KR20230082064A - Device and method for target gene selection using natural language and processing - Google Patents

Abstract

The present invention relates to a method and system for selecting a target gene using natural language processing, and specifically, extracts disease, gene, and protein-related terms from one or more bio-field document data, and each As a technology that extracts relationships between terms and uses them for target genetic selection, the first step is to separate sentences into sentences and recognize and extract entity names in conjunction with a database (DB) that stores abstracts or full texts of bio-field literature. ; And, a second step of recognizing the relationship with respect to the entity name extracted from the abstract or full sentence and generating a relationship event to build a pathway; configured to be used for target gene selection by having a natural language processing literature analysis process including do.

Description

Apparatus and method for target gene selection using natural language processing {Device and method for target gene selection using natural language and processing}

The present invention relates to an apparatus and method for selecting a target gene using natural language processing, in order to discover a new gene candidate group for new drug development or to select a gene candidate group that is highly related to a target gene and is judged to have research value. It relates to a target gene selection device and method that can efficiently process literature analysis and related gene information by natural language processing for .

In drug development and gene research targeting specific diseases, it is important to recognize the mutual relationship between genes, compounds, and diseases. The existing molecular biology research method was a method of segmenting and dividing the target for analysis. For example, if you want to study the function of gene A, what role this gene affects cells, for example, subdivide into cell movement, cell death, cell growth, etc., and the gene that this gene A controls It was a way to select only one out of 10 and observe how this subregulatory gene changes the cell's function. It was accepted that a simple combination of these subdivided gene functions could help us understand the function of cells.

However, due to the recent development of molecular biological analysis methods, the process caused by the interaction of multiple genes is known, and the latest technology (e.g., NGS sequencing technology, microarray, mass-spectrometer) that can simultaneously analyze multiple genes, etc. With this development, systems biology, which analyzes using identified genes rather than studying gene functions individually, began to emerge. Systems biology has shifted to a research paradigm in which the function of a gene can only be understood by analyzing the whole rather than simply the sum of the functions of several genes. It is a discipline that aims to analyze and invent replicas using principles. As an example, there is an abnormal increase in the expression of genes associated with cancer development or tumor suppression function in human tumors. Some genes commonly suppress the abnormal expression of tumors, but since there are more cases that show unique genetic patterns depending on the type of tumor, the origin of tumor tissue can be inferred through this. Characteristic gene and protein expression patterns as a cause of gene dysfunction underlying disease occurrence can greatly contribute to reducing social costs in the future by providing opportunities to develop markers necessary for diagnosis, prognosis, and prediction.

Despite the steady progress of genetic research over the past century, even yeast, a simple single-celled organism, has about 1,000 genes whose functions have not been experimentally verified. Considering the pleiotropy of genes, what we currently know about overall gene function is very limited.

On the other hand, among the gene candidates detected by existing gene analysis methods, there may be genes that have already been actively researched by researchers or that have been experimentally verified. Since it is almost impossible to select an effective gene by injecting manpower who can understand the above research, understanding related paper information, and updating the corresponding gene information in a situation where the information is accumulated through such, it is almost impossible to select a target gene candidate. Research is ongoing.

This screening process requires the process of collecting and organizing variously distributed data in order to uncover the relationship between disease and gene or disease and protein in the new drug development stage. Since they are distributed in a form, it requires a lot of effort to quickly identify them and identify the relationship between diseases and genes/proteins. Therefore, in the process of selecting target genes prior to new research and development to identify the relationship between disease and gene/protein, natural language processing from unstructured text included in bio-field document data is performed using a device (eg, computer) equipped with an arithmetic function. It is necessary to develop and graft technology that can perform

An object of the present invention is to provide an analysis method and system for overcoming the disadvantages of the prior art using natural language processing for bio-field literature and effectively predicting unknown functions of genes and outputting gene candidate groups.

In addition, an object of the present invention is to effectively discover a new gene candidate group list for new drug development.

In addition, an object of the present invention is to predict the similarity with a target gene or disease to discover a list of drug candidate genes for the treatment of pathogenic microorganisms.

The technical problem to be achieved by the present invention is not limited to the above-mentioned technical problem, and other technical problems not mentioned can be clearly understood by those skilled in the art from the description below. There will be.

In order to achieve the above technical problem, a method for selecting a target gene using natural language processing according to an embodiment of the present invention is a disease, gene, and protein related term from one or more bio field document data. It is a method of extracting and extracting the relationship between each term and using it for target genetic screening. In conjunction with the database (DB) where the abstract or full text of bio-field literature is stored, it is separated from sentences into sentence units, and individual names are recognized and extracted. step 1; and a second step of constructing a pathway by recognizing a relationship with respect to the entity name extracted from the abstract or full sentence and creating a relationship event.

In addition, in the target gene selection method using natural language processing according to an embodiment of the present invention, the natural language processing literature analysis process generates a multidimensional index for each location of the extracted entity name, and each genetic disease index It may be configured to further include; storing so that it can be seen from a viewpoint.

In addition, in the target gene screening method using natural language processing according to an embodiment of the present invention, the multidimensional index includes genes, diseases, drugs, pathways, compounds, A first dimension selected and configured from proteins; and, gene-gene, gene-disease, gene-drug, gene-path, gene-compound, gene-protein, disease-drug, disease-compound, disease-protein, drug-path, drug-compound, drug-protein, a second dimension selected from pathway-compound, pathway-protein, and compound-protein relationships; It can be configured to include; a third dimension composed of gene-disease-path relationships.

