of context is achieved. Accordingly, once a sufficient level of context is achieved, the mapped and/or aligned reads, in step 1080, may be sorted, as in the processes disclosed herein, by the mapped/aligned reference position. And in step 1081, any duplicate reads may be flagged and/or removed.

[0817] Additionally, in step 1082, reads from the backlog of reads overlapping each marked reference CpG location may be analyzed. When a "T" is replaced by a "C", it may be marked as an unmethylated "C" in step 1083; and when a "C" remains in the sequence, in step 1084, the "C" may be marked as a methylated "C". Finally, in step 1086, a determination and/or report on the likely methylation status of each of the CpG locations on each nucleotide strand, and a confidence in the methylation call, may also be made.

[0818] Additionally, provided herein are methods for analyzing genomic material in which part of the genetic material may have, or may otherwise be associated with, a structural variant. In particular, a structural variation is a variation in the structure of an organism's chromosome. Structural variations involve many types of variations in the genome of a species, including microscopic and submicroscopic types such as deletions, duplications, copy number variants, insertions, inversions, and translocations. Many structural variants are associated with genetic diseases. In fact, about 13% of the human genome is defined as structurally variant in the normal population, and there are at least 240 genes that exist as homozygous deletion polymorphisms in human populations. Such structural variations can comprise millions of nucleotides of heterogeneity across genomes, and are likely to make an important contribution to human disease susceptibility.

[0819] Copy number variation is a broad category of structural variation, which includes insertions, deletions, and duplications. There are several known inversions that have been implicated in human disease. For example, a recurrent 400kb inversion in the factor VIII gene is a common cause of hemophilia A, and smaller inversions affecting idunorate 2-sulfatase will cause Hunter syndrome. Further examples include Angelman syndrome and Sotos syndrome. The most common type of complex structural variation is non-tandem duplications, in which the sequence is duplicated and inserted in an inverted or forward orientation elsewhere in the genome. Other classes of complex structural variant include deletion-inversion-deletions, duplication-inversion-duplications, and tandem duplications with internal deletions. There are also cryptic translocations and uniparental segmental disomy (UPD).

[0820] However, detection of abnormal DNA structures is problematic and beyond the scope of previously known variant calling. Such structural variants that are problematic to detect include those that have: large insertions and deletions (e.g., beyond the 50 to 100 bp indel size); duplications and other copy number variations (CNVs); inversions and translocations, and aneuploidy (abnormal chromosome copy counts: monosomy, disomy, trisomy, etc.). In certain cases disclosed herein, the identified copy number variations can be tested in subjects who do not have genetic diseases, such as using quantitative SNP genotyping.

[0821] Detection of structural variation generally begins with performing a mapping and alignment operation, such as using the devices and methods disclosed herein. For example, reads from the genomic sample to be analyzed may be mapped and aligned to a reference genome, such as in a protocol that supports chimeric alignments. Specifically, some structural variants (e.g., CNVs and aneuploidy) may be detected by analysis of relative mapped coverage. However, other structural variants (e.g., large indels, inversions, translocations) may be detected by analysis of trimmed and chimeric alignments.

[0822] Specifically, each structural variant involves one or more "break" positions, where the read does not map to the reference genome, such as when the geometry changes between the sample and the reference. In such a case, the stack may be configured such that reads therein that slightly overlap structural variant breaks may be trimmed at the break, and reads that slightly overlap structural variant breaks may be chemically aligned, e.g., with two portions of a read mapping to different reference locations. However, read pairs that overlap structural variant breaks may be inconsistently aligned, with the two corresponding reads mapping to widely different reference locations, and/or with abnormal relative orientation of corresponding reads. Such obstacles may be overcome by the methods disclosed herein.

[0823] For example, in certain cases, data pertaining to known structural variants may be used to better determine the sequence of a structural variant. For example, a database listing the structural variations in the human genome may be compiled, e.g., with an emphasis on CNVs, and such data may be used in determining the sequence of particular variants, as in a suitably configured weighting protocol. In particular, when a structural variant is known, its "inside" and "outside" coordinates may be employed as a minimum and maximum range of sequence that may be affected by the structural variation. Additionally, insertion, loss, gain, inversion, LOH, everted, transchr, and UPD variations may be classified and provided to the knowledge base of the present system.

[0824] In various cases, determination of a structural variant may be performed by a suitably configured CPU/GPU/QPU software execution, such as by employing previously determined sequencing data, and in other cases, structural variant analyses may be performed as on the hardware disclosed herein. Accordingly, in particular cases, a method for analyzing genomic sequences for structural variants is provided. For example, in a first step, genomic sequence reads may be received from a nucleotide sample. In certain cases, the sequenced reads may have been derived from paired-end or matching-pair protocols to detect structural variants. Next, an index to the genomic reference sequence may be constructed, such as when the index may be a hash table or a tree, such as a prefix or suffix tree. Once the index has been constructed, sample genomic sequence reads may be compared to the index in order to determine whether the sample genomic sequence reads map to the index. In this case, sample genomic sequence reads can then be aligned to the genomic reference sequence to which they map, and an alignment score can be determined.

[0825] As indicated above, mapping and alignment may be achieved by the hardware module as described herein. In some embodiments, the mapped and aligned data may then be communicated to an associated CPU/GPU/QPU for further analysis and processing. Reads may be sorted by mapped reference position, and duplicate reads may be flagged and deleted. Chimeric reads and/or unusual relative alignments of two matching reads may be determined, and possible structural variants may be determined based on a detected chimeric read and/or unusual relative alignments (e.g., large indel, an inversion, or a translocation). Similarly, the posterior probabilities of each possible structural variant may be calculated. In some embodiments, structural variant haplotypes may be determined, such as using HMM analysis of the chimeric reads and/or unusual relative alignments. For example, pairwise HMM may be used for such determination. Pairwise HMM can be achieved by using hardware module.

[0826] Accordingly, in various instances, as can be seen with respect to Figure 43D, a method for determining variations in the chromosome structure of an organism is provided. For example, in accordance with the methods disclosed herein, in step 1000, genomic sequence data reads may be received. In step 1010, one or more genomic reference sequences may be constructed, so as to perform a comparison between the reads and the reference sequence (or reference sequences). Specifically, in step 1010, a genomic reference sequence may be constructed so as to allow the received reads to be compared to the generated reference. More specifically, for these purposes, in step 1020, an index to the genomic reference sequence may be generated, e.g., in step 1020, a hash table or prefix/suffix tree may be generated. Then, in step 1030, the sample genomic sequence reads may be compared to the generated index, as per the software and/or hardware implementations disclosed herein.

[0827] If, in step 1040, it is determined that the sample genomic sequence reads map to the index, then in step 1050, the mapped reads can be aligned to the genomic reference sequence, and an alignment score can be generated. In step 1080, the sample reads can be ranked by their mapped reference positions. At that point, in step 1081, duplicate reads can be flagged and/or removed. Additionally, in step 1090, chimeric reads and/or unusual relative alignments, e.g., of two matched reads, can be detected, and in 1092 possible structural variants can be determined, based on the detected chimeric reads and/or unusual relative alignments. Additionally, the posterior probabilities of each possible structural variant may be calculated and, optionally, in step 1096, the structural variant haplotypes may be determined, such as by using HMM analysis as described herein, of chimeric reads and/or unusual relative alignments.

[0828] Additionally, the devices, systems, and methods disclosed herein may be employed for processing RNA sequences. In particular, herein, methods for analyzing RNA sequence reads are disclosed, such as employing a spliced mapping and alignment protocol (e.g., with a suitably configured RNA mapper/aligner). For example, in one embodiment, a transcriptome segmentation may be provided, such as for ultra-fast RNA sequence data analysis. In particular, such segmentation may be configured to perform secondary analysis on RNA transcripts, such as with respect to reference-only alignment, as well as annotation-aided alignment.

[0829] Accordingly, in a first method, raw read data, e.g. in a BCL and/or FASTQ file format, may be produced by a sequencing instrument and may be input to the system, where mapping, alignment, and variant calling may be performed. However, in various cases, one or more gene annotation files (GTF) may be input to the system, so as to guide the spliced alignments, e.g., a splice junction LUT may be constructed and used. For example, alignment and splice junction accuracy tables may be employed. Accordingly, a 2-phase alignment may be performed, such as where in a first detection phase novel splice junctions may be employed, which may then be used to guide a second mapping/pass alignment phase. After variant calling, the system will output a standard VCF file ready for tertiary analysis.

[0830] In particular, once an input file is received, spliced mapping and alignment may be performed, such as on either single or paired read ends. As indicated, configurable junction filters may be employed to generate a single junction output. Position selection may be performed, which may include discretization by reference range, and then selection of discretizations by reference position, and duplicate marking may occur, such as based on starting position and CIGAR strand so as to achieve a high quality duplicate report, wherein any duplicates may be removed. Haplotype variant calling may then be performed, for example using a SW and HMM processing engine, and assembly may be performed.

[0831] Additionally, the devices, systems, and methods disclosed herein may be employed to perform somatic variant calling. For example, a somatic variant calling protocol may be employed to detect variants that may occur in cancer cells. In particular, genomic samples for somatic calling may be obtained from single or multiple tumor biopsies, or from blood. Optionally, a "normal" (non-tumor) sample may also be obtained, such as for comparison during variant calling, e.g., where somatic variants will occur in tumor cells but not in cells from the normal sample. The DNA/RNA form of the sample (or samples) may be isolated and sequenced, such as by a Next Gen sequencer. The sequence data, e.g., from each sample, may then be transmitted to the secondary processing platform, and the reads may be mapped and aligned. Additionally, the reads may be subjected to a plurality of variant calling procedures, including processing by one or both of the SW engine and pairwise HMM.

[0832] However, the system should be configured so that it can detect low variant allele frequencies, such as 3% to 10% (or higher). More particularly, a genotyping probability model may be employed, wherein the model is configured to allow for arbitrary allele frequencies. One method to allow for this is to assign each variant genotype allele frequency corresponding to the allele frequencies observed in the overlapping reads. For example, if 10% of overlapping reads exhibit a certain variant, a genotype may be tested that consists of 90% reference allele and 10% alternative allele. For dual tumor/normal samples, the posterior probability that a variant is present in the tumor sample but not in the normal sample may be estimated.

[0833] Additionally, the somatic variant caller segmentation may be configured to provide information about tumor heterogeneity, e.g., that a series of distinct mutation events have occurred, such that one or more sections of a tumor with different genotypes (a subclone) have been identified. Such subclone information may be derived from a determination of variant allele frequencies and distributions thereof, and/or by explicitly calling variants differentially among multiple tumor samples.

[0834] Accordingly, methods for detecting cancer cell sequence variants from a sample are provided. In a first step, genomic sequence reads of a nucleotide sample may be obtained from cancer and/or normal cells. The sequence reads may be from paired-end or matched-pair protocols similar to that for detecting structural variants. An index to the genomic reference sequence may be constructed, such as where the index may be a hash table or a tree, such as a prefix or suffix tree. Sample genomic sequence reads, e.g., from the tumor and/or normal sample, may be compared to the index, and it may be determined whether the sample genomic sequence reads map to the index.

[0835] The sample genomic sequence reads can then be aligned to the genomic reference sequence to which they are mapped, and an alignment score can be generated. The mapping and alignment can be achieved by a software and/or hardware module as described herein. In some embodiments, the mapped and aligned data can then be communicated to a CPU/GPU/QPU for further analysis and processing. The reads can be sorted by mapped reference position, and any duplicate reads can be flagged and deleted. Variants can be detected using a Bayesian analysis that is modified to expect arbitrary variant allele frequencies, and to detect and report low possible allele frequencies (e.g., 3% to 10%).

[0836] In some embodiments, germline variants can be detected in both non-cancerous and cancerous samples, and somatic variants can be detected only in cancerous samples. For example, germline and somatic mutations can be distinguished by relative frequency. Posterior probabilities can be calculated for each possible cancer variant, and in some embodiments, structural variant haplotypes can be determined using HMM analysis of chimeric reads and/or unusual relative alignments. For example, pairwise HMM can be used for such determination. Pairwise HMM can be achieved using hardware modules as described herein.

[0837] Accordingly, in various embodiments, a somatic variant calling procedure as exemplified in Figure 43E may be performed to calculate the probability that a variant is a cancer variant. For example, in step 1000, reads from genomic sequence samples may be generated, e.g., via NGS sequencing, and/or received, e.g., via transmission over a suitably configured cloud-based network system, such as from one or both of the cancerous and non-cancerous genetic samples. In step 1010, a genomic reference sequence may be generated, such as for comparison of reads, in step 1020, an index may be constructed from the genomic reference sequence, and in step 1030, the sample genomic sequence may be compared to the index, such as by employing the software and/or hardware implementations disclosed herein, so as to map the genomic sequence reads to the index, in step 1040. Additionally, in step 1050, the mapped reads may be aligned to the genomic reference sequence to generate an alignment score. The mapped and/or aligned reads may then be ranked relative to the reference position, in 1080, and optionally, in 1081, any duplicate reads may be traded and removed.

[0838] Additionally, once the reads have been mapped and/or aligned and/or classified and/or de-induced, then, in step 1100, variants may be detected, such as by employing a Bayesian analysis, and in 1101, germline variants in both cancerous and non-cancerous samples, as well as somatic variants therein, may optionally be detected. Similarly, in step 1094, the posterior probabilities of each possible cancer variant may be calculated. Additionally, in step 1096, cancer variant haplotypes may optionally be determined, such as by implementing an HMM analysis in software and/or hardware, as disclosed herein.

[0839] Additionally, the devices, systems, and methods disclosed herein may be configured to perform a joint genotyping operation. In particular, a joint genotyping operation may be employed to improve the accuracy of variant calling, such as by jointly considering reads from a cohort of multiple subjects. For example, in many cases, genomic variations may be highly correlated, e.g., in certain populations, where certain variants are common to a plurality of subjects. In such cases, the sensitivity and specificity of variant calling may be improved by jointly considering the evidence for each variant from multiple DNA (or RNA) samples. Specifically, sensitivity may be improved because weak evidence for a variant in one subject may be enhanced by evidence for the same variant in other samples. More specifically, sensitivity may be improved because moderate evidence for a false positive variant may be tempered by lack of evidence for the same variant in other samples. Generally, the more samples participating in pooled genotyping, the more accurate variant calling can be for any given subject.

