WO2024213903A1

WO2024213903A1 - Methods for optimisation of liquid handling processes

Info

Publication number: WO2024213903A1
Application number: PCT/GB2024/050984
Authority: WO
Inventors: Christopher Richard GRANT; Steven Richard Brown; Michael Ian SADOWSKI; Markus Christian GERSHATER
Original assignee: Synthace Limited
Priority date: 2023-04-14
Filing date: 2024-04-15
Publication date: 2024-10-17
Also published as: GB202404764D0; GB202305498D0

Abstract

Methods for implementing a biological or chemical process are provided, wherein the process comprises at least one liquid handling step. The methods comprise establishing a baseline execution strategy for the process; identifying at least one process step, that may involve liquid handling, that may be used within the baseline execution strategy; determining a value for a process or liquid handling metric; and modifying the baseline execution strategy to generate a modified execution strategy for implementing the process by optimizing the value for the process or liquid handling metric. Liquid handling apparatus configured to implement the biological or chemical process according to the described methods are also provided.

Description

METHODS FOR OPTIMISATION OF LIQUID HANDLING PROCESSES

FIELD OF THE INVENTION

The invention is in the field of laboratory automation including systems and methods that incorporate or control liquid handling robotics as well as semi-automated manually operated liquid handling systems.

BACKGROUND OF THE INVENTION

Automation of laboratory protocols, such as within a research and development environment and particularly the multiple liquid handling steps needed in an experiment, offers great promise for design and implementation of complex experiments. Converting scientific experimental laboratory protocols that are typically undertaken manually onto laboratory automation presents many benefits for reproducibility and shareability. However, in practical terms the task is very challenging and, in the past has taken great time and effort requiring a user to have a certain level of technical laboratory automation expertise in order to accomplish the task. Nevertheless, it is through automation that greater standardisation and precision may eliminate human handling errors that contribute to variation in results when performed manually. Improved understanding and implementation of liquid handling, is at the core of recently emerging disciplines such as Computer Aided Biology which facilitate the rapid translation of ideas into results.

Many research scientists are dissuaded from using laboratory automation because of a perception that the skills needed to operate such equipment, including coding, could be beyond them or are time consuming to obtain. For many devices, the expertise required includes how to programme a liquid handler, an understanding of the specific technical process constraints of the liquid handler as well as a knowledge about liquid handler specific concepts such as liquid classes, plate teaching, liquid level detection etc. Additionally, even with some process automation expertise the challenge can be daunting for a conventionally trained molecular biologist or research chemist and can require hours of careful optimisation of liquid handling conditions to suit the particular application or experiment. If such optimisation is not performed, sub-optimal behaviour may result which could cause the experiment to take longer than necessary possibly leading to degradation of samples, lower accuracy and higher consumption of consumables such as pipette tips. As a consequence, such skills remain niche within the wider industry and the efficiency gains promised by the potential of laboratory automation are yet to be fully realised within fields such as biology, chemistry, biotechnology and medicine.

One particular challenge is the translation of a manually performed protocol to an automated liquid handler. For example, instructions for manual protocols are often included in scientific literature, text books or in patent publications. Often it may not be possible to directly translate a manual protocol to automated form, and even if it is, performing such a direct translation may miss out on opportunities to exploit specific capabilities of the liquid handler in order to run the protocol optimally. This results in structural underperformance due to a poor conceptual understanding of the capabilities of the device and how these could be used to the advantage of the researcher. Unfortunately, it is often too much to expect a conventionally trained bench scientist to possess the diverse process engineering skills that may be required to recognise how diverse protocols can be optimised through automation.

Hence, it is of great benefit to be able to rapidly assess various strategies for automated execution of a protocol and either help the user to select an optimal strategy, or to use the process of automation to execute an improved strategy without need for user intervention. Such strategies may also be based on trade-offs between quality metrics relevant to the specific protocol, as well as the device(s) it will be performed on. Hence, there is a need to inform the user of the nature of such trade-offs so that decisions on optimisation can be made in a better way that allows for an improvement in experimental design and execution.

US-20020076818 A1 (Tecan Trading AG) describes a system and method for optimizing liquidhandling parameters for liquid-handling instruments based on automated use of Design of Experiments (DoE) principles. The described approach seeks to optimise liquid handling of an apparatus by defining the liquid class of the liquid being handled.

WO-2021/136932 A1 (Synthace Ltd) describes methods and devices for computer implemented improvement of process performance of automated laboratory protocols that typically require at least one liquid handling step. The methods involve selecting process factors for liquid handling steps in the protocol, wherein the process factors are selected from equipment process factors; liquid process factors; and protocol process factors. Test runs are performed in which parameter variations for the process factors are analysed for their impact on process performance.

EP-3543707 A1 (Tecan Trading AG) describes methods for optimization of liquid classes used by liquid-handling instruments, to reduce the number of processed samples during optimization of a liquid class, and to automate the optimization of liquid classes. The methods involve optimizing the liquid classes iteratively with a genetic algorithm by: applying liquid classes from a set of liquid classes to a laboratory automation device, discarding liquid classes from the set of liquid classes with a selection function and adding liquid classes to the set of liquid classes, which are generated by modifying liquid classes from the set of liquid classes. The genetic algorithm is defined as a method performed by a computer device, which optimizes a set of liquid classes by generating new liquid classes by modifying liquid classes from the set and by discarding liquid classes that are in a sense not as optimal as other ones in the set. The grade of optimization may be computed with a selection function, which is applied to process parameters generated by the liquid class. US-2006/0202922 A1 (Phynexus Inc.) describes methods for optimizing automated processes for extracting an analyte from a liquid sample using DoE principles. Accordingly, various reagent and process parameters for analyte extraction may be varied and tested in parallel to identify which are most optimal.