In addition, in the target gene screening method using natural language processing according to an embodiment of the present invention, the entity name is at least one of a disease, a drug and a pathway, a chemical, and a protein. One, and the relationship between the gene and the individual name can be configured to be extracted by dividing it into upregulator, downregulator, indirect-upregulator, and indirect-downregulator.

In addition, in the target gene selection method using natural language processing according to an embodiment of the present invention, the natural language processing literature analysis process is performed by searching and filtering based on keywords of bio field documents, or by determining the impact factor or the number of hits (Hit Searching and filtering based on detailed information of bio field documents including Count); may be configured to filter target documents or genes by further including.

In addition, the method for selecting a target gene using natural language processing according to an embodiment of the present invention includes preparing FASTQ data related to a nucleotide sequence; and performing NGS (next generation sequencing)/DEG (Differential expressed gene) analysis on the prepared nucleotide sequence; A medical statistical analysis step of determining the effectiveness of a gene and a specific disease based on machine learning and extracting a candidate gene related to the specific disease; Gene network analysis step of selecting genes through gene network analysis; It may be configured to further include at least one of them.

In addition, a target gene selection system using natural language processing according to an embodiment of the present invention is constructed using the above-described target gene selection method using natural language processing.

In addition, the program for selecting a target gene using natural language processing according to an embodiment of the present invention is stored and formed in a computer readable recording medium to execute the method for selecting a target gene using natural language processing.

The gene analysis system and method according to the present invention selects a gene candidate group with high research value through an analysis tool using natural language processing without the need to test all genes in the genome to search for a new trait gene, and the selected gene candidate group genes There is an effect that can drastically reduce time, effort, and expense by performing intensive testing on

In addition, there is an effect of reducing time and effort in selecting a candidate for a target gene by more accurately and efficiently selecting a key disease-related gene associated with a biological process in a given sample group.

In addition, the integrated analysis platform is provided in the form of a web, so that users can easily use the analysis service by entering the download address and directly downloading and analyzing from the server computer without installing a separate program or uploading the analysis file. there is.

The effects of the present invention are not limited to the above effects, and should be understood to include all effects that can be inferred from the detailed description of the present invention or the configuration of the invention described in the claims.

1 is a diagram showing the configuration of a natural language processing document analysis unit according to an embodiment of the present invention.
2 is a diagram illustrating detailed shape analysis of individual sentences for natural language processing.
FIG. 3 exemplarily shows types of relationship types for determining relationships such as genetics and diseases.
4 is a diagram showing a web UI/UX screen related to bio-field literature analysis according to an embodiment.
5 is a diagram showing the results of bio-document processing by way of example.
6 is a diagram showing the configuration of an apparatus for selecting a target gene using natural language processing according to an embodiment of the present invention.
7 is a diagram of the process of the present invention for selecting target genes.
8 is a diagram of a method for constructing a machine learning system model for extracting a candidate gene related to a specific disease by the integrated analysis system and method according to the present invention.
9 is a diagram related to a structure for natural language processing of papers, patent documents, and the like.
10 is a diagram of a method for predicting the function of a gene through gene network analysis.
11 is, as an example, a diagram showing a step of selecting a target gene using the integrated analysis system and method according to the present invention with respect to cancer tissue and normal tissue for the purpose of discovering a new gene candidate group for drug development.
12 is a graph of the output of one exemplary survival curve analysis.
13 is a graph of the output of one exemplary ROC curve analysis.

Hereinafter, the present invention will be described with reference to the accompanying drawings. However, the present invention may be embodied in many different forms and, therefore, is not limited to the embodiments described herein.

Throughout the specification, when a part is said to be "connected (connected, contacted, combined)" with another part, this is not only "directly connected", but also "indirectly connected" with another member in between. "Including cases where In addition, when a part "includes" a certain component, it means that it may further include other components without excluding other components unless otherwise stated.

Terms used in this specification are only used to describe specific embodiments, and are not intended to limit the present invention. Singular expressions include plural expressions unless the context clearly dictates otherwise. In this specification, terms such as "include" or "have" are intended to indicate that there is a feature, number, step, operation, component, part, or combination thereof described in the specification, but one or more other features It should be understood that the presence or addition of numbers, steps, operations, components, parts, or combinations thereof is not precluded.

In the case of a conventional genetic analysis system, only the relationship between entities within one sentence is extracted as an analysis method dependent on a text mining method. Although there are cases in which all disease-related factors are listed in one sentence included in document data, it is frequently the case that disease-related factors are described in multiple sentences. In the prior art, even if relationships between entities are extracted and entities are connected according to the extracted relationships, due to limitations in performing natural language processing independently for each sentence, the speed is slow and the accurate meaning of data can be grasped reliably. There were limitations in analyzing the function of .

Accordingly, the present inventors use natural language processing technology using artificial intelligence (AI) deep learning to consider the context of the entire text content, the shape of the word itself, etc. in multiple document data, It overcomes existing technical limitations by deriving the relationship between data and genes and conveying information about extracted genes and their relationships to users.

An apparatus and method for selecting a target gene using natural language processing according to an embodiment of the present invention, when it is time-consuming and costly to research and develop a target gene group that can be verified molecularly, through a natural language-based literature analysis tool. It can be used in the process of selecting effective gene candidates.