[0840] Ensemble genotyping involves estimating posterior probabilities for several subsets of all subjects who have a given variant, using prior probabilities that express the observed correlations in genetic variation. In many cases, ensemble genotyping can be performed in a single variant calling pass, in which aligned reads from multiple samples are examined by the variant caller. This is usually only practical for small numbers of samples, because when tens, hundreds, or thousands of samples are involved, the total data size becomes impractical to access and manipulate quickly.

[0841] Alternatively, joint genotyping can be done by first performing variant calling separately for each sample, then merging the results with a joint genotyping tool, which updates the variant probabilities for each subject using joint information. This method uses additional output from each single-sample variant calling pass to better measure areas of weak evidence for variants and/or regions where no variants would be called without joint processing. While the VCF format is commonly used to represent variant calls from single-sample variant calling, a special gVCF format can be used to represent first-stage variant (and non-variant) calls in preparation for merging. The gVCF format includes records for locations, and/or blocks of multiple locations, where no variants are most likely present, so this information can be fused with other gVCF or non-call calls at the same locations to produce improved joint genotype calls for each subject.

[0842] Accordingly, the joint genotyping segmentation can be configured to call variants from multiple samples faster and with greater accuracy. Additionally, the joint genotyping segmentation can be further configured to support pedigree as well as population variant calling from a cohort of samples. For example, the segmentation can be configured to handle up to 10, 15, 20, 25, even 50 or more samples at a time. In many cases, a population calling configuration can be adapted to handle sample sizes of many thousands at a time. Additionally, a combination of speed and hierarchical clustering of multiple samples provides a computationally efficient analysis solution for joint genotyping. Additionally, sequencing of samples for joint genotyping can be performed on the same flow cell of a Next Gen sequencer thus allowing the system to simultaneously map/align inputs from multiple samples thus speeding up the overall joint calling process, such as when BCL data can be fed directly to segmentation to produce unique gVCF files for each sample.

[0843] Therefore, a method for improving variant calling accuracy by jointly considering reads from a multi-subject cohort is provided herein. In a first step, genomic sequence reads from two or more samples are received. A genomic reference sequence for comparison with the reads is constructed, and from the genomic reference sequence, an index is generated. The genomic sequence reads from each sample are then compared to the index, and it is determined whether the genomic sequence reads from each sample map to the index.

[0844] The mapped reads can then be aligned to the genomic reference sequence and an alignment score can be generated. The reads can be sorted by mapped reference position, and duplicate reads can be flagged and/or removed. Additionally, overlapping reads from the read pool can then be analyzed to determine whether a majority of reads agree with the reference genomic sequence. The posterior probabilities of each possible variant are calculated, and the variant calling data from all samples can be merged in order to improve the variant calling accuracy for each individual sample. This can improve the variant calling accuracy (e.g., sensitivity and specificity) for each sample, and can be achieved as a processing step after all samples have been subjected to variant calling analysis, or can be achieved cumulatively after each sample has been subjected to variant calling analysis. The probability of non-reference alleles in regions where no variants are called can then be determined, and the determined probability of non-reference alleles in regions where no variants are called can be reported.

[0845] Accordingly, in various embodiments, a somatic variant calling procedure as exemplified in Figure 43F may be performed to calculate the probability that a variant is a cancer variant. For example, in step 1000, genomic sequence sample reads may be generated, e.g., via NGS sequencing, and/or received, e.g., via transmission over a suitably configured cloud-based network system, such as from one or both of the cancerous and non-cancerous genetic samples. In step 1010, a genomic reference sequence may be generated, such as for comparison of reads, in step 1020, an index may be constructed from the genomic reference sequence, and in step 1030, the sample genomic sequence may be compared to the index, such as by employing the software and/or hardware implementations disclosed herein, so as to map the genomic sequence reads to the index, in step 1040. Additionally, in step 1050, the mapped reads may be aligned to the genomic reference sequence to generate an alignment score. The mapped and/or aligned reads may then be ranked relative to the reference position, in 1080, and optionally, in 1081, any duplicate reads may be traded and removed.

[0846] Similarly, at 1082, overlapping reads from a pool of reads may be analyzed to determine whether one or more, e.g., a majority of the reads, agree with the reference genomic sequence (or reference genomic sequences), and at step 1094, the posterior probabilities of each possible variant may be calculated. At this point, at step 1096, variant haplotypes may be determined, if desired, such as by performing an HMM analysis, and/or at step 1120, variant calling data, e.g., from all samples, may optionally be merged so as to enhance variant calling accuracy for each individual sample. Additionally, at step 1122, the probability of non-reference alleles, e.g., in regions where no variants are called, may be determined and reported.

[0847] Additionally, as can be seen with reference to Figure 43 , in one aspect, an online app store is provided to allow users to develop, sell, and use genomic tools that can be incorporated into the system and employed to analyze genomic data transmitted to and input into the system. In particular, the genomic app store enables customers who wish to develop genetic tests, e.g., as a NICU test, and once developed can be uploaded to the system, e.g., genetic marketplace, to purchase and run as a platform therein, so that anyone running the newly developed system platform can install the uploaded tests via the web portal. More particularly, a user can browse the web portal app store, find a desired test, e.g., NICU test, download it, and/or configure the system to deploy the same, as their uploadable genetic data. The online “cohort” marketplace therefore presents a quick and efficient way to deploy new genetic analytical applications, which applications allow identical results to be obtained from any of the present system platforms running the downloaded application. More particularly, the online marketplace provides a mechanism for anyone working with the system to develop genetic analysis applications that remote users can download and configure for use in accordance with the present workflow models.

[0848] Another aspect of the cohort market disclosed herein is that which enables secure data sharing. For example, the transmission and storage of genomic data must be highly secure. However, such genetic data is often large and difficult to transfer in a secure and protected manner, such as when the identity of the subject is restricted. Accordingly, the present genetics market enables cohort participants to share genetic data without needing to identify the subject. In such a market, cohort participants can share questions and processes in order to advance their research in a protected and secure environment, without risking the identity of their respective subject genomes. Additionally, a user can enlist the assistance of other researchers in analyzing their sample sets without identifying who those genomes belong to.

[0849] For example, a user may identify subjects who have a specific genotype and/or phenotype, such as stage 3 breast cancer and/or who have been treated with a particular drug. A cohort may be formed to see how these drugs affect cancer cell growth at a genetic level. Thus, these characteristics, among others, may form cohort selection criteria that will allow other researchers, e.g. remotely located, to perform standard genetic analyses on genetic data, using uniform analytical procedures, on subjects who have access to this cohort criteria fit. In this way, a given researcher need not be responsible for identifying and securing all members of a sample set, e.g., subjects who fit the criteria, to substantiate his or her scientific query.

[0850] In particular, Researcher A may define a research cohort in the marketplace and identify appropriate selection criteria for subjects, the genomic test (or genomic tests) to be performed, and the parameters by which the testing should be performed. Researchers B and C, located remotely from Researcher A, may then register for the cohort, identify and select subjects who meet the criteria, and then perform the specified tests on their subjects using the uniform procedures disclosed herein, so as to assist Researcher A in achieving or better fulfilling his or her research objectives in a timely manner. This is beneficial since only a portion of the genetic data is transmitted, subject identity is protected, and since the data is analyzed using the same genetic analysis system employing the same parameters, the data results will be the same regardless of where and on which machine the test (or tests) is performed. Consequently, the cohort marketplace allows users to form and build cohorts by simply posting selection criteria and executing parameters in the dashboard. Clearing rates can also be posted and payouts executed using a suitably configured trading, e.g., currency exchange program.

[0851] Anyone who accepts participation in the cohort may then download the criteria and data file (or data files) and/or use genetic data from subjects that they have already generated and/or stored in performing the requested analyses. For example, each cohort participant will have, or may generate, a database of BCL and/or FASTQ files that are stored on their individual servers. These genetic files will have been derived from subjects who happened to meet the selection criteria. Specifically, such stored genetic data and/or other subject data may be screened to determine suitability for inclusion in the cohort selection criteria. Such data may have been generated for a variety of purposes, but regardless of the reasons for generation, once generated, it may be selected and subjected to the requested segmentation analyses and used for inclusion in the cohort.

[0852] Accordingly, in various embodiments, the cohort system may be a forum for connecting researchers to enable them to pool their resources and data, e.g., genetic sequence data. For example, engaging a cohort would allow a first researcher to introduce a project that requires genetic data analyses that require mining and/or examining multiple genomes from multiple subjects, such as with respect to mapping, alignment, variant calling, and/or the like. Therefore, rather than needing to gather subjects and collect sample sets individually, the cohort initiator may announce the need for a prescribed analysis procedure to be performed on sample sets previously or to be collected by others, and as such a collective approach to generating sample sets and analyzing the same is provided by the cohort organization herein. In particular, the cohort initiator may define the cohort selection, create a configuration file to be shared with potential cohort participants, create the workflow parameters, e.g., in a workflow folder, and may thereby automate data generation and analyses, e.g., through the workflow management system. The system may also enable the commercial aspect of the transaction, e.g., payment processing to compensate cohort participants for their provision of genetic datasets that may be analyzed, such as with respect to mapping, alignment, variant calling, and/or with respect to tertiary analyses.

[0853] In various embodiments, structured cohort analyses may be directed toward primary processing, e.g., of DNA or RNA, such as with respect to image processing and/or base quality score recalibration, methylation analysis, and the like; and/or may be directed toward performing secondary analysis, such as with respect to mapping, alignment, selection, variant calling, and the like; and/or may be directed toward tertiary analysis, such as with respect to array, genomic, epigenomic, metagenomic, genotyping, variant analysis, and/or other forms of tertiary analysis. Additionally, it should be understood that while many of the segmentations and analyses performed may thus involve primary and/or secondary processing, various analysis platforms herein may not be directed toward primary or secondary processing. For example, in certain cases, an analysis platform may be exclusively directed toward performing tertiary analysis, such as on genetic data, or other forms of genomic and/or bioinformatics analyses.

[0854] For example, in particular embodiments, with respect to the particular analytical procedures to be performed, the analyses to be performed may include one or more of mapping, alignment, classification, variant calling, and the like, to produce result data that may be subjected to one or more secondary and/or tertiary analytical procedures, depending on the specific segmentations selected to be performed. The workflow may be simple or it may be complex, e.g., it may require the performance of a segmentation module, e.g., mapping, or multiple modules, such as mapping, alignment, classification, variant calling, and/or others, but an important parameter is that the workflow must be identical for each person who is part of the cohort. In particular, a unique feature of the system is that the applicant establishing the cohort introduces control parameters in order to ensure that the analysis to be performed is in the same manner, regardless of where those procedures are performed and on which machines.

[0855] Accordingly, upon defining the cohort, the requester will upload both selection criteria together with a configuration file. Other cohort participants will then view the selection criteria to determine whether they have genetic information datasets that meet the criteria provided, and if so, will perform the requested analysis on the data based on the settings in the configuration file. Research subjects may be enrolled to be selected as a cohort participant, and if enrollment is large, a lottery or competition may be held to select participants. In many cases, a bidding system could be initiated. The results data generated by cohort participants may be processed on-site or in the cloud, and as long as the configuration file is followed, the data processing will be the same. In particular, the configuration file shows how the BioIT analytical device should be configured, and once the device is set up according to the prescribed configuration, a device associated with the system will perform the requested genetic analyses in the same manner, regardless of where it is located, e.g., locally or remotely. The results data can then be uploaded to the cohort marketplace and payment submitted and received in view of the results data received.

[0856] For example, the analysis of genetic data may be performed locally and the results uploaded to the cloud, or the genetic data itself may be uploaded and the analyses performed in the cloud, e.g., a server or server network, such as a quantum processing platform, associated with the cloud. In many cases, it may be useful to only upload the result data in order to better protect the identities of the subjects. In particular, by uploading only the result data, not only is security protected, but large amounts of data do not need to be transferred, thus enhancing the efficiency of the system.

[0857] More particularly, in many cases, a compressed file containing outcome data from one or more segmentations may be uploaded, and in some cases, only a file containing a description of variations need be uploaded. In some cases, only one response need be given, such as a text response, e.g., a "yes" or "no" answer. Such responses are preferred as they do not reveal the identity of the subject. However, if analyses need to be performed online, e.g., in the cloud, the selected BCL and/or FASTQ files may be uploaded, the analyses performed, and the outcome data may then be backed up to the initial submitter, who may then upload the outcome data into the cohort interface. The original raw data may then be deleted from the online memory. In these and other ways, the cohort requester will not have access to the identities of the subjects.

[0858] Compression, such as that employed in "just-in-time" (JIT) analysis, is particularly useful in enhancing cohort efficiency. For example, using typical procedures, the movement of data into and out of the cohort system is very expensive. Consequently, although in many configurations, raw and/or uncompressed data uploaded to the system may be stored therein, in particular cases, the data may be compressed prior to being uploaded, the data may then be processed within the system, and the results may then be compressed prior to being transmitted out of the system, such as where compression is performed according to a JIT protocol. In such a case, storage of such data, such as in a compressed form, is less expensive and therefore the cohort system is much more cost-effective.

[0859] Additionally, in various cases, a plurality of cohorts may be provided within an online marketplace, and given the compression processes described herein, data may be transmitted from one cohort to another to allow research from several different cohorts to share data between them, which, without the compression methods disclosed herein, would be prohibitively expensive. In particular, without the speed and efficiency of JIT, compression data, once transmitted to the cloud, would typically remain in the cloud, although accessible there for analysis and manipulation. However, JIT allows data to be rapidly transmitted to and from the cloud for both local and cloud-based processing. Furthermore, as can be seen with respect to Figure 41B and 43, in particular cases, the system 1 can be configured to subject the generated and/or processed data in a secondary manner to further processing, for example, through a local computing resource 100 and/or remote computing resource 300, such as by running the same through one or more tertiary processing threads, such as one or more of a microarray analysis thread, a genome, for example, which genome analysis thread, genotyping analysis thread, exome analysis thread, epigenome analysis thread, metagenome analysis thread, microbiome analysis thread, genotyping analysis thread including junction genotyping, variant analysis thread including structural variant threads, somatic variant threads, and GATK and/or MuTect2 threads, as well as RNA sequencing threads and/or other tertiary processing thread. Such result data from such processing may then be compressed and/or stored remotely 400 and/or be transferred in order to be stored locally 200.