These and other uses, features and advantages of the invention should be apparent to those skilled in the art from the teachings provided herein.

SUMMARY OF THE INVENTION

In a first aspect the invention provides a method for implementing a biological or chemical process, wherein the process comprises at least one liquid handling step, the method comprising: establishing a baseline execution strategy for the process; identifying at least one liquid handling step that may be used within the baseline execution strategy; determining a value for a metric for the liquid handling step; and modifying the baseline execution strategy to generate a modified execution strategy for implementing the process by optimizing the value for the liquid handling step; wherein determination of the value for the metric comprises either or both of: i. generating a prediction for the liquid handling step by applying one or more in silico rules or models of liquid handling; and/or ii. generating performance quality criteria for the liquid handling step following execution of the baseline execution strategy.

A second aspect of the invention provides a device for executing a laboratory protocol, the device comprising at least one automated liquid handling system, and at least one processor for controlling an operation of the liquid handling system, the processor being configured to perform a method as set out herein.

A third aspect of the invention provides a liquid handling apparatus configured to implement a biological or chemical process according to a method as defined according to any of the embodiments described herein.

A fourth aspect of the invention provides a method for determining one or more stock concentrations for a plurality of liquid reagents intended for use within a plurality of parallel or sequential reactions comprised within a fully or partially automated biological or chemical process, the method comprising: identifying the number of liquid reagents (N) within the plurality of reactions; identifying a first volumetric constraint that defines a total volume (TVol) for each of the plurality of reactions; and identifying a second volumetric constraint that defines a minimum volume (MinVol) for each liquid reagent comprised within the plurality of reactions; wherein the stock concentrations for the plurality of liquid reagents are set according to one or more heuristics that may be determined with reference to at least one of N, TVol and/or MinVol.

In a fifth aspect the invention provides a method for implementing a biological or chemical process, wherein the process comprises at least one liquid handling step, the method comprising: establishing a baseline execution strategy for the process; identifying at least one process step that may be used within the baseline execution strategy; determining a value for a process metric; and modifying the baseline execution strategy to generate a modified execution strategy for implementing the process by optimizing the value for the process metric; wherein determination of the value for the process metric comprises either or both of: i. generating a prediction for the process metric by applying one or more in silico rules or models for the process metric; and/or ii. specifying performance quality criteria for the process prior to or following execution of the baseline execution strategy.

Within the scope of this application it is expressly intended that the various aspects, embodiments, examples and alternatives set out in the preceding paragraphs, in the claims and/or in the following description and drawings, and in particular the individual features thereof, may be taken independently or in any combination. That is, all embodiments and/or features of any embodiment can be combined in any way and/or combination, unless such features are incompatible.

BRIEF DESCRIPTION OF THE DRAWINGS

One or more embodiments of the invention will now be described, by way of example only, with reference to the accompanying drawings, in which:

Figure 1 is a flow chart that sets out one embodiment of a method of the invention

Figure 2 data plots and a model for determining a volume threshold for pipetting, above which no failures are predicted to occur for free dispenses as well as accepting an acceptable bias level. Figure 3 shows models for predicting bias in execution of pipetting strategies, left hand panels show experimental data and tight hand panels show the predictive models.

Figure 4 is python code using the scikit learn library for a Linear Regression Model (RidgeCV)

Figure 5 shows bias models generated using linear (RidgeCV) polynomial order 3, left hand panels show experimental data, right hand panels show the models generated.

Figure 6 is model code using FASTAPI.

Figure 7 shows for liquid handling planner software code.

Figure 8 shows code setting out potential criteria which may be used to choose a suitable liquid handling policy for a given set of candidate conditions.

Figure 9 shows code for various transfer options for liquid handling.

Figure 10 shows code for using quality predictions to choose stock concentrations of source liquids to ensure free dispenses which meet quality requirements are always viable

Figure 11 is python code for a neural net machine learning (ML) model for eliminating bias in automated free dispenses.

Figure 12 shows bias models generated using neural net ML, left hand panels show experimental data, right hand panels show the models generated.

Figure 13 is python code for an autoML model (TPOT) for eliminating bias in automated free dispenses of solutions

Figure 14 shows bias models generated using AutoML (TPOT), left hand panels show experimental data, right hand panels show the models generated

Figure 15 shows probability of failure (<-75% systemic error) models generated using AutoML (TPOT), left hand panels show experimental data (probability of failure), right hand panels show the models generated.

Figure 16 shows a series of plots of experimental data showing evaporation of liquids in different multi-well plates. Figure 17 shows a chart of predicted versus actual results generated using a linear regression model fitted to experimental data (top panel) for predicting evaporation of liquids during liquid handling, and a chart showing the magnitude of a range of factors toward achieving accuracy.

Figure 18 (a) to (c) show illustrations of an experimental protocol to predict homogeneity of solutions based on mixing conditions, liquid composition and diffusion.

Figure 19 shows results of mixing experiments for aqueous solutions comprising 30% (v/v) glycerol and 70% (v/v) glycerol versus water.

Figure 20 shows the results of Figure 19 following conversion to a concentration gradient

Figure 21 shows a linear regression model (RidgeCV) (left hand panel) to predict concentration gradient based upon mix conditions of time and liquid viscosity, and a chart showing the magnitude of a range of factors toward achieving homogeneity of mixing (right hand panel)

Figure 22 shows graphs of the level of systematic error (%) versus the volume of solution transferred for a range of solutions from left to right: bovine serum albumin (BSA), glucose, glycerol, sodium chloride (NaCI) and water.

Figure 23 shows pressure profile graphs over time obtained during pipette aspirate and dispense phases of water and BSA (bovine serum albumin) solutions

Figure 24 shows a machine learning model (left hand panel) to predict volume transferred based upon pressure curve metrics.

Figure 25 shows a neural net generated model to predict precision (CV) for semi-automated pipetting.