Bio field literature targeted in the present invention may include patent literature, thesis literature, and the like, and the thesis literature includes degree thesis literature, conference thesis literature, journal thesis literature, technical reports, and the like. Hereinafter, an analysis of academic research papers as the main target is described by way of example, but is not limited thereto.

The natural language processing-based literature analysis tool is a tool that can understand, automatically classify, and reflect the contents of a database (DB) built with published papers by combining natural language processing technology and neural networks, and is a natural language processing system. ; and a database (DB) built on the basis of published papers, patent documents, and reliable academic journals; and a pipeline that transmits output results to a web page.

The database (DB) is configured by collecting abstracts of papers or patent documents, but it is also possible to collect and configure the full text.

The natural language processing system can understand the content and automatically classify candidate genes by combining natural language processing technology and a neural network based on the database (DB) constructed from published papers and the like. SyntaxNet, BERT, etc. can be used as technologies for natural language processing.

First, the natural language processing system will be described in detail.

Existing natural language processing systems find value and meaningful information by extracting patterns or relationships from unstructured text data composed of natural language based on text mining technology. That is, terms to be recognized as important terms are extracted from the text, and the importance of the extracted term characteristics is shown by calculating the weight function and the support function based on the location and number of occurrences of the word. For example, it is assumed that the importance of a word appearing several times in one document is high, but if the occurrence is high across several documents, the importance of this word is considered low.

The natural language processing system of the present invention goes one step further from the text mining technology and applies a roll-based machine learning technology, performs multi-dimensional analysis on genes or diseases, and applies multi-dimensional indexes from large-scale bio field literature to gene-gene. Effectively extract gene-gene relationship or gene-disease relationship. That is, indexes for diseases, genes, functions, biological pathways, etc. are constructed from large-capacity bio-field literature, and the built indexes are stored according to a predetermined index storage structure, and the user or A computer inputs a search word and analyzes the disease, gene, function, biological pathway, etc. from the large-volume bio field literature, which is a gene-gene relationship or a gene-disease relationship. ) can be extracted.

The overall flow of the natural language processing system in the present invention is as follows. Step 1 of separating sentences into sentences in conjunction with the database (DB) where the abstract or full text of bio-field literature is stored, and recognizing and extracting entity names; Recognizing the relationship between the entity name extracted from the abstract or full text and Step 2 of constructing a pathway by generating a relationship event; generating a multi-dimensional index for each location for the additionally extracted object name and storing it so that it can be viewed from the perspective of each genetic disease index; include

In the first step of separating the abstract or full text sentence by sentence from the database storing the bio-field literature and recognizing and extracting the entity name, the entity name can be extracted from the divided morphemes based on the artificial neural network model. The entity name refers to a noun having a unique meaning, such as a gene name, a disease name, or a chemical formula, in a sentence. Entity name recognition means extracting such entity names from documents and classifying categories. Conventional entity name recognition methods used dictionary-based and rule-based methods, but recently, artificial neural network models based on technologies such as RNN (Recurrent Neural Network) and CNN (Convolutional Neural Network) are being developed. RNN (Recurrent Neural Network) is the deepest network structure among deep learning of input values by stacking data at each moment in an artificial neural network structure. Illustratively, in one embodiment of the present invention, a recurrent neural network (RNN) model, a long short-term memory (LSTM) model, or a BERT model may be used.

In order to perform the first step of recognizing and extracting entity names based on the artificial neural network model, a segmentation unit divides the input text into predetermined units to generate segmented text; a morpheme segmentation unit that generates; an inference unit deriving an inference result, which is data in a vector form, from the divided morphemes; and an object name extraction unit extracting the object name based on the reasoning result and deriving the object name result.

The inference unit may include a first inference unit deriving a first inference result; A second reasoning unit that derives a second reasoning result, which is data in the form of a vector for the entity name, based on the first reasoning result. In addition, the first reasoning unit includes two or more learned artificial neural network models, the first reasoning result includes two or more derivation results derived from the two or more learned artificial neural network models, and the second reasoning unit includes one or more A trained artificial neural network model may be included.

In addition, the artificial neural network model of the first inference unit is a first model that is learned in word units and derives word detail results for input words, and a second model that is learned in parts of speech units and derives parts of speech detail results for input words. models can be included.

Next, the second step of establishing a pathway by recognizing a relationship with the entity name extracted from the database (DB) and generating a relationship event will be described.

Here, pathway means in-depth biological knowledge that expresses the dynamics or interactions between biological elements such as proteins, genes, and cells in a network format.

In the above step, a web search is performed based on the deduction result derived by the entity name extraction unit in the above step 1, documents in which entities appear and locations in cells obtained by analyzing base sequences of entities are collected, and the collected information A relationship event by, for example, a relationship between two entities, a disease related to two entities, a relationship event such as location information of each entity is generated, and based on this, a pathway displaying the corresponding entities at the corresponding place within the cell is created.

For example, a context is extracted using a predetermined context pattern, and a relationship recognition method using the extracted context is an example of an interaction relationship such as 'activate' or 'inhibit' among verbs that appear with high frequency along with gene or protein entity names. It is possible to analyze patterns by extracting event-type verbs and recognize relationships between entities by utilizing the analyzed pattern information.

In order to generate the above pathway, the present invention includes a relationship event generating unit generating a relationship event based on collected information; A pathway construction system including a pathway creation unit that displays corresponding objects at a corresponding place in a cell based on the recognized relationship event and creates a pathway.

Next, a step of generating a multidimensional index for each position of the extracted individual name and storing it so that it can be viewed from the viewpoint of each genetic disease index will be described.