[0860] In particular, one or more, e.g., all, of these functions may be performed locally, e.g., at site 10, in a local cloud 30, or via controlled access via hybrid cloud 50. In such a case, a developer environment is created which allows a user to control the functionality of the system to meet his or her individual needs and/or allow access thereto to others seeking the same or similar results. Accordingly, the various components, processes, procedures, tools, layers, and hierarchies of the system may be configurable, such as via a GUI interface which allows the user to select which components of the system to operate on which data at what time and in what order according to user-determined desires and protocols in order to generate relevant data and connections between data which may be securely communicated throughout the system locally or remotely. As indicated, these components may be made to communicate seamlessly together, e.g., independently of location and/or as connected, such as being in a tightly coupled configuration and/or a seamless cloud-based coupling, and/or being configurable, e.g., via a JIT protocol, in order to operate the same or similar processes in the same or similar manner, such as by employing corresponding API interfaces dispersed throughout the system, the employment of which allows various users to configure the various components to operate the various procedures in a similar manner.

[0861] For example, an API may be defined in a header file with respect to the processes to be operated by each particular component of the system 1, wherein the header describes the functionality and determines how to call a function, such as the parameters that are passed, the inputs received and the outputs transmitted and the manner in which this occurs, what goes in and how, what goes out and how, and what is returned and in what form. For example, in various embodiments, one or more of the components and/or elements thereof, which may form one or more pipelines of one or more layers of the system, may be configurable such as by means of instructions entered by a user and/or one or more second and/or third party applications. These instructions may be communicated to the system via corresponding APIs that communicate with one or more of the system's various drivers, instructing the driver (or drivers) which parts of the system, e.g. which modules and/or which processes thereof, should be activated, when, and in what order, given a pre-selected parameter configuration, which may be determined by a user-selectable interface, e.g. GUI.

[0862] In particular, the one or more DMA drivers of system 1 may be configured to operate correspondingly, such as at the kernel level of each component and of system 1 as a whole. In such a case, one or more of the provided kernels may have their own very low-level core API that provides access to the hardware and functions of the various components of system 1 in order to access applicable registers and modules in order to configure and direct the processes and the ways in which they are operated in system 1. Specifically, on top of this layer, a virtual layer of service functions may be constructed in order to form the building blocks that are used for a multitude of functions that send files to the kernel (or kernels) and get results back, encode, encrypt and/or transmit the relevant data, and additionally perform higher-level functions thereon. On top of that layer, an additional layer may be built which uses these service functions, which may be at the API level with which a user can interface, which may be adapted to function primarily for configuring the system 1 as a whole or its component parts, downloading files and uploading results, where the files and/or results may be transmitted across the system locally or globally. Additional APIs may be configured and included as discussed in greater detail above in relation to secure data storage.

[0863] Such configuration of the various APIs, memories and/or system firmware may include communicating with registers and also performing functional calls. For example, as described herein above, one or more function calls necessary and/or useful to perform the steps, e.g., sequentially, to perform a mapping and/or an alignment and/or a classification and/or a variant call or other secondary and/or tertiary functions as described herein, may be implemented in accordance with the hardware operations and/or related algorithms in order to generate the necessary processes and perform the required steps.

[0864] Specifically, because in certain embodiments one or more of these operations may be based on one or more structures, the various structures required to implement these operations may need to be constructed. Therefore, a function call will perform this function, wherein the function call will cause the required structure to be constructed to perform the operation, and because of this, a call will accept a file name of where the structure parameter files will be stored and will then generate one or more data files that contain and/or configure the required structure. Another function call may be to load the structure that was generated by the respective algorithm and transfer it to on-chip and/or system memory 1 and/or place it in the correct location where the hardware is expecting it to be. Obviously, various data will need to be downloaded to the chip and/or otherwise transferred to the system generator, as well as to perform various other selected functions of system 1, and the configuration manager may perform these functions, such as loading whatever needs to be there in order for the pipeline modules of the layers of the chip platforms and/or system as a whole to perform their functions, into on-chip and/or system memory attached to or otherwise associated with them.

[0865] Additionally, the system may be configured to allow various system components to communicate with each other, such as to allow one or more chips of system 1 to interface with the circuit board of sequencer 121, computing resource 100/300, transformer 151, analyzer 152, interpreter 310, collaborator 320, or other system component, when included therein, in order to receive FASTQ and/or other genetic sequencing files generated and/or processed directly from the sequencer or other processing component, such as immediately once they have been generated and/or processed, and then transfer that information to the configuration manager, which then directs that information to the appropriate memory banks in the hardware and/or software that make that information available to the pertinent modules of the hardware, software, and/or system as a whole so that they may perform their designated functions on that information in order to call bases, mapping, aligning, classifying, etc. the sample DNA/RNA against the reference genome and/or performing associated secondary and/or tertiary processing operations on the same.

[0866] Accordingly, in various embodiments, a client-level interface (CLI) may be included, wherein the CLI may follow the user to call one or more of these functions directly. In various embodiments, the CLI may be a software application, e.g., that has a GUI, which is adapted to configure the accessibility and/or use of the hardware and/or various other software applications of the system. The CLI, therefore, may be a program that accepts instructions, e.g., arguments, and makes functionality available simply by calling an application program. As indicated above, the CLI may be command-line based or GUI (graphical user interface) based. The line-based commands occur at a level below the GUI, wherein the GUI includes a window-based file manager with click-through function boxes that delineate which modules, which pipelines, which layers, from which platforms will be used, and the parameters of their use. For example, in the operation, if instructed, the CLI will locate the reference, determine if a hash table and/or index needs to be generated, or if already generated, locate where it is stored and direct the upload of the generated hash table and/or index, etc. These types of instructions may appear as user options in the GUI where the user can select the associated chip(s)/system 1 to perform.

[0867] Additionally, a library may be included, wherein the library may include pre-existing, editable configuration files, such as files geared toward user-selected operation of the associated hardware and/or software, such as with respect to a portion or a complete genome and/or protein analysis, for example, for various analyses, such as personal medical history and ancestry analysis, or disease diagnosis, or drug discovery, therapeutic substances, and/or one or more other analyses, etc. These types of parameters may be predefined, such as for performing such analyses, and may be stored in the library. For example, if the platform described herein is employed, such as for NIPT, NICU, Cancer, LDT, AgBio, and related research on a collective level, the predefined parameters may be configured differently than if the platform were simply directed to research genomics and/or genealogy-based research, such as on an individual level.

[0868] More particularly, for the specific diagnosis of an individual, accuracy may be an important factor. Therefore, system parameters may be set to ensure increased accuracy although at the cost of possibly decreasing speed. However, for other genomic applications, speed may be the key determinant and therefore system parameters may be set to maximize speed, which may however sacrifice some accuracy. Accordingly, in various embodiments, parameter settings typically used to perform different tasks may be predefined in the library to facilitate ease of use. Such parameter settings may also include the necessary software applications and/or hardware configurations employed in the operation of the system 1. For example, the library may contain code that executes the API and may additionally include sample files, scripts, and any other auxiliary information necessary to operate the system 1. Accordingly, the library may be configured to compile software to operate the API as well as several of the executables.

[0869] Additionally, as can be seen with respect to Figure 42C and 43, the system can be configured such that one or more of the system components can be performed remotely, such as when the system component is adapted to perform one or more comparative functions on the data, such as an interpretive function 310 and/or collaborative function 320. For example, when an interpretive protocol is employed on the data, the interpretive protocol 312 can be configured to analyze and draw conclusions about the data and/or determine various relationships thereto, one or more other analytical protocols can also be performed and include annotating the data 311, performing a diagnosis 313 on the data, and/or analyzing the data to determine the presence or absence of one or more biomarkers 314. As indicated, one or more of these functions can be directed by the WMS and/or performed by the A/I module disclosed herein.

[0870] Additionally, when a collaborative protocol is performed, the system 1 may be configured to provide an electronic forum where data sharing 321 may occur, wherein the data sharing protocol may include user-selectable security 324 and/or privacy 322 settings that allow the data to be encrypted and/or password protected so that the identity and sources of the data may be hidden from a user of the system 1. In particular cases, the system 1 may be configured to allow a third-party analyzer 121 to run virtual simulations on the data. Additionally, once generated, the interpreted data and/or data subjected to one or more collaborative analyses may be stored remotely 400 or locally 200 in order to be made available to remote 300 or local 100 computing resources, such as for further processing and/or analysis.

[0871] In another aspect, as can be seen with respect to Figure 44, a method for using the system to generate one or more data files upon which one or more secondary and/or tertiary processing protocols can be operated is provided. For example, the method can include providing a genomic infrastructure, such as for one or more of on-premises, cloud-based, and/or hybrid genomic and/or bioinformatics generation and/or processing and/or analysis.

[0872] In such a case, the genomic infrastructure may include a bioinformatics processing platform that has one or more memories that are configured to store one or more configurable processing structures for configuring the system to be capable of performing one or more analytical processing functions on data, such as data that includes a genomic sequence of interest or processed result data that pertains thereto. The memory may include the genomic sequence of interest to be processed, e.g., once generated and/or acquired, one or more genetic reference sequences, and/or may further include an index of the one or more genetic reference sequences and/or a list of splice junctions that pertain thereto. The system may also include an input that has a platform application programming interface (API) for selecting from a list of options of one or more of the configurable processing structures, such as for configuring the system, such as selecting which processing functions of the system will be operated on the data, e.g., the pre- or pre-processed genomic sequences of interest. A graphical user interface (GUI) may also be present, such as operationally associated with the API, in order to present a menu by which a user can select which of the available options he wishes to have operated on the data.

[0873] Thus, in these and/or other such cases, the hybrid cloud 50 may be configured to allow uninterrupted and protected transmission of data by all components of the system, such as when the hybrid cloud 50 is adapted to allow the various users of the system to configure its component parts and/or the system itself, for example, by means of the WMS, in order to meet the user's research and/or development, diagnostic, therapeutic and/or prophylactic testing needs. In particular, the hybrid cloud 50 and/or the various components of the system 1 may be operationally connected to compatible and/or corresponding API interfaces that are adapted to allow a user to remotely configure the various components of the system 1 in order to install the desired resources in the desired manner and to do so additionally locally, remotely or in a combination thereof, as based on the demands of the system and the particulars of the analyses being performed, being able at all times to communicate in a secure, encryptable environment.

[0874] As described above, the system may be implemented on one or more integrated circuits that may be formed by one or more sets of configurable digital logic circuits, e.g., preconfigured and/or hardwired, that may be interconnected via a plurality of physical electrical interconnects. In such a case, the integrated circuit may have an input, such as a memory interface, for receiving one or a plurality of the configurable framework protocols, e.g., from memory, and may be further adapted to implement the one or more frameworks on the integrated circuit in accordance with the configurable framework processing protocols. The input memory interface may also be configured to receive genomic sequence data, which may be in the form of a plurality of genomic data reads. The interface may also be adapted to access the one or more genetic reference sequences and the index (or indices).

[0875] In various instances, the digital logic circuitry may be arranged as a set of processing engines that are each formed by a subset of the digital logic circuitry. The digital logic circuitry and/or the processing engines may be configured to perform one or more preconfigurable steps of a primary, secondary, and/or tertiary processing protocol to generate the plurality of genomic sequence data reads and/or to process the plurality of genomic data reads, such as in accordance with the reference gene sequence(s) or other genetic sequence-derived information. The integrated circuit may additionally have an input for outputting result data from the primary, secondary, and/or tertiary processing, such as in accordance with the platform application programming interface (API).

[0876] Particularly, in various embodiments, the digital logic circuits and/or sets of processing engines may form a plurality of genomic processing pipelines, such as where each pipeline may have an input that is defined in accordance with the platform application programming interface for receiving primary and/or secondary processing result data by the bioinformatics processing platform and for performing one or more analytical processes thereon to produce result data. Additionally, the plurality of genomic processing pipelines may have a common pipeline API that defines a secondary and/or tertiary processing operation to be operated on the result data of the primary and/or secondary processed data, such as where each of the plurality of genomic processing pipelines is configured to perform a subset of the secondary and/or tertiary processing operations and to output secondary and/or tertiary processing result data in accordance with the pipeline API.

[0877] In such cases, a plurality of genomic analysis applications may be stored in memory and/or an associated searchable application repository, such as where each of the plurality of genomic analysis applications is accessible via an electronic medium by a computer, such as for execution by a computer processor, in order to perform a targeted analysis of pre- or post-processed genomic data from primary, secondary, and/or tertiary processing result data, such as by one or more of the plurality of genomic processing pipelines. In particular cases, each of the plurality of genomic analysis applications may be defined by the API and may be configured to receive primary, secondary, and/or tertiary processing result data and/or to perform targeted analysis of the pre- or post-processed genomic data and to output the targeted analysis result data to one of one or more genomic databases.

[0878] The method may further include selecting, for example, from a GUI menu, one or more genomic processing pipelines from a plurality of the system's available genomic processing pipelines; selecting one or more genomic analysis applications from the plurality of genomic analysis applications that are stored in an application repository; and executing, using a computer processor, the one or more selected genomic analysis applications to perform a targeted analysis of genomic data from the primary, secondary, and/or tertiary processing result data.

[0879] Additionally, in various embodiments, all of the mapping, alignment and classification, and variant calling may occur on-chip, and local realignment, duplicate marking, base quality score recalibration may also occur, and/or one or more of the tertiary processing protocols and/or pipelines, in various embodiments, may occur on-chip or in software, and in various cases, various compression protocols, such as SAM and/or BAM and/or CRAM, may also occur on-chip. However, once the primary, secondary, and/or tertiary processed data has been produced, it may be compressed, such as before being transmitted, such as by being sent through the system, being sent to the cloud, such as for performance by the variant calling module, a secondary, tertiary, and/or other processing platform, such as including an interpretive and/or collaborative analysis protocol. This can be useful especially given the fact that variant calling, including tertiary processing thereof, can be a moving target, e.g. there is no agreed standard according to the algorithm that the industry uses.