Figure 26 shows the experimental data compared to results of a predictive model for predicting CV by volume for liquid types on a Dragonfly dispenser generated using the Ridge modelling methodology.

Figure 27 shows a model generated by Ridge for process metric prediction of molecules of equivalent fluorescein (MEFL) and maximum rate of MEFL production (MEFL/min). DETAILED DESCRIPTION OF THE INVENTION

All references cited herein are incorporated by reference in their entirety. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

Prior to setting forth the specific embodiments of the invention, a number of definitions are provided that will assist in the understanding of the invention.

As used in this description, the singular forms “a,” “an,” and “the” include plural referents unless the context clearly dictates otherwise. Thus, for example, the term “a system” is intended to mean a single system or more than one system or to an assembly comprising a plurality of systems operating in combination. Additionally, any reference referred to as being “incorporated herein” is to be understood as being incorporated in its entirety.

As used herein, the term "systems" also contemplates devices, apparatus, compositions, assemblies, kits, etc., and vice versa. Similarly, the term "method" also contemplates processes, procedures, steps, etc., and vice versa. Moreover, the term "products" also contemplates devices, apparatus, compositions, assemblies, kits, etc., and vice versa.

As used herein, the term "comprising" means any of the recited elements are necessarily included and other elements may optionally be included as well. "Consisting essentially of’ means any recited elements are necessarily included, elements that would materially affect the basic and novel characteristics of the listed elements are excluded, and other elements may optionally be included. "Consisting of’ means that all elements other than those listed are excluded. Embodiments defined by each of these terms are within the scope of this invention.

The term “substantially” refers to the complete or nearly complete extent or degree of an action, characteristic, property, state, structure, item, or result. The exact allowable degree of deviation from absolute completeness may in some cases depend on the specific context, as would be understood to the person of skill in the art. However, in general terms the nearness of conformity to the absolute will be such as to have the same overall result - e.g. functional equivalence - as if total conformity were achieved. For example, when referring to substantially all of a biological or chemical laboratory protocol it may be accepted as meaning a majority, or at least 80%, or 90% or 95% or even 99% of the liquid handling steps defined in that protocol.

The term “execution strategy” is used herein to denote the instructions for a series of experimental stages or steps that may be comprised within a specified chemical or biological process in order to allow for the completion of the process or at least a part of the process. The execution strategy when implemented may comprise sequential process steps, or alternatively may comprise one or more parallel steps, or even a combination of sequential and parallel steps. The implementation of the execution strategy will typically include one or more steps that involve reagent and/or liquid handling that may contribute towards performance of a chemical and/or biological reaction and/or growth of a bio-organism, and/or other process objective. In accordance with at least one embodiment of the invention the execution strategy when implemented may comprise a plurality of liquid handling steps and associated operations that contribute to all or a part of a laboratory protocol. The laboratory protocol may include steps or stages that are either analytic or synthetic in intention.

As used herein the term “liquid handling step” Refers to a discrete stage within a larger process, protocol or execution strategy in which one or more liquids is subject to some form of physical manipulation and/or processing. By way of non-limiting example, manipulation and/or processing can include a step in which liquid media, reagents, buffers or samples are dispensed, aspirated, mixed, or otherwise transformed or transported from one location to another. Processing or manipulation may occur within a single location or container, or the protocol may involve transportation of a liquid from a first location to a second, third, fourth or other locations within a defined physical space. The manipulation and handling of liquid may occur within a defined configuration of space, such as within the geometric boundaries of a multi-well plate. In this instance a first location would be defined as a first well and second or more locations would be defined as other destination wells within the defined geometry of the multi-well plate. Liquid handling would involve aspiration and dispense of liquids and, thus, transference of liquids from one well to another in order to undertake the reactions defined within the specified protocol. In embodiments of the invention, the liquid handling steps occur within a laboratory automation robot or apparatus, or within the area/field of operation of such a robot or apparatus. In embodiments of the invention, the liquid handling steps occur within a plurality of laboratory automation robots or apparatus, or within the area/field of operation of such robots or apparatus. Suitably the plurality of laboratory automation robots or apparatus may be operating cooperatively as part of a larger system or production line.

In embodiments of the present invention, a majority (i.e. greater than 50%) of the liquid handling steps within a process or protocol may occur in the sub-millilitre (e.g. < 1 ml) range, hence, typically a majority of the liquid handling steps within an execution strategy relate to liquid transfers of less than about 1000 pl; suitably less than about 500 pl; and optionally less than about 100 pl. In specific embodiments of the present invention, at least one of the liquid handling steps within an execution strategy of a process or protocol comprises at least one transfer of less than around 50 pl; typically less than around 20 pl; suitably less than around 10 pl; optionally less than around 5 pl. In particular embodiments of the invention, the process is entirely carried out within the bounds of microlitre scale liquid volumes - e.g. all liquid handling steps within a protocol involve transfers of between about 0.1 pl and about 10,000 pl.

The term “liquid” may refer to any non-gaseous fluid material that can be readily subjected to automated physical manipulation and/or processing within a laboratory protocol. Suitably, the liquids subject to processing or manipulation may include biological or environmental samples, buffers, solvents, chemical reagents, biological reagents (e.g. enzymes), solutions, cell or bacterial cultures, culture media, foams, emulsions, suspensions, and ionic liquids. Typically, the various liquids utilised within such protocols will exhibit a range of physical properties, such that different reagents and components comprised with a process or protocol will require handling that is best optimised to the specific physical needs of the liquid as well as to accommodate any other liquid that it may come into contact with/be mixed with. Physical properties of a liquid, therefore, may be considered to include: viscosity (kinetic and dynamic); surface tension; charge; hydrophobicity; conductivity and/or resistivity; volatility; density; stability; temperature; adhesion; cohesion; vapour pressure; and/or sheer sensitivity. Specific parameters associated with these physical properties may also be important considerations such as melting point, evaporation/dew point, flash point, contact angle, and/or glass transition temperature.