In the present invention, in order to dramatically increase the performance and accuracy of retrieval by utilizing the index without directly accessing the entire sentences of bio-field documents, which increase by several thousand each year, the present invention builds a multi-dimensional index, and prepares the built multi-dimensional index in advance. It is stored according to the defined index storage structure.

For example, in the present invention, the first dimension is composed of genes, diseases, drugs, pathways, chemicals, and proteins, and the second dimension is gene-gene, gene-disease, gene-drug, gene-pathway, gene-compound, gene-protein, disease-drug, disease-compound, disease-protein, drug-route, drug-compound, drug-protein, pathway-compound, pathway- It is composed of protein, compound-protein relationship, etc., and is configured to analyze the 3-dimensional gene-disease-path relationship.

As an example of the first dimension, for the gene index, information on the GEO ID, name, organism, gene expression value, etc. is stored, and the standard gene name and synonym information for the gene name are related to this can be stored

In addition, for the disease index, pubmed ID, sentence number, disease ID and disease name, start position, and end position information are stored, and standard disease name and synonym information for the disease may be related and stored.

For other analysis dimensions, only index information needs to be appropriately added as needed with reference to the above, so that a multidimensional analysis model can be established by easily adding another analysis dimension.

In the present invention, a gene-disease-compound relationship can be extracted through the multidimensional index technology by storing sentences in the database so as to be viewed from the perspective of each gene disease index.

The database (DB) of the present invention is built on the basis of published papers, patent documents, reliable academic journals, etc. and is stored in the memory. It is possible to apply an algorithm that immediately reflects and updates the data of

The update solution structure of the natural language processing database is as follows.

First, a learning dataset is built using the thesis information database built based on the rules. Then, we design a model that searches and classifies the desired feature by taking the Parse tree as an input. When a sentence is input, the model goes through a preprocessing process that removes pre-set elements such as parentheses and redundant spaces, and based on rules, separates sentences according to the nature of conjunctions included in sentences from which unnecessary elements are removed, or sentences to remove some In addition, a learned feature extraction model and its implementation can be calculated through the process of recognizing sentence components such as subjects, verbs, objects, complements, and modifiers constituting sentences in separated or partially removed sentences.

A natural processing analysis device or means for sentences input from bio field literature may include a gene recognition module, a relevance determination module, a relationship derivation module, a connection type classification module, a learning module, and the like, and, if necessary, an input module, It may be configured to further include an output module and the like. (See FIG. 1)

The gene recognition module finds the gene to be analyzed in the thesis database and recognizes it as an entity, and the relevance determination module analyzes the recognized gene entity and sentences in the thesis to determine relevance with the object. The relationship derivation module derives a correlation between genes and objects through sentence analysis based on the results derived from the relevance judgment module. The connection type classification module classifies the relationship between genes and objects obtained by the relationship derivation module by type (see Fig. 2).

In the gene selection device of the present invention, in the input part of the natural language-based search filter, the bio field documents matching the gene list are extracted according to the input object value, and the object is gene, disease, drugs, pathway, chemical, It is configured to select one or more categories among proteins. Through this, related terms such as gene, disease, drug, and pathway are extracted from the bio-field literature data, and a relationship between each term is derived.

The natural language processing method of the present invention analyzes sentences in bio field documents into morphemes and learns words that are considered components that determine the correlation between genes and objects using an artificial neural network model. According to the above process, the gene recognition module and the relationship derivation module are trained in advance and configured to perform relationship derivation from thesis data.

Upregulation, downregulation, and indirect upregulation of the correlation between the gene and the object in the bio field literature extracted using the literature analysis tool constructed with the gene list and the natural language processing-based database indirect-upregulator) and indirect-downregulator. (See Fig. 2)

It can be further subdivided into not, inhibitor, activator, antagonist, agonist-inhibitor, agonist-activator, substrate, product_of, Up to 13 correlations including substrate synthesis product (substrate_product) can be extracted.

For example, the learned artificial neural network model extracts context using the context pattern of the paper and uses the extracted context to perform a relation recognition method among verbs that appear with high frequency along with one gene or protein entity name in the gene list. It is possible to recognize the relationship between entities by extracting event-type verbs representing interactive relationships such as 'activate' or 'inhibit', analyzing patterns, and utilizing the analyzed pattern information. If one gene of the gene list has a correlation with upregulation in relation to a specific disease, it can be inferred that the candidate gene is a gene that is upregulated and expressed in a specific disease. Or, if there is no relation (not), it can be derived that the candidate gene cannot find a sentence related to a specific disease.

Genes in the gene list derived as a result of the search can be determined as genes with research value when they have a correlation with respect to genes and genes, genes and diseases, genes and pathways, and genes and drugs.

In order to strengthen the reliability of the user's use of the gene, the resulting value is provided including the sentence derived from the thesis database that has been determined to have a correlation with the object.

Genes with high matching between the candidate gene and the bio-field literature can be eliminated and extracted from the output unit through the literature analysis tool in which the natural language processing-based database is built, or the related thesis information can be extracted.

In the present invention, it may be configured to provide a download link for referring to bio-field literature related to a corresponding gene through a literature analysis tool in which the database based on natural language processing is constructed.