[0880] Accordingly, different algorithms may be employed, such as by remote users, in order to achieve a different type of result as desired, and as such having a cloud-based module to perform this function may be useful to allow flexibility in selecting which algorithm is useful at any particular time, and also for serial and/or parallel processing. Accordingly, any of the modules disclosed herein may be deployed as hardware, e.g., on-chip, or software, e.g., in the cloud, but in certain embodiments, all of the modules may be configured such that their function may be performed on-chip, or all of the modules may be configured such that their function may be performed remotely, such as in the cloud, or there will be a mix of modules where some are located on one or more chips and some are located in the cloud. Additionally, as indicated, in various embodiments, the chip(s) itself may be configured to function in conjunction with, and in some embodiments, in immediate operation with, a genetic sequencer, such as an NGS and/or a sequencer-on-a-chip.

[0881] More specifically, in various embodiments, an apparatus of the disclosure may be a chip, such as a chip that is configured to process genomic data, such as by employing a segmentation of data analysis modules. Accordingly, as can be seen with respect to Figure 45 , a genomic segmentation processor chip 100 is provided in conjunction with associated hardware of a genomic segmentation processor system 10. The chip 100 has one or more connections to external memory 102 (at “DDR3 Memory Controller”) and a connection 104 (e.g., “PCIe Interface or QPI”) to the outside world, such as a host computer 1000, for example. A crossbar 108 (e.g., switch) provides access to the memory interfaces to various requestors. DMA engines 110 transfer data at high speeds between the host and external memories 102 of processor chip 100 (via crossbar 108) and/or between the host and a central controller 112. Central controller 112 controls chip operations, particularly by coordinating the efforts of multiple processing engines 13. Processing engines are comprised of a set of hard-wired digital logic circuits that are interconnected via physical electrical interconnects and are organized into engine clusters 11/114. In some implementations, engines 13 in a cluster 11/114 share a crossbar port, via an arbiter 115. Central controller 112 has connections to each of the engine clusters. Each engine cluster 11/114 has a plurality of processing engines 13 for processing genomic data, including a mapper 120 (or mapping module), an aligner 122 (or alignment module), and a classifier 124 (or classification module), and one or more processing engines for performing other functions, such as variant calling, may also be provided. Thus, an engine cluster 11/114 may also include other engines or modules, such as a variant caller module.

[0882] In accordance with a data flow model consistent with the implementations described herein, the host CPU 1000 sends commands and data via the DMA engines 110 to the central controller 112, which load balances the data to the processing engines 13. The processing engines return processed data to the central controller 112, which continuously transmits the same back to the host via the DMA engines 110. This data flow model is suitable for mapping and alignment and variant calling. As indicated, in various cases, communication with the lost CPU may be via a relatively loose or tight coupling, such as a high-bandwidth, low-latency interconnect, such as a QPI, to maintain cache coherence between the associated memory elements of the two or more devices.

[0883] For example, in various cases due to various power and/or space constraints, such as when performing big data analytics such as mapping/alignment/variant calling in a hybrid software/hardware accelerated environment as described herein, where data needs to be moved both rapidly and continuously between system devices, a cache coherent tight coupling interface may be useful for performing such data transmissions throughout the system to and from the coupled devices, such as to and from the sequencer, DSP (digital signal processor), CPU and/or GPU or CPU/GPU hybrid, accelerated integrated circuit, e.g., FPGA, ASIC (on network card), as well as other intelligent network accelerators in a cache coherent and fast manner. In such cases, a tightly coupled interconnect with adequate cache coherence may be one or more of a single interconnect technology specification that is configured to ensure that processing, such as between a plurality of processing platforms using different instruction set architectures (ISAs), can coherently share data between the different platforms and/or with one or more associated accelerators, e.g., such as a wired FPGA-deployed accelerator, in order to enable efficient heterogeneous computing and thereby significantly improve the efficiency of the system, which in various cases may be configured as a cloud-based server system. Thus, in certain cases, a low-latency, high-bandwidth, cache-coherent interconnect protocol such as QPI, Coherent Processor Accelerator Interface (CAPI), NVLink/GPU, or other suitable interconnect protocol may be employed to expedite various data transmissions between various system components, such as those pertaining to mapping, alignment, and/or variant calling computation functions that may involve the use of acceleration mechanisms whose operation requires the need to access, process, and move data seamlessly between various system components where the various data to be processed resides in the system. And, where such data is retained within an associated memory device such as RAM or DRAM, the transmission activities may also involve fast coherent lookup and in-memory database processing.

[0884] Particularly, in particular embodiments, such heterogeneous computing may involve a multiplicity of processing and/or acceleration architectures that may be interconnected in a reduced instruction set computing format. In such a case, such an interconnect device may be a six coherent connection interconnect (CCVI) device, which is configured to allow all computing components within the system to address, read from, and/or write to one or more associated memories in a coherent, consistent, and unique manner. More particularly, a CCVI interconnect may be employed to connect multiple system devices, such as the CPU and/or GPU or CPU/GPU hybrid, FPGA and/or associated memories, etc. to each other, such as in a high bandwidth manner that is configured to increase transfer rates between the multiple components while evidencing greatly reduced latency rates. Specifically, a CCVI interconnect may be employed and configured to allow system components to access and process data regardless of where the data resides, and without the need for complex programming environments that would otherwise need to be implemented to make the data coherent. Other such interconnects that may be employed in order to speed up, e.g., decrease, processing time and increase accuracy include QPI, CAPI, NVLink, or other interconnect that may be configured to interconnect the various system components and/or sit on top of an associated PCI Express peripheral interconnect.

[0885] Thus, in accordance with an alternative data flow model consistent with the implementations described herein, host CPU 1000 continuously transmits data to external memory 1014, either directly via DMA engines 110 and crossbar 108 or via central controller 112. Host CPU 1000 sends commands to central controller 112, which sends commands to processing engines 13, which instruct the processing engines what data to process. Due to the tight coupling, the processing engines 13 access input data directly from the external memory 1014 or a cache memory associated therewith, process it, and write the results back to the external memory 1014, such as over the tightly coupled interconnect 3, reporting the situation to the central controller 112. The central controller 112 either continuously transmits the result data back to the host 1000 from the external memory 1014 or notifies the host to itself fetch the result data via the DMA engines 110.

[0886] Figure 46 illustrates a genomic segmentation system and processor 20, showing a full complement of processing engines 13 within an engine cluster 11/214. Segmentation processor system 20 may include one or more engine clusters 11/214. In some implementations, segmentation processor system 20 includes four or more engine clusters 11/214. Processing engines 13 or processing engine types may include, without limitation, a mapper, an aligner, a classifier, a local realigner, a base quality recalibrator, a duplicate marker, a variant caller, a compressor, and/or a decompressor. In some implementations, each engine cluster 11/214 has one of each processing engine type. Accordingly, all processing engines 13 of the same type may access crossbar 208 simultaneously, through different crossbar ports, because they are each in a different engine cluster 11/214. Not every processing engine type need be realized in every engine cluster 11/214. Processing engine types that require massive parallel processing or memory bandwidth, such as the mapper (and attached aligner(s)) and the sorter, may appear in every engine cluster of the segmentation processor system 20. Other engine types may appear in only one or some of the engine clusters 214, as needed to satisfy their performance requirements or the performance requirements of the segmentation processor system 20.

[0887] Figure 47 illustrates a genomic segmentation processor system 30 showing, in addition to the engine clusters 11 described above, one or more embedded central processing units (CPUs) 302. Examples of such embedded CPUs include standard Snapdragons® or ARM® cores, or in other cases, may be an FPGA. These CPUs execute fully programmable bio-IT algorithms, such as advanced variant calling, such as constructing a DBG or performing an HMM. Such processing is accelerated by computing functions in the various engine clusters 11, which may be called by CPU cores 302 as needed. Additionally, uniform engine central processing, such as mapping and alignment, may be managed by CPU cores 302, giving them increased programmability.

[0888] Figure 48 illustrates a processing flow for a genomic segmentation processor system and method. In some preferred implementations, there are three passes through the data. The first pass includes mapping 402 and alignment 404, with the complete read set streamed continuously through engines 13. The second pass includes classification 406, in which a large block to be classified (e.g., a substantial portion or all of the reads previously mapped to a single chromosome) is loaded into memory, classified by the processing engines, and returned to the host. The third pass includes downstream stages (local realignment 408, duplicate flagging 410, base quality score recalibration (BQSR) 412, SAM output 414, reduced BAM output 416, and/or CRAM compression 418). The steps and functions of the third pass may be done in any combination or subcombination, and in any order, in a single pass.

[0889] Thus, in this manner, data is passed in a relatively continuous fashion from the one or more processing engines to the host CPU, such as in accordance with one or more methodologies described herein. Thus, a virtual segmentation architecture, such as as described above, is used to continuously stream reads from the host to circular buffers in memory, through one processing engine after another in sequence, and back to the host. In some implementations, CRAM decompression may be a separate streaming function. In some implementations, SAM output 414, reduced BAM output 416, and/or CRAM compression 418 may be replaced with variant call, compression, and decompression.

[0890] In many cases, a hardware implementation of a sequence analysis segment is described. This can be done in many different ways, such as in an FPGA or ASIC or structured ASIC implementation. The functional blocks that are implemented by the FPGA or ASIC or structured ASIC are shown in Figure 49. Accordingly, the system includes several blocks or modules to perform sequence analysis. The input to the hardware implementation can be a FASTQ file, but is not limited to that format. In addition to the FASTQ file, the input to the FPGA or ASIC or structured ASIC consists of secondary information, such as Flow Space Information from the technology, such as from NGS. The blocks or modules can include the following blocks: Error Control, Mapping, Alignment, Classification, Local Realignment, Duplicate Marking, Base Quality Recalibration, BAM and Secondary Information reduction, and/or variant calling.

[0891] These blocks or modules may be present within, or implemented by, the hardware, but some of these blocks may be omitted or other blocks added to achieve the purpose of performing a sequence analysis segmentation. Blocks 2 and 3 describe two alternatives of the sequence analysis segmentation platform. The sequence analysis segmentation platform comprising an FPGA or an ASIC or a structured ASIC and software supported by a host (e.g., PC, server, cluster or cloud computing) with cloud storage and/or cluster. Blocks 4 through 7 describe different interfaces that the sequence analysis segmentation may have. In Blocks 4 and 6, the interface may be a PCIe interface and/or QPI/CAPI/CCVI/NVLink interface, but is not limited to a PCIe, QPI or other interface. In Blocks 5 and 7, the hardware (FPGA or ASIC or structured ASIC) may be directly integrated into a sequencing machine. Blocks 8 and 9 describe the integration of hardware sequence analysis segmentation embedded in a host system, such as a PC, a server cluster, or a sequencer. A plurality of DDR3 memory elements and a PCIe/QPI/CAPI/CCVI/NVLink interface surround the FPGA or ASIC or hardware-structured ASIC. The board with the FPGA/ASIC/sASIC connects to a host computer, which consists of a host CPU and/or GPU, which may be a low-power CPU, such as an ARM®, Snapdragon®, or any other processor. Block 10 illustrates a hardware sequence analysis segmentation API that can be accessed by third-party applications to perform tertiary analysis.

[0892] Figures 50A and 50B depict an expansion card 104 having a processing chip 100, e.g., an FPGA, of the disclosure, as well as one or more associated elements 105 for coupling the FPGA 100 with the host CPU/GPU, such as for transferring data, such as data to be processed and result data, back and forth from the CPU/GPU to the FPGA 100. Figures 50B depict the expansion card of Figure 50A having a plurality, e.g., 3, of slots containing a plurality, e.g., 3, of processing chips of the disclosure.

[0893] Specifically, as depicted in Figures 50A and 50B, in various embodiments, an apparatus of the disclosure may include a computing architecture, such as embedded in a silicon field-programmable gate array (FPGA) or application-specific integrated circuit (ASIC) 100. The FPGA 100 may be integrated onto a printed circuit board (PCB) 104, such as a Peripheral Component Interface - Express (PCIe) card, which may be connected to a computing platform. In various cases, as shown in Figure 50A, the PCIe card 104 may include a single FPGA 100, wherein the FPGA may be surrounded by local memories 105, however, in various embodiments, as depicted in Figure 50B, the PCIe card 104 may include a plurality of FPGAs 100A, 100B, and 100C. In various instances, the PCI card may also include a PCIe bus. This PCIe card 104 may be added to a computing platform to execute algorithms on extremely large data sets. In an alternative embodiment, as noted above with respect to Figure 34, in various embodiments, the FPGA may be adapted to be directly associated with the CPU/GPU, such as through an interpolator, and tightly coupled thereto, such as through a QPI, CAPI, CCVI interface. Accordingly, in various instances, the overall genomic sequencing workflow involving the FPGA may include the following: Sample Preparation, Alignment (including mapping and alignment), Variant Analysis, Biological Interpretation, and/or Specific Applications.

[0894] Thus, in various embodiments, an apparatus of the disclosure may include a computing architecture that achieves high-performance execution of algorithms, such as mapping and alignment algorithms, that operate on extremely large data sets, such as when the data sets exhibit poor locality of reference (LOR). These algorithms are designed to reconstruct a complete genome from millions of short-read sequences, from modern so-called next-generation sequencers, which require multi-gigabyte data structures that are randomly accessed. Once reconstruction is achieved, as described herein above, additional algorithms with similar characteristics are used to compare a genome to libraries of others, perform gene function analysis, etc.