The rheological state of a given liquid may be considered especially in relation to the composition of a liquid. Rheology of a given liquid will influence the compatibility of handling steps the liquid will be subject to and the interactions with the various parts of an automated liquid handler and labware involved in such transfers. For example, flow properties will depend upon the composition of the liquid and whether it comprises particles, biological material such as cells (e.g. bacterial or eukaryotic) or vesicular components (e.g. exosomes, liposomes or other emulsion systems). Further whether the liquid is non-newtonian fluid, or a foam/solution/emulsion/suspension, should also be considered as factors that may affect liquid handling and could contribute to error if not accommodated accordingly. One key measure of rheological state relates to the assessment of viscosity and can be assessed as either the dynamic viscosity or the kinematic viscosity of the liquid. Dynamic viscosity, q, can be obtained by multiplying the kinematic viscosity, v, by the density, p, of the liquid (e.g. see ASTM test method D445-03). The SI unit typically used for kinematic viscosity is mm²/s, and for dynamic viscosity is mPa s. Density is a fundamental physical property that can be used in conjunction with other properties to characterize the liquid being handled. Density of a liquid will usually vary according to the temperature. The adhesion (or ‘adhesiveness’) of a liquid refers to the tendency of the liquid to stick to other materials that it comes into contact with. Such materials may include pipette tips or other items of laboratory ware. The adhesive force is typically measured by the work (e.g. J/m²) required to break the adhesive force. The cohesion of a liquid relates to the tendency of the atoms and molecules with the liquid to stick to each other. Cohesion may be measured by determining the work done per unit area required to divide a homogeneous liquid, for example the work required to create droplets from a body of liquid. Adhesion and cohesion are also related to surface tension.

Chemical properties of a liquid that is subject to handing within a laboratory protocol may be contingent upon reactivity, oxidising or reducing properties, radioactive decay, ionic content (e.g. Na⁺, K⁺, Ca²⁺, or Cl- content), pH, or total organic carbon content. Liquids handled within processes, such as chemical or biological processes, may also possess discrete properties that are distinctive and contributory to potential liquid handling variability if overlooked. Such properties may include cellular/optical density of microbial (e.g. bacterial) or eukaryotic cells (e.g. animal, fungal, or plant/plant protoplast cells); biomolecular composition (e.g. nucleic acid, protein, peptide, cytokine or oligo-/polysaccharide concentration); concentration of biomolecules, including metabolites and waste products; properties that can be analysed and are indicative of cell health, cell viability or cell reproducibility; and biopolymer integrity (e.g. integrity of nucleic acid - single-stranded and double-stranded; proteins, polypeptides, polysaccharides etc.).

Examples of liquid handling systems suitable for the performance of automated laboratory protocols may include Freedom EVO (Tecan), Fluent (Tecan), JANUS® (PerkinElmer), Biomek® (Beckman Coulter), Microlab STAR® (Hamilton Robotics) Microlab VANTAGE® (Hamilton Robotics), EpMotion® (Eppendorf), Echo® (LabCyte), Mosquito® (TTP Labtech), OT-1 and OT-2 (Opentrons), LYNX® (Dynamic Devices), PIPETMAX® (Gilson), and Bravo (Agilent). Examples of dispensers suitable for the performance of automated laboratory protocols may include SPT Dragonfly Discovery®, Formulatrix Mantis®, and Thermo Scientific Multidrop. Examples of acoustic liquid handlers suitable for the performance of automated laboratory protocols may include Beckman Coulter Echo Acoustic series liquid handlers. Examples of optofluidic systems suitable for the performance of automated laboratory protocols include Berkley Lights The Beacon®, The Lightening™ and The Culture Station™ platforms

In accordance with embodiments of the invention, the method provides for the bringing together of various steps to help a user or a processor to rapidly assess and/or select an execution strategy from a range of potential strategies based on predictions derived from an understanding of the quality metrics that underpin key steps within a biological or chemical process. Embodiments of the invention comprise generating a prediction for an automated liquid handling step by applying one or more in silico rules or models of liquid handling. These generated predictions may relate to liquid handling metrics that affect performance of, say, pipetting quality and/or resource consumption (e.g. time, pipette tips, reagent quantity). Such predictions may be used to assess and optimise execution conditions for a given scientific protocol. Hence in this way, a user or process controller can improve a baseline execution strategy, one that provides for the least optimised functional approach to mere completion of the process, by selecting a modified execution strategy that is improved in one or more steps so as to optimise a metric associated with the performance of an automated liquid handling step.

In an alternative or additional embodiment, the quality metrics that underpin key steps within a biological or chemical process may be assessed and determined by generating performance quality criteria for the automated liquid handling step following execution of the baseline execution strategy. Hence, performance quality criteria may be generated following the performance of all or a part of a process according to the baseline execution strategy. This approach allows for iterative improvement and further optimisation of an execution strategy.

The modification of the baseline execution strategy may be manifested by ascribing one or more values for a key performance liquid handling metric. Determining how those values may change in different execution strategies allows a user or computer processor to compare the strategies and help direct the selection of modified strategies that favour better experimental choices. In alternative embodiments the process may be fully automated such that the best overall modified execution strategy is selected in accordance with predefined criteria, such as reduction in resource consumption, reduction in time taken for the process to be completed, improvement in yield or accuracy. According to specific embodiments a modified execution strategy is selected that represents a ‘best-fit’ or most acceptable compromise solution within the constraints of the design space available.