First, the text collected from the bio field literature database is segmented to create segmented text, and then segmented into morpheme units. In order to remove meaningless parts in sentences, parentheses are removed, and conjunctions, coordinating conjunctions, subordinate conjunctions, and object clause conjunctions of the object clause are separated. The remaining words after removal are divided into sentence form elements, and the subject, verb, object, and complement corresponding to the sentence components of the word are extracted. (See Fig. 3)

In the present invention, a thesis database can be provided. Specifically, it may be an abstract of the thesis, but is not particularly limited thereto. By providing the thesis database in a refined form of the thesis data, which is unstructured data, it is possible for the user to shorten the time required to derive the analysis result value.

First, in the natural language processing thesis analysis system, an artificial neural network model having a structure preset by an artificial intelligence learning module derives relational information of a gene list extracted by analyzing and selecting from a thesis database.

Paper analysis based on natural language processing of the present invention consists of three steps including auxiliary pre-processing (see Fig. 4).

The first step is to use the keyword-based search filter of the thesis, the second step is to search the database with the detailed information-oriented search filter of the thesis, and the third step is to search for correlations within the text of the thesis based on natural language.

In the keyword-based search filter of the thesis in the first step, when the gene list is input in the input unit, the data of the thesis stored in the thesis database and including the keyword of the input gene list can be searched, and the genetic object is obtained from the keyword data of the thesis. It is possible to derive a relationship between them. The input unit supports multi-keyword search for searching genes centered on the thesis keywords extracted from the thesis database. In addition, a search function by setting the period in which the thesis was published is further provided as needed.

As a result of the search, the number of papers including keywords related to the input gene and the most recent representative papers are derived and provided. As detailed information, the name of the paper, the name of the author, and the name of the publishing institution can be presented. In addition, the subject of the thesis is classified into natural science, engineering, and pharmaceutical science, and the type of thesis is divided into academic journals, academic conference materials, professional magazines, and research reports, so that users can easily understand the progress or achievements of research on the gene. let it

In the second-stage paper detailed information-oriented search filter, the number of Impact Factors and Hits of the paper are added to the results of the first-stage search filter, or the importance of the paper related to the gene list to be studied is considered. By screening, the value of the corresponding gene is confirmed and it helps to determine whether to add or delete the final gene. For example, if the paper related to the gene has a high number of Impact Factors and Hits, it is a well-known gene and is not suitable as a gene candidate for predicting new function, so it can be decided to delete it from the final gene. (See Fig. 5)

Thereafter, in the third step, in the natural language-based intra-sentence correlation search, the gene list derived from the analysis and the correlation within the thesis sentence after filtering the thesis abstract are searched through the thesis database. (See Figure 1)

The present invention includes the target gene selection device 10 using natural language processing and the method, as shown in FIG. 6 , including the above-described natural language processing analysis device as one natural language processing literature analysis module. In addition, a web UI/UX-based medical statistical analysis integrated tool that analyzes the overall process of RNA sequencing is provided to perform DEG (Differential expressed gene) analysis using NGS (next generation sequencing) detection data. A nucleotide sequence analysis module that performs; and a medical statistical analysis module that determines the effectiveness of a gene and a specific disease based on machine learning and extracts a candidate gene related to the specific disease; and gene network analysis that selects genes through gene network analysis. module; may be included.

Here, the target gene or candidate gene may refer to all of the candidate genes extracted by performing the analysis system and method according to the present invention as well as the gene to be finally studied.

The web UI/UX-based medical statistical analysis integrated tool of the present invention refers to a system and method that analyzes raw sequence data using an Internet public network and provides output results, and server resources and sequencing results, A pipeline capable of analyzing FASTQ format files and GUI analysis software based on web UI/UX are provided together to enable one-stop analysis and to select meaningful target genes. The FASTQ format here refers to a text-based format for storing both biological sequences (usually nucleotide sequences) and their quality scores. General-purpose tools that analyze and graph FASTQ files are common, but they require a lot of computational resources and memory space to analyze raw sequence data. In the present invention, an analysis that provides a high-capacity FASTQ file together with server resources so that individual users can easily analyze FASTQ files without installing a separate program by providing a server with a built-in pipeline that analyzes FASTQ files without having to meet these resource conditions. systems and methods are provided.

Steps for selecting a target gene according to the present invention include a first step of performing differential expressed gene (DEG) analysis using NGS (next generation sequencing) detection data; 2nd step of analyzing through a machine learning-based medical statistical analysis tool; Step 3 of analyzing through a natural language-based literature analysis tool; 4 steps of analysis with a focused search tool based on gene network analysis; included (see FIG. 7).

First, the first step of performing DEG (Differential expressed gene) analysis using NGS (next generation sequencing) detection data will be described.

The NGS analysis (next generation sequencing) is a method for high-speed analysis of genome sequences and is also called high-throughput sequencing, massive parallel sequencing, or second-generation sequencing. Unlike the existing Sanger sequencing, it is characterized by processing a large number of DNA fragments in parallel. In other words, it is a method of rapidly deciphering vast genome information by disassembling a genome into numerous pieces, reading each piece simultaneously, and then combining them using computational technology. With the advent of next-generation sequencing, the cost required for genome analysis has drastically decreased, and it is widely used in many fields.

The DEG (Differential expressed gene) analysis relates to an analysis method of analyzing the difference in the average expression level of the same gene under different conditions according to a statistical method, and a general calculation method is as follows.