[0895] There are two other typical architectures that can generally be constructed to perform one or more of the operations described in detail herein, such as including multi-core purpose CPUs and general-purpose graphics processing units (GPGPUs). In such a case, each CPU/GPU in a multi-core system can have a classic cache-based architecture, where instructions and data are fetched from a level 1 cache (L1 cache) that is small but has extremely fast access. Multiple L1 caches can be connected to a larger but slower shared L2 cache. The L2 cache can be connected to a large but slower DRAM (Dynamic Random Access Memory) system memory, or it can be connected to an even larger but slower L3 cache, which can then be connected to DRAM. One advantage of this arrangement may be that applications in which programs and data exhibit locality of reference behave almost as if they were running on a computer with a single memory as large as DRAM but as fast as L1 cache. Because fully custom, highly optimized CPUs operate at very high clock rates, e.g., 2 to 4 GHz, this architecture may be essential to achieving good performance. Additionally, as discussed in detail with respect to Figure 33, in various embodiments, the CPU may be tightly coupled to an FPGA, such as an FPGA configured to perform one or more functions related to the various operations described herein, such as through a low latency, high bandwidth interconnect such as a QPI, CCVI, CAPI, in order to further enhance performance as well as the speed and coherence of data transferred throughout the system. In such a case, cache coherence may be maintained between the two devices, as noted above.

[0896] Additionally, GPGPUs can be used to extend this architecture, such as by deploying very large numbers of small CPUs, each with its own small L1 cache, where each CPU executes the same instructions on different subsets of the data. This is an architecture called SIMD (Single Instruction Stream, Multiple Data Stream). Economy can be gained by sharing the instruction fetch and decode logic across a large number of CPUs. Each cache has access to multiple large external DRAMs via an interconnect network. Assuming that the computation to be performed is highly parallelizable, GPGPUs have a significant advantage over general-purpose CPUs due to having large numbers of computing resources. However, they still have a cache architecture and their performance suffers for applications that do not have a high enough degree of locality of reference. This leads to a high cache miss rate and processors that are idle while waiting for data to arrive from the external DRAM.

[0897] For example, in many cases, Dynamic RAMs can be used for system memory because they are less expensive than Static RAMs (SRAMs). The elevator rule used to be that DRAMs had 4x the capacity for the same cost as SRAMs. However, due to the decreasing demand for SRAMs in favor of DRAMs, this difference has increased considerably due to economies of scale favoring DRAMs, which are in high demand. Regardless of cost, DRAMs are 4x denser than SRAMs placed on the same silicon area because they require only one transistor and one capacitor per bit compared to 4 transistors per bit to implement the SRAM flip-flop. DRAM represents a single bit of information such as the presence or absence of charge on a capacitor.

[0898] One problem with this arrangement is that the load decays over time, so it has to be refreshed periodically. The need to do this has led to architectures that organize memory into independent blocks and access mechanisms that deliver multiple words of memory per request. This compensates for the times when a given block is unavailable while it is being refreshed. The idea is to move as much of the data as possible while a given block is available. This is in contrast to SRAMs in which any location in memory is available in a single access in a constant amount of time. This characteristic allows memory accesses to be single-word oriented rather than block oriented. DRAMs work well in a cache architecture because each cache is redirected to a block of memory that is read from the DRAM. The theory of locality of reference is that if it recently accessed word N, then it will probably access words N+1, N+2, N+3, and so on.

[0899] Figure 51 provides an exemplary implementation of a system 500 of the disclosure including one or more expansion cards of Figure 50, such as for bioinformatics processing 10. The system includes a Bio IT processing chip 100 that is configured to perform one or more functions in a processing pipeline, such as base calling, error correction, mapping, alignment, classification, assembly, variant calling, and the like, as described herein.

[0900] The system 500 further includes a configuration manager that is adapted to configure the integrated operation of the one or more processors 100. Specifically, in various embodiments, the configuration manager is adapted to communicate instructions to the internal controller of the FPGA, e.g., firmware, such as via a suitably configured driver in a loose or tightly coupled interconnect, in order to configure the one or more processing functions of the system 500. For example, the configuration manager may be adapted to configure the internal processing clusters 11 and/or engines 13 associated therewith in order to perform one or more desired operations, such as mapping, alignment, classification, variant calling, and the like, in accordance with received instructions. In such a manner, only the clusters 11 that contain the processing engines 13 to perform the requested processing operations on the data provided from the host system 1000 to the chip 100 may be engaged to process the data in accordance with the received instructions.

[0901] Additionally, in various embodiments, the configuration manager may be further adapted so that it can be adapted by itself, e.g., remotely, by a third-party user, such as in an API connection as described in greater detail herein above, such as by a user interface (GUI) presented by an application of the system 500. Additionally, the configuration manager may be connected to one or more external memories, such as a memory that forms or otherwise contains a database, such as a database that includes one or more reference or individually sequenced genomes and/or an index thereof and/or one or more previously mapped, aligned, and/or classified genomes or portions thereof. In various cases, the database may further include one or more genetic profiles that characterize a disease state, such as for performing one or more tertiary processing protocols, such as on aligned and newly mapped genetic sequences or a VCF pertaining thereto.

[0902] The system 500 may also include a web-based access to enable remote communications such as over the Internet, to form a cloud or at least one hybrid cloud communications platform 504. In such a manner, processed information generated from the Bio IT processor, e.g., results data, may be encrypted and stored as an electronic health record, such as in an external, e.g., remote, database. In various cases, the EMR database may be searchable, such as with respect to genetic information stored therein, in order to perform one or more statistical analyses on the data, such as to determine trends or disease states or for the purpose of analyzing the effectiveness of one or more prophylaxis or treatments pertaining thereto. Such information together with the EMR data may then be further processed and/or stored in an additional database 508 in a manner that ensures confidentiality of the source of the genetic information.

[0903] More particularly, Figure 51 illustrates a system 500 for performing a segmentation sequence analysis on genetic sequence data. The system 500 includes a configuration manager 502 that includes a computing system. The computing system of the configuration manager 502 may include a personal computer or other computer workstation or may be implemented by a networked computer cluster. The configuration manager 502 may further include one or more third party applications connected to the computing system by one or more APIs, which, with one or more proprietary applications, generate a configuration for processing genomic data from a sequencer or other genomic data source. The configuration manager 502 further includes drivers that load the configuration into the genomic segmentation processor system 10. The genomic segmentation processor system 10 may output result data to, or be accessed through, the Web 504 or other network, for storage of the result data in an electronic health record 506 or other knowledge database 508.

[0904] As discussed at various points in this document above, the chip implementing the genomic segmentation processor may be connected to or integrated with a sequencer. The chip may also be connected to or integrated, e.g., directly via an interpolator or indirectly, e.g., with an expansion card, such as via PCIe, and the expansion card may be connected to or integrated with a sequencer. In other implementations, the chip may be connected to or integrated with a server computer that is connected to a sequencer, to transfer genomic reads from the sequencer to the server. In still other implementations, the chip may be connected to or integrated with a server in a cloud computing cluster of computers and servers. A system may include one or more sequencers connected (e.g., via Ethernet) to a server containing the chip, where genomic reads are generated by the multiple sequencers, transmitted to the server, and then mapped and aligned on the chip.

[0905] For example, in general next-generation DNA sequencer (NGS) data pipelines, the primary analysis stage processing is typically specific to a given sequencing technology. This primary analysis stage functions to translate physical signals detected within the sequencer into nucleotide sequence “reads” with associated quality (confidence) scores, e.g., FASTQ format files, or other formats that contain sequence and typically quality information. Primary analysis, as noted above, is typically quite specific in nature to the sequencing technology employed. In many sequencers, nucleotides are detected by sensing changes in fluorescence and/or electrical charges, electrical currents, or radiated light. Some primary analysis pipelines typically include: Signal processing to amplify, filter, separate, and measure sensor output; Data reduction, such as through quantization, decimation, weighting, transformation, etc.; Image processing or numerical processing to identify and enhance meaningful signals and associate them with specific reads and nucleotides (e.g., image offset calculation, cluster identification); Algorithmic processing and heuristic arithmetic to compensate for sequencing technology artifacts (e.g., phasing estimates, crosstalk matrices); Bayesian probability calculations; Hidden Markov Models; Base calling (which selects the most likely nucleotide at each position in the sequence); Base calling quality (confidence) estimation, and the like. As discussed in this document above, one or more of these steps may be benefited by implementing one or more of the necessary processing functions in hardware, such as implemented by an integrated circuit, e.g., an FPGA. Additionally, after such a format is achieved, secondary analysis proceeds, as described herein, to determine the DNA (or RNA, etc.) content of the sequenced sample, such as by mapping and aligning reads to a reference genome, classification, duplicate marking, base quality score recalibration, local realignment, and variant calling. Tertiary analysis may then proceed to extract medical or research implications from the determined DNA content.

[0906] Accordingly, given the sequential nature of the above processing functions, it may be advantageous to integrate primary, secondary, and/or tertiary processing acceleration into a single integrated circuit or multiple integrated circuits positioned on a single expansion card. This may be beneficial because sequencers produce data that typically requires both primary and secondary analysis in order to be useful and can be further used in multiple tertiary processing protocols, and integrating the same into a single device is most efficient in terms of cost, space, power, and resource sharing. Thus, in a particular aspect, the disclosure is directed to a system, such as a system for performing a segmentation sequence analysis on genetic sequence data. In various cases, the system may include an electronic data source, such as a data source that provides digital signals, e.g., digital signals representing a plurality of genomic data reads, wherein each of the plurality of genomic data reads includes a nucleotide sequence. The system may include one or more of a memory, such as a memory that stores one or more genetic reference sequences, and/or an index of the one or more genetic reference sequences; and/or the system may include a chip, such as an ASIC, an FPGA, or a sASIC.

[0907] One or more aspects or features of the subject matter described herein may be embodied in digital electronic circuitry, integrated circuitry, specially designed application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or structured ASIC computer hardware, firmware, software, and/or combinations thereof.

[0908] These various aspects or features may include implementation in one or more computer programs that are executable and/or interpretable on a programmable system that includes at least one programmable processor, which may be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device. The programmable system or computing system may include clients and servers. A client and a server are generally remote from each other and typically interact through a communications network. The client and server relationship arises by virtue of computer programs that operate on the respective computers and that have a client-server relationship with each other.

[0909] These computer programs, which may also be referred to as programs, software, software applications, applications, components or code, include machine instructions for a programmable processor and may be implemented in a high-level procedural and/or object-oriented programming language and/or assembly/machine language. As used herein, the term “machine-readable media” refers to any computer program product, apparatus and/or device, such as, for example, magnetic disks, optical disks, memory and Programmable Logic Devices (PLDs), used to provide machine instructions and/or data to a programmable processor, including a machine-readable media that receives machine instructions as a machine-readable signal. The term “machine-readable signal” refers to any signal used to provide machine instructions and/or data to a programmable processor. The machine-readable media may store such machine instructions in a non-transitory manner, such as, for example, as would a non-transitory solid-state memory or a magnetic hard disk or any equivalent storage medium. The machine-readable media may alternatively or additionally store such machine instructions in a transient manner, such as, for example, as would a processor cache or other random-access memory associated with one or more physical processor cores.

[0910] Additionally, due to the immense growth in data acquisition and production in the 21st century, a need has developed for increased processing power that can handle the increasingly computationally intensive analyses upon which modern development is based. Supercomputers have been introduced and have been instrumental in advancing technological development across a wide range of platforms. However, while supercomputing is useful, it has proven insufficient for some of the many complex computing problems that many modern technology companies face. Particularly since the sequencing of the human genome, technological advancement in biological techniques has been exponential. However, given the high rate and increasing complexity of raw data produced every day, a problematic hurdle has evolved in the processing and analysis of the generated data. Quantum computers have been developed to help solve this hurdle. Quantum computing represents a new frontier in computing, providing an entirely new approach to solving the world's most challenging computational needs.

[0911] Quantum computing has been known since 1982. For example, in the International Journal of Theoretical Physics, Richard Feynman theorized a system for performing quantum computation. Specifically, Feynman proposed a quantum system that could be configured for use in simulating other quantum systems so that conventional computer processing functions could be performed quickly and efficiently. See Feynman, 1982, International Journal of Theoretical Physics 21, pp. 467–488, which is incorporated herein by reference in its entirety. In particular, a quantum computer system can be designed to exhibit exponential time savings in complex computations. Such controllable quantum systems are commonly known as quantum computers, and have been successfully developed into general-purpose computing computers that can not only be used to simulate quantum systems, but can also be adapted to run specialized quantum algorithms. More particularly, complex problems may be modeled in the form of an equation, such as a Hamiltonian, which can be represented in the quantum system in such a way that the behavior of the system provides information concerning the solution to the equation. See Deutsch, 1985, Proceedings of the Royal Society of London A 400, pages 97–117, which is incorporated herein by reference in its entirety. In such cases, solving a model for the behavior of the quantum system may be configured to involve solving a differential equation related to the wave mechanical description of a particle, e.g., the Hamiltonian, of the quantum system.

[0912] In essence, quantum computing is a computational system that uses quantum mechanical phenomena, e.g., superposition and/or entanglement, to perform various calculations on large amounts of data extremely quickly. As such, quantum computers are a vast improvement over conventional digital logic computers. Specifically, conventional digital logic circuits operate using binary digital logic gates that are formed by directly connecting electronic circuitry to a conductive substrate. In a digital logic circuit, an “on/off” state of a transistor serves as a basic unit of information, e.g., a bit. In particular, a common digital computer processor employs digits, e.g., bits, in an “on” or “off” state, e.g., as a 0 or 1, to encode data. Quantum computing, on the other hand, employs an information device that uses superpositions of entangled states, called quantum bits or qubits, to encode data.

[0913] The basis for performing such quantum computations is an information device, e.g., a unit, which forms the quantum bit. The qubit is analogous to the digital "bit" in traditional digital computers, except that the qubit has greater computational potential than a digital bit. In particular, as described in greater detail in this document, instead of simply encoding one of two distinct states, such as a "0" and a "1", as found in a digital bit, a qubit can also be placed in a superposition of "0" and "1". Specifically, the qubit can exist in both the "0" and "1" states at the same time. Consequently, the qubit can perform a quantum computation in both states simultaneously. In general, N qubits can be in a superposition of 2N states. Quantum algorithms, therefore, can make use of this superposition property to speed up certain computations.

[0914] A qubit, therefore, is analogous to a bit in a traditional digital computer and is a type of information device that exhibits coherence. In particular, a quantum computing device is constructed from a plurality of information devices, e.g., qubit, building blocks. For example, the computing power of a quantum computer increases as the information devices that form its building blocks are coupled, e.g., entangled, together in a controllable manner. In such a case, the quantum state of one information device affects the quantum state of each of the other information devices to which it is coupled.