In a further embodiment of the invention modification of the baseline execution strategy may be made in relation to one or more process metrics. As used herein the term “process metric” refers to quantifiable measurements of process parameters than may be used to evaluate and/or monitor the the performance of a biological or chemical process, or the performance of one or more steps within such processes. Process metrics serve as indicators of the overall process performance and execution in comparison to an expected or specified output result. Process metrics may encompass various parameters that allow assessment of different aspects of a process. In some instances a process metric may provide insights into how well the process is functioning, identify areas for improvement, and guide decision-making. Some non-limiting examples of process metrics may include:

- Product Yield: The proportion of desired product obtained from starting reagents during a chemical reaction or biological process.

- Conversion Efficiency: A measurement of how effectively reactants are converted into desired products or outputs.

- Throughput: The rate at which products are produced or processed within a given time frame.

- Cycle Time: The time taken to complete the execution of one full cycle of the specified process.

- Quality Metrics: Parameters related to either product or intermediates, including quality, purity, concentration, activity, binding or other specific characteristics.

- Assay Quality: Parameters related to the quality of an assay, for example, Z prime (Z factor), standard deviation, signal to noise ratio, sensitivity.

- Product Activity: parameters related to the activity of the product (e.g. binding or enzymatic conversion)

- Product specificity: Parameters related to the specificity of the product to bind or react to a specific substrate or ligand and avoid binding or activity on off-target substrates and ligands. - Product stability: parameters related to the product stability, i.e. the resistance to degradation and longevity/maintenance of function.

- Cost: Parameters related to the overall cost of the process or surrogate process for which this process is being used as a model.

In specific embodiments of the invention process metrics which may be useful in biological processes, such as bio-assays, may include cell density (g/L); live cell density (cells/ml); recombinant protein expression titre (g/L); rate of protein expression (g/L/h); specific rate of protein expression (g/g/h); protein activity (U/ml); and levels of reporter protein activity, such as fluorescence, luminescence etc. All of these aforementioned process metrics may relate end products/outputs of a given biological process or may also include intermediate products generated at key steps (e.g. milestones) within the process.

It is an advantage of the present invention that the general approach lowers the expertise barrier for researchers and also the time needed to devise successful automated experimental protocols. In certain embodiments, principles that underpin approaches such as Design of Experiments (DoE) may be incorporated into aspects of an in silico rule or model allowing for significant improvements in process performance to be attained especially when compared to, say, an unmodified baseline execution strategy.

A computer implemented, or in silico, model may be based partly or fully on experimental data or mathematical/engineering principles which are applied to a system or apparatus operating the methods of the invention. Hence, in embodiments of the invention a non-transitory computer readable storage medium, comprised of a processor and one or more process control elements, is provided that implements the disclosed methods via laboratory automation. Such models may be used to estimate the expected performance of a given execution strategy that may comprise combinations of metrics, such as liquid transfer; liquid volume; liquid type; and environmental combinations. In embodiments, the models utilise these estimates to firstly select an execution strategy with the best transfer performance at a given volume (referred to as “Auto set liquid class”). In an alternative embodiment, an execution strategy that favours selection of the minimum volume at which the required performance of a free dispense can be achieved. This can be simulated for all transfers required of the liquid(s) in the protocol thereby enabling the system or apparatus to select stock concentrations to ensure only, or a predominance of, free dispenses of liquids will be performed in a given process step, or even for the process as a whole. Since the liquid handling properties will impact the expected performance, where possible experimental data obtained on the accuracy and precision of transfers of a proxy forthe liquid being transferred using various candidate liquid handling related parameters can be used to generate the models used to estimate performance.

In specific embodiments physical properties exhibited by liquids, rather than liquid composition, may be used as model parameters. This can allow for improved generalisation across different liquid compositions for which there could be an infinite number of possible combinations of mixtures. Using this approach the physical liquid properties may be estimated based on literature models (e.g. using a model which precisely estimates viscosity based on volume fraction of, say, glycerol content and at a given temperature). Alternatively, direct empirical models may be used for cases where specific sub components of known importance to a use case are characterised. In one embodiment, the models used in the methods of the invention may be tailored to be specific to a particular automation system set up of a type of apparatus device, or also generated to incorporate steps that require manual or semi manual pipetting. In specific embodiments, the physical properties of the liquids being handled may represent an assay result in themselves, for example, if the properties change from one particular ‘signature’ or ‘profile’ to another during execution of a liquid handling strategy. Metrics based around a change in ‘signature’ or ‘profile’ may be recorded or used to inform the in silico model further.

In an embodiment of the invention an apparatus is configured to perform the methods described herein prior to executing a laboratory protocol that comprises one or more liquid handling steps. In an alternative embodiment the device is configured to perform the methods fully or partially concurrently with the execution of a laboratory protocol that comprises one or more liquid handling steps. In embodiments of the invention the apparatus may include a (computer) system. The system can be configured for engineering compliant communications. The system can comprise one or more processors and one or more non-transient computer-readable storage media. The computer readable storage media can have stored thereon computer-executable instructions that are executable by the one or more processors to cause the computer system to perform the methods and procedures described herein. Hence, it will be appreciated that the present invention may be a system, an apparatus, a method, and/or a computer program product. The computer program product may include a computer readable storage medium (or media) having computer readable program instructions thereon for causing a processor to carry out aspects of the present invention.

Typically, a processor(s) adjusts the laboratory protocol by regulating one or more operations of an automated liquid handling system. In this way process parameters that may contribute to metric values within a process are adjusted. Optionally, the processor adjusts a laboratory protocol by modifying an input reagent requirement/specification, such as by changing parameters that relate to the liquid type or properties, or by changing other liquid handling parameters such as a flow rate or delay time prior to aspiration.