First, the gene expression data are used to obtain sample statistics (mean, standard deviation, etc.). The gene expression data collection may be performed using any method known in the art. For example, microarray gene expression data, multiplex polymerase chainreaction (PCR), quantitative reverse transcription polymerase chain reaction (RT-PCR), transcriptome analysis using tiling arrays, short read sequencing ( short read sequencing), but is not limited thereto. Then, after assuming the distribution of the population through sample statistics, the difference between the experiment / control group is represented by a significant probability (p-value), for example, a gene with a p-value <0.05 is generally determined as a DEG. By analyzing gene expression patterns in this way, it is possible to obtain DEGs (Differentially Expressed Genes) that show high or low expression under specific conditions. That is, the effect of the expression of a specific gene on the expression of other genes can be analyzed.

Although the analysis process for the NGS detection data, the DEG selection, and the medical statistics analysis process require integrated progress, there is still no service that provides information by integrating them, and in the present invention, the analysis efficiency and accuracy are integrated It provides improved effects in terms of, etc.

In the present invention, NGS detection data analysis is performed by extracting DEG (Differential expressed gene) between multiple analysis groups, statistical verification using t-test (p-value < 0.05), Go-term (gene ontology term) and Reactome pathway DB ( A module that can analyze data detected in the NGS (next generation sequencing) process and perform DEG (Differential expressed gene) analysis by providing various gene analysis tools such as pathway analysis using a database for biological pathways provides That is, if the analysis module according to the present invention is used, the analysis server can perform analysis by identifying at least one of gene ontology information or pathway information stored in another DB based on at least one target gene.

Next, the second step of performing the analysis through the machine learning-based medical statistical analysis module will be described.

The above machine learning-based medical statistical analysis module utilizes a running machine-based graph analysis technique or statistical technique such as survival curve analysis, ROC curve analysis, and LASSO regression analysis. It means an analysis tool that can perform medical statistical analysis.

The survival curve analysis is an analysis technique that analyzes and predicts the time taken from a defined starting point to the occurrence of an event. For example, by using the TCGA Data set, a genome open database of more than 10,000 cancer patients in the US, only genes that are statistically correlated with the survival time of cancer patients are selected, and the number of candidate gene groups is reduced from the existing 1000-2000 to 100-200. Can be selected from within or without dogs.

The ROC curve analysis is a graph analysis technique visualized so that evaluation of classification analysis models can be easily compared. The closer the value of AUROC, which represents the area under the ROC curve in the graph, to 1, the better the performance of the model. Using the analysis technique of the ROC curve analysis, it is possible to determine the effectiveness of the corresponding gene and disease diagnosis, so even a clinician can easily and quickly conduct molecular biological research.

The LASSO regression analysis is one of the important variable selection methodologies used in regression models. The complexity of the regression model can be reduced by removing insignificant variables from a large amount of data. can

Next, a method of constructing a machine learning system model for extracting a candidate gene related to a specific disease is as follows.

Using DEG files for diseases for which genetic information is known, the first type of training dataset labeled with upper boundaries (gene data with positive foldchange values) and the second type of training dataset with labeled lower boundaries (negative foldchange values) genetic data with foldchange values) is constructed. Next, after learning a plurality of first models using the first training dataset, the model with the best performance among the plurality of first models is used as a model for calculating the upper boundary, and using the second training dataset After training a plurality of second models, the model with the best performance among the plurality of second models is used as a model for calculating the lower boundary (see FIG. 8). Here, the model for calculating the upper boundary and the model for calculating the lower boundary may be different models.

As described above, in many cases, DEG (Differentially expressed gene) analysis was performed by distinguishing two sample groups using general statistical methods (fold change and t-test) of existing NGS analysis. However, the above analysis technique detects an average of about 1000 candidate genes, making it difficult to selectively identify specific target gene information.

Here, when the machine learning-based medical statistical analysis tool is performed using the present invention, it is determined whether the 1000 to 2000 gene candidates extracted in step 1 have a clear medical relationship with the target gene or related disease group. to about 200 effective candidate genes.

Next, through the natural language processing analysis process, a parse tree is created that structures sentences and composes rules through SyntaxNet, BERT, etc., and identifies and tags parts of speech of words in sentences. carry out Afterwards, build a Rule-based feature extraction pipeline through Parse tree search. Here, the feature can be subject, verb, object, complement, etc. Build a thesis information database (DB) that combines the output results of the constructed pipeline and labeling (misclassification correction by pipeline, correlation coefficient between each feature, direction expression, etc.) (see Fig. 9).

In the present invention, by excluding gene groups whose contents have been previously published from the 100-200 effective candidate gene groups selected in step 2 through the literature analysis tool in which the natural language processing-based database was constructed, a 5- About 20 candidate gene groups can be provided.

Next, the four stages of analysis with the intensive search tool based on gene network analysis will be described.

The gene network analysis is an analysis method for deriving associations between genes or identifying related diseases based on a network analysis technique for gene-gene interactions or gene-disease interactions. For example, to understand the mechanism of disease, such as gene function and expression information, G Protein signal transduction process using systems biology, immune response or cancer, etc. It is possible to identify the inter-relationships between the genes and to predict the unknown functions of related genes.

The usefulness of gene network analysis for predicting the functions of genes has been demonstrated through several studies. If the function of an unknown gene can be predicted, it is highly likely that it will be studied as a new drug target that can treat diseases. Therefore, corresponding analysis is essential when selecting a target gene for new drug development.

The network analysis algorithm for performing the intensive search method based on gene network analysis may utilize a Guilt-By Association (GBA) technique, a network connectivity-based prediction technique, an essential gene-based prediction technique, and the like.