[0915] Consequently, like the bit in classical digital computing, the qubit in quantum computing serves as the basic unit for encoding information, such as quantum information. Similar to a bit, the qubit encodes data in a two-state system, which in this case is a quantum-mechanical system. Specifically, for the qubit, the two quantum states involve entanglement, as does the polarization of a single photon. Hence, whereas in a classical system, a bit must be in one state or the other, in a quantum computing platform, the qubit can be in a superposition of both states at the same time, such a property being fundamental to quantum processing. Consequently, the distinguishing feature between the qubit and the classical bit is that multiple qubits exhibit quantum entanglement. Such entanglement is a non-local property that allows a set of qubits to express greater correlation than is possible in a classical system.

[0916] In order to function, such information devices, e.g., quantum bits, must meet several requirements. First, the information device must be reducible to a two-level quantum system. This means that the information device must have two distinguishable quantum states that can be used to perform computations. Second, the information devices must be capable of producing quantum effects such as entanglement and superposition. Additionally, in certain cases, the information device may be configured to store information, e.g., quantum information, such as in a coherent form. In such cases, the coherent device may have a quantum state that persists without significant degradation over a long period of time, such as on the order of microseconds or longer.

[0917] In particular, quantum entanglement is the physical phenomenon that occurs when a pair or group of particles is generated or otherwise configured to interact in such a way that the quantum state of one particle cannot be described independently of the other, despite the space separating them. Consequently, rather than describing the state of one particle in isolation from the others, a quantum state needs to be described for the system as a whole. In such cases, measurements of various physical properties, such as position, momentum, spin and/or polarization, performed on entangled particles are correlated. For example, if a pair of particles is generated so that their total spin is known to be zero, and one particle is found to have clockwise spin on a certain axis, the spin of the other particle, measured on the same axis, will be found to be counterclockwise, as would be expected due to their entanglement.

[0918] Thus, one particle in an entangled pair simply “knows” what measurement was performed on the other, and with what result, despite there being no known means for such information to be communicated between the particles, which at the time of the measurement may be separated by arbitrarily large distances. Because of this relationship, unlike classical bits which can have only one value at a time, entanglement allows multiple states to act simultaneously. It is these unique entangled relationships and quantum states that have been capitalized upon for the development of quantum computing.

[0919] Consequently, there are several types of physical operations employing pure qubit states that can be performed. For example, a quantum logic gate can be formed and configured to operate on the basic qubit, where the qubit undergoes a unitary transformation, such as where the unitary transformations correspond to rotations or other quantum phenomena of the qubit. Indeed, any two-level system can be used as a qubit, such as photons, electrons, nuclear spins, coherent states of light, optical lattices, Josephson junctions, quantum dots, and the like. Specifically, a quantum gate is the basis for a quantum circuit that operates on a small number of qubits. For example, a quantum circuit is comprised of quantum gates that act on fixed numbers of qubits, such as two or three, or more. Qubits, therefore, are the building blocks of quantum circuits, as classical logic gates are for conventional digital circuits. Specifically, a quantum circuit is a model for quantum computation in which the computation is a sequence of quantum gates that are reversible transformations on a quantum mechanical analogue of an n-bit register. Such analogous structures are referred to as n-qubit registers. Thus, unlike classical logic gates, quantum logic gates are always reversible.

[0920] In particular, as described herein, a digital logic gate is a hard-wired device that can be implemented using one or more diodes or transistors that act as electronic switches to perform logical operations, e.g., Boolean functions, on one or more binary inputs to produce a single binary output. With amplification, logic gates can be cascaded in the same manner that Boolean functions can be composed, allowing for the constriction of a physical model of all Boolean logic, and thus all algorithms and mathematics that can be described with Boolean logic can be performed by digital logic gates. Similarly, a cascade of quantum logic gates can be formed for the performance of Boolean logic operations.

[0921] Quantum gates are typically represented as matrices. In many implementations, a quantum gate acts on k which can be represented by a unitary matrix of 2k x 2k. In such cases, the number of qubits at the input and output of the gate must be equal, and the action of the gate on a particular quantum state is found by multiplying the vector representing the state by the matrix representing the gate. Hence, given this configuration, quantum operations can be performed on a very small number of quantum bits. For example, there are quantum algorithms that are configured to perform much more complex computations faster than any possible probabilistic classical algorithm. In particular, a quantum algorithm is an algorithm that operates on a quantum circuit model of computation.

[0922] Where a classical algorithm is a finite sequence of step-by-step procedures or instructions that can be performed by the digital logic circuits of a classical computer; a quantum algorithm is a step-by-step procedure, each of the steps of which can be performed on a quantum computer. However, although quantum algorithms exist, such as Shor's, Grovar's and Simon's algorithms, all classical algorithms can also be performed on a quantum computer with the right settings. Quantum algorithms are typically used for those algorithms that are inherently quantum, for example involving quantum superposition or entanglement. Quantum algorithms can be referred to in various models of quantum computing, such as the Hamiltonian oracle model.

[0923] Consequently, as the classical computer has a memory composed of bits, where each bit is represented by either a "1" or a "0", a quantum computer supports a sequence of qubits where a single qubit can represent a one, a zero, or any superposition of those two qubit states. Consequently, a pair of qubits can be in any 4-state quantum superposition, and three qubits can be in any 8-state superposition. In general, a quantum computer with n qubits can be in an arbitrary superposition of up to 2n different states simultaneously, which compares to a normal computer that can only be in one of those 2n states at any given time. Therefore, qubits can hold exponentially more information than their classical counterparts. In action, a quantum computer operates by setting the qubits to a bias that solves the problem by manipulating those qubits with a fixed sequence of quantum logic gates. It is this sequence of quantum logic gates that forms the operations of quantum algorithms. The computation ends with a measurement, collapsing the system of qubits into one of 2n pure states, in which each qubit is either a “0” or a “1”, decomposing into a classical state. Thus, traditional algorithms can also be performed on a quantum computing platform, where the result is typically n classical bits of information.

[0924] In standard notation, the basis states of a qubit are denoted as the "0" and "1" states. However, during quantum computing, the state of a qubit can in general be a superposition of the basis or base states, such that the qubit has a non-zero probability of occupying the "0" basis state and a simultaneous non-zero probability of occupying the "1" basis state. Consequently, the quantum nature of the qubit is derived in large part from its ability to exist in a coherent superposition of basis states, and for the qubit state to have a phase. A qubit will retain this ability to exist as a coherent superposition of basis states as long as the qubit is sufficiently isolated from sources of decoherence.

[0925] Accordingly, to complete a computation using a qubit, the state of the qubit is measured. As indicated above, when a measurement of the qubit is made, the quantum nature of the qubit may be temporarily lost and the superposition of the basis states may collapse to the basis state "0" or the basis state "1". In this way, the qubit regains its similarity to a conventional digital "bit". The actual state of the qubit after it has collapsed will depend on the various probability states present immediately before the measurement operation. In this way, qubits can be used to form quantum circuits that can themselves be configured to form a quantum computer.

[0926] There are several general approaches to the design and operation of a quantum computer. One approach that has been presented is that of a circuit model for quantum computation. Circuit model quantum computation requires long quantum coherence, so the type of information device used in quantum computers that support such an approach might be the qubit, which by definition has long coherence times. Consequently, the circuit model for quantum computation is based on the premise that qubits can be formed from and act on logic gates, much like bits, and can be programmed using quantum logic to perform computations, such as Boolean computations. Research has been done to develop qubits that can be programmed to perform quantum logic functions in this manner. For example, see Shor, 2001, arXiv.org:quant-ph/0005003, which is incorporated herein by reference in its entirety. Similarly, a computer processor can take the form of a quantum processor, such as a superconducting quantum processor.

[0927] A superconducting quantum processor may include multiple qubits and associated local bias devices, e.g., two, three, or more superconducting qubits. Accordingly, while in various embodiments a computer processor may be configured as a non-traditional superconducting processor, in other embodiments the computer processor may be configured as a superconducting processor. For example, in some embodiments a non-traditional superconducting processor may be configured to not focus on quantum effects such as superposition, entanglement, and/or quantum tunneling, but may rather operate by emphasizing different principles, such as those principles that govern the operation of classical computer processors. In other embodiments, the computer processor may be configured as a traditional superconducting processor, such as being adapted to process through various quantum effects such as superposition, entanglement, and/or quantum tunneling.

[0928] Accordingly, in many instances, there may be certain advantages to the implementation of such superconducting processors. In particular, due to their natural physical properties, superconducting processors in general may be capable of higher switching speeds and shorter computation times than non-superconducting processors and therefore may be more practical for solving certain problems in superconducting processors. Additionally, details and embodiments of exemplary quantum processors that may be used in conjunction with the present devices, systems, and methods of their use are described in USSNs: 11/317,838; 12/013,192; 12/575,345; 12/266,378; 13/678,266; and 14/255,561; as well as various divisions, continuations, and/or continuations in part thereof; including U.S. Patent 7,533,068; 7,969,805; 9,026,574; 9,355,365; 9,405,876; and all their foreign counterparts, which are incorporated herein by reference in their entirety.

[0929] Additionally, in addition to the above quantum systems and devices, methods for their use in solving complex computational problems are also disclosed. For example, the quantum systems and devices disclosed herein may be employed to control the quantum state of one or more information systems and/or devices, in a coherent manner, in order to perform one or more steps in a bioinformatics and/or genomics processing segmentation, such as for performing one or more operations in an image processing, phase calling, mapping, alignment, classification, variant calling, and/or other genomics and/or bioinformatics segmentation. In particular embodiments, the one or more operations may include performing a Burrow-Wheeler operation, a Smith-Waterman operation, and/or an HMM operation.

[0930] In particular, solving computational problems in genomics and/or bioinformatics using a quantum computing device may include generating one or more qubits and using the same to form a quantum logic circuit representation of the computational problem, encoding the logic circuit representation as a discrete optimization problem, and solving the discrete optimization problem using the quantum processor. The representation may be an arithmetic and/or geometric problem for solution by means of an addition, subtraction, multiplication, and/or division circuit. The discrete optimization problem may be composed of a set of minimal optimization problems, where each minimal optimization problem encodes a respective logic gate from the logic circuit representation. For example, a mathematical circuit may employ binary representations of factors, and these binary representations may be decomposed to reduce the total number of variables required to represent the mathematical circuit. Accordingly, in accordance with the teachings herein, a computer processor may take the form of a digital and/or analog processor, e.g., a quantum processor, such as a superconducting quantum processor. A superconducting quantum processor may include multiple qubits and associated local bias devices, e.g., two or more superconducting qubits, which may be formed into one or more quantum logic circuit representations.

[0931] More particularly, in various embodiments, a superconducting integrated circuit may be provided. Specifically, in particular embodiments, such a superconducting integrated circuit may include a first superconducting current path that is disposed in a metal, e.g., first metal layer. A dielectric layer, e.g., first dielectric layer, may also be included, such as wherein at least a portion of the dielectric layer is associated with and/or charged to the first metal layer. A second superconducting current path may also be included and disposed in a second metal layer, such as the metal layer that is charged to or otherwise associated with the first dielectric layer. In such an embodiment, at least a portion of the second superconducting current path may overlap at least a portion of the first superconducting current path. Similarly, a second dielectric layer may also be included, such as wherein at least a portion of the second dielectric layer is associated with or charged to the second metal layer. Additionally, a third superconducting current path may be included and disposed in a third metal layer that may be associated with or carried in the second dielectric layer, such that at least a portion of the third superconducting current path may overlap at least a portion of one or both of the first and second superconducting current paths. One or more additional metal layers, dielectric layers, and/or current paths may also be included and configured accordingly.

[0932] Additionally, a first superconducting connection may be positioned between the first superconducting current path and the third superconducting current path, such as wherein the first superconducting connection extends through both the first dielectric layer and the second dielectric layer. A second superconducting connection may also be included and positioned between the first superconducting current path and the third superconducting current path, such as wherein the second superconducting connection may extend through both the first dielectric layer and the second dielectric layer. Additionally, at least a portion of the second superconducting current path may be enriched by an external superconducting current path that may be formed by at least a portion of one or more of the first superconducting current path, at least a portion of the second superconducting current path, and/or the first and second superconducting connections. Consequently, in such cases, the second superconducting current path may be configured to couple, e.g. inductively couple, a signal to the external superconducting current path.

[0933] In some embodiments, a mutual inductance between the second superconducting current path and the external superconducting current path may be sublinearly proportional to a thickness of the first dielectric layer and a thickness of the second dielectric layer. The first and second superconducting connections may also include at least one respective superconducting via. Additionally, in various embodiments, the second superconducting current path may be a portion of an input signal line and one of or both of the first and third superconducting current paths may be coupled to a superconducting programmable device. In other embodiments, the second superconducting current path may be a portion of a superconducting programmable device and both of the first and third superconducting current paths may be coupled to an input signal line. In particular embodiments, the superconducting programmable device may be a superconducting qubit that may then be coupled, e.g., quantumly coupled, to one or more other qubits to form a quantum circuit, such as a quantum processing device.

[0934] Accordingly, provided herein are devices, systems, and methods for solving computational problems, especially problems related to solving the genomics and/or bioinformatics hurdle described herein above. In various embodiments, these devices, systems, and methods introduce a technique such that a logic circuit representation of a computational problem can be solved directly and/or can be encoded as a discrete optimization problem, and the discrete optimization problem can then be solved using a computer processor, such as a quantum processor. For example, in particular embodiments, solving such a discrete optimization problem can include executing the logic circuit to solve the original computational problem.

[0935] Accordingly, the devices, systems, and methods described herein may be implemented using any form of computer processor, such as including traditional logic circuits and/or logic circuit representations, as configured for use as a quantum processor and/or in superconducting processing. In particular, various steps in performing image processing, basis calling, mapping, alignment and/or variant calling, bioinformatics segmentation may be encoded as distinct optimization problems and, as such, may be particularly well suited to being solved using the quantum processors disclosed herein. In other cases, such computations may be solved more generally by a computer processor that leverages quantum effects to achieve such computation; and/or in other cases, such computations may be performed using a dedicated integrated circuit, such as an FPGA, ASIC, or structured ASIC, as described in detail herein. In some embodiments, the discrete optimization problem is viewed as a problem by configuring the logic circuits, qubits, and/or couplers in a quantum processor. In some embodiments, the quantum processor may be specifically adapted to facilitate solving such discrete optimization problems.