The computer readable storage medium can be a tangible device that can retain and store instructions for use by an instruction execution device, such as an automated liquid handling apparatus or system. The computer readable storage medium may be, for example, but is not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. A non-exhaustive list of more specific examples of the computer readable storage medium includes the following: a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD- ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanically encoded device such as punch-cards or raised structures in a groove having instructions recorded thereon, and any suitable combination of the foregoing. Computer readable storage media may be accessible within a local area network in the form of one or more linked servers, or located remotely in cloud based virtual machines or servers. Cloud based services may be accessed via wired or wireless (wi-fi) telecommunications, such as over the internet. A computer readable storage medium, as used herein, is not to be construed as being transitory signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through a waveguide or other transmission media (e.g., light pulses passing through a fiber-optic cable), or electrical signals transmitted through a wire.

In embodiments of the invention the automated liquid handling apparatus and/or system may comprise one or more of: a pipette; a pipette tip feeder; a plate reader; a plate handling system; a thermocycler; an agitating/vibrational mixer; an aspirator; an ultrasound mixer; an incubator; a chiller unit; and a fluid dispenser.

The concept of a “liquid class” is intended to arrange and describe the information regarding various liquid parameters and properties that are required to ensure high pipetting accuracy for a given type of liquid when using liquid-handling instruments. Typically, the term “liquid class” itself does not necessarily describe the specific liquid properties or characteristics inherent to the liquid being transferred. Instead it refers to the settings adopted by the automated liquid handling apparatus and/or system which are used when a liquid labelled with that class is to be transferred. Although liquid classes are often given names which are descriptive of liquid properties such as “Aqueous”, “Solvent”, “Serum” are often used, the liquid class itself does not necessarily describe the liquid properties or characteristics of the liquid being transferred. For removal of ambiguity, the term “Liquid Type” is used herein as the definition for a descriptive label which is used to describe one or more inherent properties of liquid itself, either based on observed properties (e.g. “quite viscous”, “very viscous”) or via similarity to other archetype liquids which may require specific liquid handling conditions (“BSA-like”) or the presence of key sub-component types (“High Protein Content”). There are multiple parameters that can contribute to the determination of the key metric values that define a liquid type and these may include, physical parameters and/or volumetric constraints.

The term “free dispense” as used herein refers to the process of contact-free dispensing of a volume of liquid by a liquid handling apparatus, such as a pipetting device or robot. Contact-free dispensing of liquid is advantageous because, as the term implies, it allows for a precise volume of liquid to be dispensed into a receiver vessel or reaction mixture without the dispenser apparatus coming into contact with the receiver vessel or reaction mixture. In addition, the process of a free dispense may also ensure that any clinging retention volume within a pipette tip is avoided due to a higher flow rate during the dispense. The combined benefits are such that the free dispense reduces contamination of the pipetting apparatus and, where disposable pipette tips are used, also reduces the consumption of pipette tips during implementation of an execution strategy. Hence, in embodiments of the invention, the modification of a baseline execution strategy may maximise the number of free dispenses employed in a chemical or biological process thereby reducing consumption of resources (e.g. pipette tips), reducing the risk of contamination and allowing faster execution times.

According to embodiments of the invention, a machine learning (ML) approach may be employed to generate algorithms, models and rules that are capable of performing an evaluation of the baseline execution strategy. This may allow, in certain embodiments, a simulation of the baseline execution strategy to be performed without need to perform a wet run. As part of the simulation, metrics values are applied for each liquid handling step based upon information provided in the baseline execution strategy. For example if the baseline execution strategy is obtained from a text source, such as from a publication or literature reference, the metrics value will be based upon the values provided in the text source. Alternatively, if the baseline execution strategy is derived from an automation protocol used in a different system or set up, the metrics value may be imported appropriately. In instances where information regarding one or more of the metrics values are incomplete the baseline execution strategy may be estimated or filled in by reference to one or more standard protocols, or by way of a best guess approach that seeks conformity with known or previously run protocols either in the wider literature or that have been run on the system previously. In the latter case this may require comparison with execution strategies saved within a database or generally within the non- transitory computer memory comprised within, or in communication with, the system or apparatus. If too many process steps or metrics values are missing from the baseline execution strategy this may be communicated to the user appropriately, for example as a null run. In such instances, the system may provide one or more prompts that enable the user to select appropriate metric values to enable the completion of the baseline execution strategy to at least a functional level.

Mere optimisation of chemical or biological processes occurs typically by changing one or at best a few parameters within a protocol. Often, when protocols are performed manually this occurs by intuition of the operator, or to respond to a particular constraint that has been imposed, such as a change in availability of a reagent or presence of a new piece of equipment. However, whilst the drive to perform routine optimisation within the art exists, this is typically highly limited and processes rarely change significantly from their original baseline execution strategy. This is often driven by the so-called parsimony principle, of ensuring minimal changes are made in order to avoid substantial unpredictable departures from the original approach. Whilst the parsimony principle prevents highly unpredictable outcomes, it also hinders the researcher seeking to leverage the full capabilities of laboratory automation. Indeed, the reluctance or technical inability to interrogate chemical and biological laboratory processes at an operational level has tended to result in an under performance of such processes when transferred from manual to automated execution. Given that the great majority of chemical and biological processes are currently recorded in the literature as protocols for manual execution there is potentially a significant shortfall between the levels of process improvement that may be attainable and those that are currently practiced. This is further limited by restrictions placed by the capability of the human mind to even contemplate the multifactorial or multidimensional design space available when utilising laboratory automation. Embodiments of the present invention seek to address this technical challenge through the use of predictive models generated though machine learning. These predictive models can perform accurate simulations of a baseline execution strategy and then undertake modifications of different metrics value within the liquid handling steps in order to propose one or more modified execution strategies.