The GBA technique is a method of predicting a new function of a specific gene using previously known functional genes among all genes linked to a specific gene under the assumption that all genes linked to each other in the network will be highly functionally related.

The network connectivity-based prediction technique defines clusters of genes using connectivity in the network, and uses the functions of known genes showing high frequency in each cluster to determine the functions of constituent genes whose functions of the same cluster are unknown. way to predict It is similar to the GBA method, but the GBA method is capable of only supervised learning, while the network connectivity-based prediction method is also capable of unsupervised learning. That is, even if there are no functionally known genes in a defined cluster, it is possible to hypothesize that a given cluster itself represents an unknown functional module by identifying a pattern in a data set.

Next, the essential gene-based prediction technique predicts the essentiality of the gene for an organism using the centrality in each gene network. Since network analysis is one of the graph models, it is possible to leak biological hypotheses using graph theory, and it has been reported through a number of network models that centrality and vitality are highly correlated. (Jeong, H., et al., Lethality and centrality in protein networks. Nature, 2001. 411(6833): p.41-2.).

Discovery of survival essential genes is very important, especially in pathogenic microorganisms, because survival essential genes of pathogenic microorganisms can provide important target genes for the development of new drugs to be used to inhibit the pathogen.

A method of predicting the function of a gene through such gene network analysis is exemplarily as follows.

If two proteins interacting with each other are discovered and either one of them is identified as a cancer-related protein, it is judged that the function of the Unknown protein is likely to be related to cancer (see Fig. 10 (a)) .

As another example, in the case of gene A, its cross-correlation is higher than that of other genes, so its functional importance is predicted to be higher than that of other genes. In addition, if it is known in advance that the function of gene G is related to breast cancer, it can be estimated that the Unknown-1 and Unknown-2 genes are also highly likely to be related to breast cancer. Therefore, in the case of gene A, it is predicted that it is highly likely to be an essential gene for cell survival, and in the case of genes Unknown-1 and Unknown-2, although research has not yet been conducted, it is highly likely to be selected as a candidate gene for new drug development (Fig. 10(b)).

In the present invention, the GBA prediction model, network connectivity-based prediction technique, and prediction of essential genes can be used as a gene network analysis algorithm to select target genes for new drug development for two purposes.

First, a new drug development target candidate gene group for disease treatment is selected using the GBA prediction model. If a gene whose function is known or is associated with a disease is known, it is assumed that the interacting gene is highly likely to be associated with that gene. In addition, the function of a new gene linked to the disease-related gene to be identified is predicted through a network connectivity-based prediction technique, and a list of target genes to be tested with priority is less than 10, preferably around 2-3. .

Second, a candidate gene group for drug development target for the treatment of pathogenic microorganisms is selected using prediction of essential genes. Genes with high centrality are predicted to function as very important genes for survival, and by targeting the genes to develop new drugs, target genes for drug development that can show therapeutic effects against pathogenic microorganisms are selected.

As an example, steps for selecting target genes using the integrated analysis tool according to the present invention with respect to cancer tissues and normal tissues for the purpose of discovering new gene candidates for drug development are as follows. First, the data (FASTQ format) detected through NGS for cancer genes and normal genes are imported into the integrated analysis tool according to the present invention, and then DEG (Differential expressed gene) analysis is performed to select about 2000 candidate genes. select Afterwards, the number of candidate genes is reduced to about 200 through a machine learning-based medical statistical analysis tool. Next, when the analysis is performed using a natural language-based literature analysis tool, the number of candidate genes remains about 20, and the number of candidate genes is analyzed with a gene network analysis-based intensive search tool to select 10 or less target genes (FIG. 11). reference)

The structure of the tool applying the GBA-based algorithm in the present invention is as follows. A web UI/UX is provided so that natural language processing-based database information can be used in the DEG analysis selection process, and interlocks with the natural language processing-based document database and DEG gene database. External databases or internal databases may be used as the literature database and gene database. In addition, genes related to diseases selected by the user among the DEG gene list are tagged. Apply an algorithm that selects the functions of unknown genes linked to tagged genes by applying a clustering algorithm, and transmit the results calculated through the algorithm in conjunction with the network analysis tool. Here, for user's convenience, the corresponding analysis output can be added to the report by linking with the existing PDF-based analysis report.

The structure of the network analysis tool is as follows. A UI/UX that can perform node analysis based on network analysis of the gene entity is provided on the web page, and after expressing the gene entity as a network node, the number of nodes and input data are displayed. Centrality-based survival genes and important gene selection process functions are implemented. Here, a system is constructed so that the DEG analysis data and the natural language processed database are interoperable, and the network analysis tool can be designed directly using a network analysis algorithm, but STRING, Cytoscape, Yeastnet, bioPIXE, Wormnet, etc. can be applied. there is.

The integrated analysis tool according to the present invention can perform gene network analysis in two ways by utilizing a database based on natural language processing. The first is to screen for unknown genes. This is a method of selecting candidate genes based on the assumption that the relationship between genes is less known and the research value increases as the number of papers studied decreases. That is, it is desirable to use it for the purpose of studying those for which the relationship between unexplored genes has not been defined, or when specific characteristics have been identified. The second is to search for and select candidate genes for which functions or relationships between genes have been secured. In the present invention, the relationship between genes is largely divided into 4 cases and subdivided into 13 cases. That is, unrelated genes are filtered out and only related candidate genes are searched and selected. Here, gene selection and classification according to the relationship can be further subdivided or reduced as needed.