[0936] As disclosed throughout this specification and in the appended claims, reference is often made to a “logic circuit representation,” e.g., of a computational problem. Depending on the context, a logic circuit may incorporate a set of logic inputs, a set of logic outputs, and a set of logic gates (e.g., NAND gates, XOR gates, and the like) that transform the logic inputs into logic outputs via a set of intermediate logic inputs and intermediate logic outputs. A complete logic circuit may include a representation of the input (or inputs) to the computational problem, a representation of the output (or outputs) of the computational problem, and a representation of the sequence of intermediate steps between the input (or inputs) and the output (or outputs).

[0937] Thus, for various purposes of the present devices, systems, and methods, the computational problem may be defined by its input(s), its output(s), and the intermediate steps that transform the input(s) into the output(s), and a “logic circuit representation” may include all of these elements. Those skilled in the art will appreciate that encoding a “logic circuit representation” of a computational problem as a discrete optimization problem, and subsequently mapping the discrete optimization problem to a quantum processor, may result in any number of layers involving any number of qubits per layer. Additionally, such mapping may implement any inter-qubit coupling scheme to enable any interlayer coupling scheme (e.g., coupling between qubits of different layers) and intralayer coupling (e.g., coupling between qubits within a particular layer).

[0938] Accordingly, as indicated, in some embodiments, the structure of a logic circuit may be stratified into layers. For example, the logic input (or logic inputs) may represent a first layer, each sequential logic (or arithmetic) operation may represent a respective additional layer, and the logic output (or logic outputs) may represent another layer. And as previously described, a logic operation may be performed by a single logic gate or by a combination of logic gates, depending on the specific logic operation being performed. Thus, a “layer” in a logic circuit may include a single logic gate or a combination of logic gates depending on the particular logic circuit being implemented.

[0939] Accordingly, in various embodiments, such as the fact that the logic circuit structure stratifies into layers (e.g., with the logic input(s) representing a first layer, each sequential logic operation representing a respective additional layer, and the logic output(s) representing a further layer), each layer may be embodied by means of a respective set of qubits in the quantum processor and/or superconductor. For example, in one embodiment of a quantum processor, one or more, e.g., each of the array of qubits may be programmed to represent a respective layer of a quantum logic circuit. That is, particular qubits may be programmed to represent the inputs to a logic circuit, other qubits may be programmed to represent a first logical operation (performed by any one or a plurality of logic gates), additional qubits may be programmed to represent a second logical operation (similarly performed by any one or a plurality of logic gates), and still other qubits may be programmed to represent the outputs of the logic circuit.

[0940] Additionally, with multiple sets of qubits representing multiple layers of the problem, it may be advantageous to enable independent dynamic control of each respective set. Additionally, in various embodiments, multiple serial logic circuits may be mapped to the quantum processor, and the respective qubits mapped to facilitate functional interactions for quantum processing in a manner suitable for enabling independent control thereof. From the above, those skilled in the art will appreciate how a similar objective function may be defined for any logic gate. Thus, in some embodiments, the problem representing a logic circuit may be essentially comprised of a plurality of minimal optimization problems, wherein each gate in the logic circuit corresponds to a particular minimal optimization problem.

[0941] Accordingly, exemplary logic circuit representations may be generated using systems and methods that are known in the art. In one example, a logic circuit representation of the computational problem, e.g., the genomics and/or bioinformatics problem, may be generated and/or encoded using a classical digital computer processor and/or a quantum and/or superconducting processor, as described herein. Accordingly, a logic circuit representation of the computational problem may be stored on at least one processor- or computer-readable storage medium, such as a non-transitory computer-readable storage medium or memory (e.g., volatile or non-volatile). Therefore, as discussed herein, the logic circuit representation of the computational problem may be encoded as a discrete optimization problem, or a set of optimization objectives, and in various embodiments, where a classical digital computer processing paradigm is configured to solve the problem, the system may be configured such that bit strings that satisfy the logic circuit have energy of zero and all other bit strings have energy greater than zero, wherein the discrete optimization problem may be solved in a manner so as to establish a solution to the original computational problem.

[0942] Additionally, in other embodiments, the discrete optimization problem may be solved using a computer processor, such as a quantum processor. In such a case, solving the discrete optimization problem may then involve, for example, evolving the quantum processor to the configuration that minimizes the energy of the system in order to establish a bit string that satisfies the optimization objective(s). Accordingly, in some embodiments, the action of solving a discrete optimization problem may include three actions. First, the discrete optimization problem may be mapped to a computer processor. In some embodiments, the computer processor may include a quantum and/or superconducting processor, and mapping the discrete optimization problem to the computer processor may include programming the elements (e.g., qubits and couplers) of the quantum and/or superconducting processor. The mapping of the discrete optimization problem to the computer processor may include the discrete optimization problem onto at least one computer-readable storage medium or processor, such as a non-transitory computer-readable storage medium or memory (e.g., volatile or non-volatile).

[0943] Accordingly, in view of the foregoing, in various instances, there is provided a device, system, and method for performing a sequence analysis segmentation, such as on genomic material. For example, the genomic material may include a plurality of genomic data reads, such as in an image file, BCL, FASTQ file, and the like. In various embodiments, the device and/or system may be employed to perform a sequence analysis on genomic data, e.g., genomic data reads, such as using an index of one or more reference genetic sequences, e.g., stored in a memory, e.g., wherein each genomic data read and each reference sequence represents a nucleotide sequence.

[0944] Particularly, in various embodiments, the device may be a quantum computing device, such as formed by a set of quantum logic circuits, e.g., hard-wired quantum logic circuits, wherein the logic circuits are interconnected to one another. In various cases, the quantum logic circuits may be interconnected by one or more superconducting connections. Additionally, the one or more superconducting connections may include a memory interface, such as for accessing memory. Together, the logic circuits and interconnections may be configured to process information represented as a quantum state that is itself represented as a set of one or more qubits. More particularly, the set of hardwired quantum logic circuits may be arranged as a set of processing engines, such as wherein each processing engine may be formed by a subset of the hardwired quantum logic circuits and may be configured to perform one or more steps in segmenting sequence analysis on reads of genomic data.

[0945] For example, the set of processing engines may be configured to include an image processing, base calling, mapping, alignment, classification, variant calling, and/or other genomics and/or bioinformatics processing module. For example, in various embodiments, a mapping module may be included, such as in a first wired configuration. Additionally, in additional embodiments, an alignment module may be included, such as in a second wired configuration. Additionally, a classification module may be included, such as in a third wired configuration. And, in additional embodiments, a variant calling module may be included, such as in a fourth wired configuration. Still additionally, in various embodiments, an image processing and/or base calling module may be included in additional wired configurations, such as wherein one or more of these wired configurations may include wired quantum logic circuits that may be arranged as a set of processing engines.

[0946] More particularly, in particular instances, a quantum computing system and/or device may include a mapping module, wherein the mapping module comprises a set of quantum logic circuits that are arranged as a set of processing engines, one or more of which are configured to perform one or more steps of a mapping procedure. For example, the one or more quantum processing engines may be configured to receive a readout of genomic data, such as through one or more of a plurality of superconductor connections. Additionally, the one or more quantum processing engines may be configured to extract a portion of the readout to generate a seed, such as wherein the seed may represent a subset of the nucleotide sequence represented by the readout. Additionally, the one or more quantum processing engines may be configured to calculate a first address within the index based on the seed, and access the address in the index in memory in order to receive a record from the address, such as wherein the record represents positional information in the reference genetic sequence. Additionally, the one or more quantum processing engines may be configured to determine, for example, based on the record, one or more corresponding positions from the readout to the reference genetic sequence; and output at least one of the corresponding positions to memory via the memory interface.

[0947] Additionally, the mapping module may include a set of quantum logic circuits that are arranged as a set of processing engines configured to calculate a second address within the index, for example, based on both the record and a second subset of the nucleotide sequence that is not contained in the first subset of the nucleotide sequence. The processing engine(s) may then access the second address in the index in memory in order to receive a second record from the second address, such as wherein the second or a subsequent record includes position information in the reference genetic sequence. The processing engine may be further configured to determine, based on the position information, one or more corresponding positions from the read to the reference genetic sequence.

[0948] Additionally, in various instances, a quantum computing system and/or device may include an alignment module, wherein the alignment module comprises a set of quantum logic circuits that are arranged as a set of processing engines, one or more of which are configured to perform one or more steps of an alignment procedure. For example, the one or more quantum processing engines may be configured to receive a plurality of mapped positions for reading from memory, and to access memory to retrieve a segment of the reference genetic sequence that corresponds to each of the mapped positions. The one or more processing engines formed as an alignment module may be further configured to calculate a read alignment for each retrieved segment of the reference genetic sequence in order to generate a score for each alignment. Additionally, once one or more scores have been generated, at least one best-scoring read alignment may be selected. In particular cases, the quantum computing device may include a set of quantum logic circuits that are arranged as a set of processing engines that are configured to perform a gapped or gapless alignment, such as a Smith Waterman alignment.

[0949] Additionally, in certain instances, a quantum computing system and/or device may include a variant calling module, wherein the variant calling module comprises a set of quantum logic circuits that are arranged as a set of processing engines, one or more of which are configured to perform one or more steps of a variant calling procedure. For example, the quantum computing variant calling module may include a set of quantum logic circuits that are adapted to perform an analysis on a plurality of genomic data reads, such as using one or more candidate haplotypes, e.g., stored in a memory, wherein each genomic data read and each candidate haplotype represents a sequence of nucleotides.

[0950] Specifically, the quantum logic circuitry may be formed as one or more quantum processing engines that are configured to receive one or more genomic data readouts and generate and/or receive the one or more candidate haplotypes, e.g., from memory, such as through one or more of a plurality of superconductor connections. Additionally, the one or more quantum processing engines may be configured to receive one or more genomic data readouts and the one or more candidate haplotypes from memory, and compare nucleotides in each of the one or more readouts to the one or more candidate haplotypes in order to determine a probability that each candidate haplotype represents a correct variant call. Additionally, the one or more quantum processing engines may be configured to generate an output based on the determined probability.

[0951] Additionally, in various instances, the quantum logic circuitry may be formed as one or more quantum processing engines that are configured to determine a probability of observing each readout of the plurality of readouts based on at least one candidate haplotype that is a true nucleotide sequence, e.g., from a source organism of the plurality of readouts. In particular instances, with respect to determining the probability, the one or more quantum processing engines may be configured to execute a hidden Markov model. More particularly, in additional embodiments, the one or more quantum processing engines may be configured to fuse the plurality of readouts into one or more contiguous nucleotide sequences and/or to generate the one or more candidate haplotypes from the one or more contiguous nucleotide sequences. For example, in various embodiments, the fusion of the plurality of readouts includes the one or more quantum processing engines that construct a de Bruijn graph.

[0952] Accordingly, in light of the foregoing, a system is provided for performing various computations in solving problems related to genomics and/or bioinformatics processing. For example, the system may include one or more of an on-site automated sequencer, e.g., NGS, and/or a processing server, either or both of which may include one or more CPUs, GPUs, and/or other integrated circuits, such as including an FPGA, ASIC, and/or structured ASIC, that are configured as described herein to perform one or more steps in a sequence analysis segmentation. In particular, the Next Gen sequencer may be configured to sequence a plurality of nucleic acid sequences in order to generate one or more image, BCL, and/or FASTQ files representing the sequenced nucleic acid sequences, such nucleic acid sequences being a DNA and/or RNA sequence. Such sequence files may be processed by the sequencer itself or by an associated server unit, such as wherein the sequencer and/or associated server includes an integrated circuit, such as an FPGA or ASIC, configured as described herein to perform one or more steps in a secondary sequence analysis segment.

[0953] However, in various cases, such as where the automated sequencer and/or an associated server is not configured to perform secondary sequence analysis on data generated from the sequencer, the generated data may be transmitted to a remote server that is configured to perform secondary and/or tertiary sequence analysis on the data, such as through a cloud-mediated interface. In such a case, the cloud-accessible server may be configured to receive the generated sequence data, such as in the form of image, BCL, and/or FASTQ, and may be further configured to perform an image processing analysis, e.g., primary and/or secondary and/or tertiary processing analysis, such as a segmentation sequence analysis, on the received data. For example, the cloud-accessible server may be one or more servers that include a CPU and/or a GPU, one or both of which may be associated with an integrated circuit, such as an FPGA or ASIC, as described herein. In particular, in certain cases, the cloud-accessible server may be a quantum computing server as described in this document.

[0954] Specifically, the cloud-accessible server may be configured to perform primary, secondary, and/or tertiary genomics and/or bioinformatics analysis on the received data, such analyses may include performing one or more steps in one or more image processing protocols, base calling, mapping, alignment, classification, and/or variant calling. In certain cases, some steps may be performed by one processing platform, such as a CPU or GPU, and others may be performed by another processing platform, such as a tightly coupled associated integrated circuit, such as an FPGA or ASIC, that is specifically configured to perform multiple steps in the sequence analysis segmentation. In such cases, where data and analysis results must be transferred from one platform to another, the system and its components may be configured to compress the data prior to transfer, and decompress the data once transferred, and as such, the system components may be configured to generate one or more of a SAM, BAM, or CRAM file, as required for transfer. Additionally, in various embodiments, the cloud-accessible server may be a quantum computing platform that is configured herein to perform one or more steps in sequence analysis segmentation as described herein, and may include performing one or more secondary and/or tertiary processing steps in accordance with one or more methods disclosed herein.

[0955] Additionally, with respect to quantum computing, details and embodiments of exemplary quantum processors and methods of their use that may be employed in conjunction with the present devices, systems, and methods are described in U.S. Patents 7,135,701; 7,533,068; 7,969,805; 8,560,282; 8,700,689; 8,738,105; 9,026,574; 9,355,365; 9,405,876; as well as the various counterparts thereof, which are incorporated herein by reference in their entireties.