The modified execution strategies may demonstrate an improvement in one or more key performance parameters compared to the baseline execution strategy. Alternatively, one or more modified execution strategies may be provided that perform less well over one or more key performance parameters compared to the baseline execution strategy. It may sound counterintuitive to provide modified execution strategies that perform less well than a baseline strategy in some parameters, however, often the knowledge derived from how a process or experiment might fail or underperform can be very valuable to a researcher. This is especially so when the information is made available without having to consume resources testing a hypothesis. In addition, there may also be acceptable trade-offs in performance that allows for a given execution strategy to be performed in a less resource intensive manner. By way of example, a modified execution strategy that takes longer to perform compared to a baseline execution strategy may not be considered non- advantageous especially if it utilises fewer reagents, less energy or makes use of off-peak equipment time (e.g. overnight or out of normal working hours). In another example, it may be advantageous to seek a modified execution strategy that seeks to maximise the number of free dispenses, thereby reducing the risk of cross contamination, increasing speed and utilisation of fewer pipette tips. In the latter case it may be acceptable to accept a reduction in experimental precision in order to achieve the stated benefits of the modified execution strategy. Hence, in embodiments of the present invention the user/operator may be provided with a plurality of modified execution strategies in the form of a matrix, in which various value metrics are presented and the effect of the modification indicated. The indication may be in the form of a graph, a colour coding (e.g. traffic light system or heat map), a graphical indicator (e.g. tick, cross or thumbs up/down). Visual comparison of the matrix may allow the user/operator to select the most appropriate modified execution strategy for their needs. Alternatively, the system may be configured to select the best overall execution strategy based upon a set of pre-defined thresholds that correspond to one or more of the liquid handling and/or process metrics, or a best fit compromise solution where appropriate. Hence, the system selection may effectively be automatic; via the action of a processor and without need for human input at the operational level.

Embodiments of the present invention are described herein with reference to flowchart illustrations and/or block diagrams. It will be understood that each block of the flowchart illustrations and/or block diagrams, and combinations of blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer readable program instructions as necessary. These computer readable program instructions may be provided to a processor of a computer or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions/acts specified in the flowchart and/or block diagram block or blocks. These computer readable program instructions may also be stored in a computer readable storage medium that can direct a computer, a programmable data processing apparatus, and/or other devices to function in a particular manner, such that the computer readable storage medium having instructions stored therein comprises an article of manufacture including instructions which implement aspects of the function/act specified in the flowchart and/or block diagram block or blocks.

Improved execution strategies

Figure 1 shows a flow chart that sets out a method for implementing a biological or chemical process according to an embodiment of the present invention. In a first step 101 a baseline execution strategy for the process is identified from a source such as an internal manual protocol, a standard automated protocol (e.g. an OEM protocol for a liquid handler), an external literature or other text source (such as a published patent application). The baseline execution strategy comprises at least one, typically more than one, liquid handling step and in the second step is resolved to a workflow 102 comprising instructions for liquid handling steps comprised within the baseline execution strategy. The liquid handling steps comprised within the workflow 102 will include a range of variable factors that can affect the performance of that step. These variable factors may be considered individually or as a collection of related contributory factors. Irrespective of their individual or collective consideration the factor(s) are defined as metrics (liquid handling or process related) that are associated with the performance of the process as a whole. One or more of the metrics have a first value ascribed based upon the selected value in the baseline execution strategy, this may be defined as the baseline value for the given metric. In the next step 103 this baseline value is subject to interrogation via one or both methods described further below.

A first method that may be employed in the interrogation step 103 comprises generating a prediction for the automated liquid handling step by applying one or more in silico rules or models of liquid handling or process execution to the metric. This predictive modelling approach will typically generate a range of variations of the baseline metric and then test them virtually within a simulation. The simulation may be generated via the use of a machine learning algorithm, or other artificial intelligence based approach, that has been trained with datasets appropriately generated for each metric using real or virtual data. Values for a liquid handling or process metric which are deemed to represent a potential improvement overthe baseline value may be selected, or prioritised, as meeting the requirement for optimization of the value for the metric. The meeting of a requirement for optimization can include falling within predefined thresholds for optimization set in advance either by the system and/or in conjunction with a user. Where a plurality of values exist these may be ranked according to predetermined thresholds and presented to a system operator (user or controller) to allow for a decision on which optimization is most appropriate for the process as shown in step 104. The decision may involve input from the user or, if automated, may follow one or more logic rules, pre-set constraints and/or a decision matrix to allow for the most appropriate selection to be made. If using a decision matrix approach, the user may apply weightings to the decision matrix to ensure that values are selected that best suit the desired criteria for optimization. Without wishing to be limiting, such weightings may include favouring process outcomes that result in any one or more of: increased liquid transfer accuracy; reduced consumption of resources; reduced probability of process failure; improved process quality; reduced execution time; improved precision; improved accuracy; and improved process yield, for example.

A second method that may be employed either instead of or in addition to the first method in the interrogation step 103 comprises generating performance quality criteria for the automated liquid handling step following execution of the baseline execution strategy. This approach involves carrying out all or a part of the baseline execution strategy as a trial run and inspecting performance criteria that are applicable to the metric. The performance criteria may be established by one or more in silico rules that ascertain whether for a given metric the performance is the most optimal or whether it may be optimized further. For a given metric the baseline value may be modified according to the one or more rules in order to generate one or more optimized values. As with the first method, described above, if a plurality of potential optimized values exist these may be ranked according to predetermined thresholds and presented to a system operator (user or controller) to allow for a decision on which optimization is most appropriate for the process as shown in step 104.

A modified execution strategy in which a value for one or more metrics, such as within the automated liquid handling step(s), is optimized is thereby obtained by selecting the most appropriate combination of metric values in step 104. This modified execution strategy may then be implemented as modified biological or chemical process using automation or manual approaches in step 105.