In the present invention, a target gene can be selected by selecting an unknown gene using a gene network analysis intensive search tool or by analyzing a candidate gene having a function or relationship between genes.

The four-step process of performing the gene network analysis intensive search tool based on the gene network analysis is as follows. A gene network is created using the data of 5 to 20 candidate genes obtained after completion of the above 3-step analysis. Each cluster is defined by extracting one or more clusters from the generated gene network. Using the functions of known genes showing high frequency connectivity within each cluster, the functions of constituent genes whose functions are unknown within the same cluster are predicted. Through this, it is possible to derive about 10 target gene candidates that are not known but have research value or are highly related to the possibility to be studied.

Next, it will be described in more detail through the configuration of a web page provided as an embodiment according to the present invention. First, the web UI/UX-based analysis tool according to the present invention provides a login screen for linking accounts by inputting user information, and a home screen for creating a project after completing the authentication process. Users can create a project, upload a gene list file and import a gene list. Here, the gene list file may be in a FASTQ file format. Afterwards, the uploaded gene list can be subjected to one-click automatic analysis without user options, or manual analysis can be performed by setting options for each step by the user.

When one-click automatic analysis is performed without a user's option, analyzing through a machine learning-based medical statistical analysis tool; and analyzing through a natural language-based literature analysis tool; Target genes are automatically selected through the step of analyzing with a concentrated search tool based on gene network analysis; and the step of outputting the analysis result. After the analysis is finished, the selected target gene list is output, and the user can check the report to review the analysis result.

When performing manual analysis by setting options for each step, analyzing through a machine learning-based medical statistical analysis tool; and analyzing through a natural language-based literature analysis tool; The process of selecting genes according to the options given in each step, including the step of analyzing with a concentrated search tool based on gene network analysis; and the step of outputting the analysis result, is performed. First, in the step of analyzing through a machine learning-based medical statistical analysis tool, a machine learning model can be selected. The machine learning model may include multidimensional scaling (MDS), survival curve, and regression analysis (LASSO, Ridge, Elastic net). Among the machine learning models, multidimensional scaling (MDS) is an analysis method for removing unnecessary data and dimensions based on finding a low-dimensional coordinate system that maximally preserves the distance of an object in a d-dimensional space.

In addition to this, the description of the present invention described above is for illustrative purposes, and those skilled in the art will understand that it can be easily modified into other specific forms without changing the technical spirit or essential features of the present invention. You will be able to. Therefore, the embodiments described above should be understood as illustrative in all respects and not limiting. For example, each component described as a single type may be implemented in a distributed or divided manner, and likewise, components described as distributed or divided may be implemented in a combined form within the understanding of a person skilled in the art. there is. Further, a method step may be performed alone or multiple times or in combination with at least one other step.

The scope of the present invention is indicated by the following claims, and all changes or modifications derived from the meaning and scope of the claims and equivalent concepts should be interpreted as being included in the scope of the present invention.

Claims (8)

A method of extracting disease, gene, and protein related terms from one or more bio field document data and extracting the relationship between each term to use for target genetic screening,
Step 1 of separating sentences from sentences in conjunction with a database (DB) storing abstracts or full text of bio-field literature and recognizing and extracting object names; and,
A target using natural language processing, characterized in that it is provided with a natural language processing literature analysis process including; a second step of recognizing the relationship with respect to the entity name extracted from the abstract or full sentence and generating a relationship event to build a pathway Genetic screening methods.
According to claim 1,
The natural language processing literature analysis process,
Generating a multi-dimensional index for each location of the extracted entity name and storing it so that it can be viewed from the perspective of each gene disease index; Characterized in that it is configured to further include, target gene selection method using natural language processing.
According to claim 2,
The multidimensional index,
a first dimension selected from Gene, Disease, Drug, Pathway, Chemical, and Protein; and,
Gene-gene, gene-disease, gene-drug, gene-path, gene-compound, gene-protein, disease-drug, disease-compound, disease-protein, drug-path, drug-compound, drug-protein, pathway- a second dimension selected from compounds, pathway-proteins, and compound-protein relationships; A method for selecting a target gene using natural language processing, characterized in that it includes; a third dimension composed of gene-disease-path relationships.
According to claim 3,
The object name is at least one of a disease, a drug and a pathway, a chemical, and a protein,
The relationship between the gene and the individual name is characterized by being extracted by being divided into upregulator, downregulator, indirect-upregulator, and indirect-downregulator, through natural language processing Target gene screening method used.
According to claim 1,
The natural language processing literature analysis process,
Searching and filtering based on the keywords of the bio field literature, or searching and filtering based on the detailed information of the bio field literature including Impact Factor or Hit Count; Characterized in that filtering, a method for selecting a target gene using natural language processing.
According to claim 1,
Preparing FASTQ data for nucleotide sequences; and,
performing NGS (next generation sequencing)/DEG (differential expressed gene) analysis on the prepared nucleotide sequence;
A medical statistical analysis step of determining the effectiveness of a gene and a specific disease based on machine learning and extracting a candidate gene related to the specific disease;
Gene network analysis step of selecting genes through gene network analysis; Characterized in that it further comprises at least one of, a target gene selection method using natural language processing.
A system constructed using the target gene selection method using natural language processing according to any one of claims 1 to 6.
A computer program stored in a computer readable recording medium to execute the target gene selection method using natural language processing according to any one of claims 1 to 6.

Images