[0956] Additionally, with respect to the artificial intelligence module disclosed above, in one aspect, a cloud-accessible artificial intelligence module is provided, and is configured to be communicatively and operatively coupled to one or more of the other BioIT segmentation components disclosed herein. For example, the A/I module may work closely with the WMS to effectively direct and/or control the various processes of the system disclosed herein. Accordingly, in various embodiments, an A/I module is provided, wherein the A/I module is configured to act as an interface between the genomic world and the clinical world.

[0957] For example, in various instances, the BioIT system may be configured to receive clinical data. In such an instance, the workflow management system may be configured to analyze the clinical and other such data, and implement one or more deterministic rule systems to derive outcome data in accordance with its analysis of the clinical data. For example, in certain embodiments, the various databases of the system may be configured to have a relational architecture.

[0958] These constructs may be represented by one or more table structures. A series of tables, for example, may then be employed through which correlations may be made by the WMS in an iterative fashion. For example, in various usage models, a first correlation may be made against a subject's name with a medical condition. Another table may then be employed to correlate the subject's medical condition with his or her medication. Similarly, an additional table may be used to correlate the medication's progress toward alleviating symptoms and/or the disease itself. A key may be used to correlate the tables, such a key may be accessed in response to a question prompt or command. The key may be any common identifier, such as a name, a number, e.g., a social security number, tax identification number, employee number, a telephone number, and the like, through which one or more tables may be accessed, correlated, and/or a question answered. Consequently, without the key, it becomes more difficult to build correlations between the information in one table with that in another.

[0959] However, in other cases, the A/I module may be configured to provide more comprehensive analysis on generated and/or provided data. For example, the A/I module may be configured to deploy one or more machine learning protocols on system data that are designed to teach the AI module to make correlations between genomic data, e.g., generated by the system, and a clinical deposition of one or more subjects, such as in view of EMR and other clinically relevant data input into the system.

[0960] Specifically, the A/I module may include programming directed at training the system to recognize more quickly, e.g., instantly, how an output was achieved based on the type and characteristics of the input received. The system is therefore configured to learn from the inputs it receives, and the results it outputs, so that it is able to extract correlations more quickly and accurately based on the initial data input received. Typically, the input data may be of two general types. In the first case, the data may be of a type where the output, e.g., the response, is known. This type of data may be input to the system and used for training purposes. The second type of data may be that where the response is unknown and therefore needs to be determined, such data likely being genomic data, upon which analysis is to be done, or clinical data for which clinically relevant results are to be determined. Specifically, these methods can be used to enhance the ability of A/I modules to learn from the first type of input data in order to better predict the outcome for the second type of input data. Specifically, based on historical evidence, the A/I module can be configured to learn to predict outcomes based on previously observed data.

[0961] More specifically, a clinical genomics platform is presented herein, wherein the clinical genomics platform is configured to correlate clinical outcomes of diseases with genomic data. In such a case, clinical profiles of subjects can be input into the system and can be evaluated together with their determined genomic profile. Particularly, upon combining these two data sets, the A/I module is configured to determine the various interrelations between them. Accordingly, in a first step, a graph database or knowledge graph can be constructed. For example, in this case, the knowledge graph can be composed of three typical elements, which basically include a subject, a predicate and an object, which can form nodes, and the relationship between the nodes needs to be determined. Any particular data point can be selected as a node, and the nodes can vary based on the queries that are performed. There are several different types of relationships that can be determined. For example, relationships can be determined based on their effects, e.g., they are effect-based; or they can be determined based on inferences, for example, relationships that are unknown but determinable.

[0962] Accordingly, with respect to the construction of the knowledge graph, any particular data point may form a node. For example, on one side of the graph, a disease condition may form a node, and on the other side of the graph a genotype, e.g., a sequence of variances, may form a node. Between these two nodes may be a third node, e.g., a series of third nodes, such as one or more symptoms, one or more medications, one or more allergies, one or more other conditions or phenotypic traits, e.g., blood pressure, cholesterol, etc. Additionally, between these nodes are relationships that can be determined.

[0963] Specifically, by constructing the knowledge graph, clinical data entered into the system, such as from medical record facilities, e.g., electronic medical records, family history of medical conditions, etc., can be encrypted and securely transferred electronically. Similarly, genomic data from the subject can be sequenced and generated according to the secondary processing steps presented herein. Additionally, once these two nodes have been established, one or more third nodes can be inserted into the system, from the presence of which the relationship (or relationships) between the two original nodes can be determined.

[0964] For example, in one example, a first node may be represented by the medical records of a person or a population of people, and a second node may be represented by a disease characteristic. In such a case, one or more third nodes may be input into the system and generated within the graph, such as where the third node may be a medication; a physical, biological, mental condition and/or characteristic; an allergy; geographic region; diet, a food item and/or ingredient; an environmental condition; a geographic condition; power lines, central towers; and/or the like. A series of relationships may then be determined by analyzing various connection points between these three items. In particular, in a particular case, one node may represent a patient suffering from a disease condition, a second node may be the patient's genomic data, and among the third nodes may be the patient's genomic variations, e.g., the subject's mutations, chromosome by chromosome, his or her medication, physiological conditions, and the like. Similarly, this process can be repeated for multiple subjects who have the same diagnosis and/or condition. Thus, in this way, the correlation between the clinical and genomic worlds can be determined.

[0965] Accordingly, one step in constructing a clinical and genomic graph is to define anchor nodes, which represent the two bounding elements between which all the various similarities are defined and explored. Then, an additional step is to define all the known correspondences between the two anchor nodes, which can be represented in the graph as a third node. These known correspondences can be constructed around the details of the effects caused by and/or characteristics of one node or another. These form the known and/or observable relationships between the nodes. From these known relationships, a second type of relationship can be explored and/or determined, such relationships can be built into inferences. Additionally, to better determine causal and/or predictable outcomes, the various different relationships can be weighted, such as based on degree of certainty, number of similarities, number of cases sharing the node, number of common relationships, and similarities.

[0966] Thus, the construction and deployment of a dynamic knowledge graph is the heart of the genomic and clinical processing platform. As indicated, the various processing platforms of the overall system may be coupled together in order to seamlessly transfer data between its various components. For example, as indicated, mapping, alignment, and/or variant calling segments may be configured to transmit their data, e.g., results data, to the artificial intelligence module. In particular, the A/I module may be configured to receive data inputs from one or more secondary processing platform components and/or one or more other system components. More particularly, the A/I module is configured to receive mapping, alignment, and/or variant calling data from the mapper, aligner, and/or variant calling processing engines, and to take that data and use it to generate one or more nodes within the knowledge graph. Additionally, as indicated, the A/I module may be configured to receive input data from one or more other sources, such as from a physician's office, a health care service provider, a research laboratory, a records storage facility, and the like, such as where the records include data pertaining to the physical, mental, and/or emotional well-being of one or more subjects, and to take such data and use the same to generate one or more nodes within the knowledge graph.

[0967] Additionally, once the knowledge graph architecture has been constructed, it may be continually updated and developed by adding more and more pertinent data to the knowledge structure, building more and more potential relationships and/or nodes. In such a case, the bounding nodes may be any combination of nodes, and as such, in certain cases, may be user selectable. For example, in various embodiments, the system may be configured to be accessible by third parties. In such a case, the user may access the A/I module, e.g., through a suitably configured user interface, upload pertinent information to the system, and/or determine the relevant nodes by which to limit an inquiry, e.g., by clicking or dragging and dropping them, and may formulate a relevant question to be answered by the A/I module. Consequently, the user can analyze and/or select the limiting nodes and then allow the system to generate a suitable knowledge map that employs the selected nodes, and determine the relationships between the nodes, from which various inquiries can be queried and answered, or at least inferred, for example, by the A/I system.

[0968] For example, in one usage model, a user may be a physician who wishes to know how a certain drug dosage is affecting a patient with respect to a certain disease. Accordingly, the physician may upload the patient's EMR, disease condition, and drug dosage, and with this data the A/I module may generate a suitable knowledge graph (and/or add to an existing knowledge graph), from which knowledge graph, boundary nodes may be selected and relationships determined. Additionally, in various cases, the user may upload the patient's genetic data, such data may undergo secondary processing, and the results thereof, e.g., result data mapped, aligned, and/or variant calling, and uploaded to the A/I module. In such a case, the disease and/or EMR and/or family medical history data may be correlated with the genomic data, from which relationships may be determined, inferences evaluated, and predictions made.

[0969] Specifically, a VCF of the subject may be input into the system, e.g., all determined chromosomal properties may be uploaded, e.g., as a constellation of nodes, such nodes may be used to determine various relationships pertinent to the subject, such as by querying the system and allowing the system to generate appropriate connections from which an answer may be inferred. More specifically, one or more phenotypic characteristics of the subject, e.g., the human phenotype ontology, may be uploaded into the system to generate an additional constellation of nodes. For example, when the genomic and/or medical histories of two individuals are input into the system, any relationships between them may be determined by the A/I module, such as with respect to genotypes, phenotypes, conditions, environments, geographies, allergies, ethnic-cultural backgrounds, common medications, and the like.

[0970] Additionally, relationships between two or more features within a subject, or between subjects, may be determined. For example, a relationship between a subject's systolic and diastolic blood pressure may be determined by the system. Specifically, a series of historical systolic and diastolic readings may be input to the system, such that the system's machine learning platform may analyze the readings and/or determine one or more relationships between the two, such that if a given systolic input is input to the system, the predicted diastolic output may be given, taking into account predictive weightings between the two. It should be noted that while the foregoing example is given in relation to blood pressure, within a single subject, the same will apply to any given nodes that are in a mathematical relationship to each other, such as in relation to a multiplicity of subjects and/or a variety of conditions.

[0971] Additionally, while in some cases, relationships may be configured in a linear array, such as to form a neural network of information, in many other cases, relationships may be formed in a multiplicity of stages, such as in a deep learning protocol. For example, the A/I system may be adapted to process information in a layered or multi-stage fashion, such as for the purpose of deep learning. Accordingly, the system may be configured to evaluate data in stages. Specifically, the A/I module may be adapted so that as it examines multiple pieces of data, such as by performing a learning protocol, stage by stage, each connection between data becomes weighted by the system, for example, based on historical features and/or evidence of relationships.

[0972] The more learning stages are initiated within the system, the better the weighting between joins will be, and the deeper the learning will be. Furthermore, uploading the data in stages allows for better convergence of the data within the system. In particular, various feature extraction paradigms may also be employed in order to better organize, weight, and analyze the most salient features of the data to be uploaded. Additionally, in order to better correlate the data, one or more users may input and/or modulate basic weighting functions, although the next system may employ a more advanced weighting function based on the active learning protocols.

[0973] To enable interaction with a user, one or more aspects or features of the subject matter described herein may be implemented in a computer that has a display device, such as, for example, a cathode ray tube (CRT), a liquid crystal display (LCD) monitor, or a light emitting diode (LED) for displaying information to the user, and a keyboard and a cursor device, such as, for example, a mouse or a trackball, through which the user can provide input to the computer. Other types of devices may also be used to enable interaction with a user. For example, the feedback provided to the user may be any form of sensory feedback, such as, for example, visual feedback, auditory feedback, or tactile feedback; and the user input may be received in any form, including, but not limited to, acoustic, speech, or tactile input. Other possible input devices include, but are not limited to, touchscreens or other touch-sensitive devices, such as single-point or multi-point resistive or capacitive trackpads, speech recognition hardware and software, optical readers, optical cursors, digital image capture devices and associated interpretation software, and the like.

[0974] The subject matter described herein may be embedded in systems, apparatus, methods, and/or articles, depending on the desired configuration. The implementations presented in the foregoing description do not represent all implementations consistent with the subject matter described herein. Rather, they are merely some examples consistent with related aspects of the subject matter described. While some variations have been described in detail above, other modifications or additions are possible. In particular, additional features and/or variations may be provided beyond those presented herein. For example, the implementations described above may be directed to various combinations and subcombinations of the features disclosed and/or combinations and subcombinations of various additional features disclosed above. Furthermore, the logic flows depicted in the accompanying Figures and/or described herein do not necessarily require the particular order shown, or sequential order, to achieve desirable results. Other implementations may fall within the scope of the following claims.

Claims (7)

1. A method for improving variant calling accuracy by jointly evaluating reads that map to two or more regions of a reference sequence that are homologous, the method comprising: accessing, via one or more computers, a joint stack of a plurality of sequence reads, wherein the joint stack includes a first stack of reads that have been aligned to a first region of the reference sequence and at least a second stack of reads that have been aligned to a second region of the reference sequence, wherein the first region and the second region are homologous to each other; determining, via one or more computers, a set of candidate variants from the joint stack; defining, via one or more computers, a processing order of the candidate variants; evaluating, via one or more computers, each of the candidate variants from the set of candidate variants based on the defined processing order; generate, by means of one or more computers and based on the evaluation of the candidate variants, a variant call file that identifies one or more of the candidate variants. 2. The method of claim 1, wherein the method comprises: obtaining multiple homologous regions of a reference sequence from one or more memory devices. 3. The method of claim 1, wherein determining a set of candidate variants using the joint stacking comprises: using a De Bruijn graph to extract the candidate variants from the joint stacking. 4. The method of claim 3, wherein the nodes in the graph represent the list of candidates, and wherein using the De Bruijn graph includes generating the De Bruijn graph using each region of the reference sequence as a parent chain and aligning each of the candidate variant positions to the world coordinates. 5. Method according to claim 1, characterized in that defining, by means of one or more computers, a processing order of the candidate variants comprises: defining, by means of one or more computers, a processing order of the candidate variants as a function of the read length or insert size. 6. The method of claim 5, wherein defining a processing order of candidate variants as a function of read length or insert size comprises: generating a connection matrix that defines the processing order of candidate variants as a function of read length and insert size. 7. The method of claim 1, wherein the evaluation, by means of one or more computers, of each of the candidate variants based on the defined processing order comprises: for each candidate variant of the set of candidate variants: generating candidate joint diplotypes, calculating an a posteriori probability of each of the joint diplotypes, computing a genotype matrix, pruning the candidate joint diplotypes, and including a next active position as evidence for a current position.