In a specific embodiment of the invention, in step 102 a plurality of values is established for a plurality of liquid handling metrics for each of a plurality of liquid handling steps within the process. In case, the plurality of values for the metric(s) are propagated across the baseline execution strategy to yield summarised metric values for the basic execution strategy. A similar approach may be assumed for process metrics within the specific biological or chemical process. The summarised metric values are presented to a user or system operator (e.g. a processor or controller) in step 104. Thus, the plurality of values are propagated across the modified execution strategy to yield summarised metric values for the modified execution strategy. This allows the summarised liquid handling and/or process metric values for the modified execution strategy to be presented to a user or system operator as a comparison with the summarised metric values for the basic execution strategy in step 104, such as in a comparison matrix. Intermediate or sub-optimal solutions may not be presented to the user or system operator, such that in certain embodiments algorithmic (e.g. genetic programming, and/or multi-objective - i.e. Pareto - optimization) approaches are utilised to iterate towards the best fit solution for the plurality of metrics for each of a plurality of liquid handling and process steps within the process.

In silico rules or models of liquid handling can be generated by performing experiments using a range of liquid handling conditions and equipment. By way of non-limiting example, data sets useful for training machine learning models may be derived from existing automated protocols with one or more additional randomised setpoints introduced, purposely for the purpose of model generation. In addition, targeted experiments may be employed to generate suitable datasets for algorithm training on a range of process parameters that contribute to key performance metrics commonly encountered in chemical and/or biological processes. For instance, experiments may be performed to ascertain the effect of experimental and system parameters selected from:

• Liquid properties/composition - e.g. varying glycerol concentration

• Liquid volumes used - e.g. 0.1-10,000 pL

• Liquid type - e.g. serum content, aqueous solution, organic solvent

• Dispense mode - e.g. Free, Wet, Multi, Dry

• Device class - e.g. Hamilton, Tecan, Perkin-Elmer etc. or manual

• Environmental conditions - e.g. humidity, temperature, atmospheric pressure, vibration

• Number of replicates - e.g. to enable calculation of a % coefficient of variation (%CV)

• Aspirate/Dispense pressure profiles - e.g. maximum pressure difference, time of maximum pressure difference, slope of pressure change during aspiration, coefficient of variance of these metrics between replicates.

• Types of available sensors/assays - presence or absence of a multichannel verification system (MVS) and/or total aspirate and dispense monitoring (TADM)

• Type of hardware available - e.g. types of multi-well plates, pipette tips etc.

Measures that might influence the transferability of a process from one automated liquid handling system to another should also be accounted for. Such measures can contribute to systemic issues that influence the reliability or accuracy of a predictive model and, thus, may be accounted for in rule and model generation. For example, the relative sensor availability and compatibility between devices may have an influence on transferability. Liquid handling systems may incorporate a multichannel verification system (MVS) that facilitates rapid calibration and verification of dispensed volumes. Presence of absence of an MVS may influence the in silico rules or models of liquid handling, for example by making assumptions regarding the %CV, given that more precise liquid handling is expected in the presence of an MVS thereby reducing expected %CV.

In generation of a predictive model there is a need to be aware of and to address any predictive bias that may occur. This may involve establishing rules that prevent or correct formation of any unintended bias, for example by removing outlier results or failed experiments present in training data prior to generating predictive models. The level of correction for bias may vary between metrics, for some metrics it may be more prevalent (e.g. pipetting of liquid volume, or liquid type and/or class) and for other metrics it may be less so. In instances where baseline execution strategies are based upon protocols that are derived from text sources in the wider literature, there may be a risk of perpetuation of an inherent bias present within the wider art. Hence, it is an advantage of the present approach that by use of independently generated training data, the influence of wider historical biases is minimised or even eliminated.

Various in silico modelling techniques can be used to generate algorithms that will analyse and perform variations of metric values that are used in the baseline execution strategy.

In a specific embodiment, a widely used class of machine learning algorithms involves simple linear models. Linear models are some of the most straightforward to use in machine learning approaches and make a prediction by using a linear function of the input features. Known linear models may include linear regression, linear regression (ordinary least squares), ridge regression, Lasso and polynomial regression. In common with all linear regression models is the need to consider the given data points (the training data) and plot a best fit line to fit the model in the best way possible and to thereby allow predictions to be made with a high level of accuracy. Regression techniques, such as those described above and that are more widely known in the art, may be used to generate a range of in silico rules and models based upon training data sets comprising multiple metric values. Linear regression models are particularly useful for extrapolation, where there is a need to estimate values beyond the observational range provided within a training data set.

In an alternative embodiment of the invention, the machine learning models utilise a neural network approach. A neural network is a model containing an interconnected group of processing elements or "neurons" that process information using a connectionist approach to computation. Neural networks are often used to model complex relationships between inputs and outputs or to find patterns within data. Typically, neural networks process data in a non-linear, distributed, parallel fashion. Often a neural network is an adaptive system that changes its structure during a learning phase. Functions are performed collectively and in parallel by the processing elements, rather than there being a clear delineation of subtasks to which various units are assigned. Generally, a neural network involves a network of simple processing elements that exhibit complex global behaviour determined by the connections between the processing elements and element parameters. Neural networks may be used with algorithms designed to alter the strength of the connections in the network to produce a desired signal flow. The strength, also known as a weighting, is altered during the training or learning phase.

A further embodiment of the invention provides for the use of decision tree algorithms in the creation of in silico models and rules. In particular, a random forest approach comprises a supervised machine learning algorithm that is constructed from decision tree algorithms. This algorithm is suitably used to predict behaviour and outcomes in a given set of circumstances. The term ‘random forest’ refers to the use of a combination of classification tree predictors such that each tree depends on the values of a random vector sampled independently and with the same distribution for all trees in the so-called ‘forest’. A random forest is a learning ensemble consisting of a bagging of un-pruned decision tree learners with a randomized selection of features at each split of the decision tree. Where ‘bagging’ is an ensemble meta-algorithm that improves the accuracy of the machine learning algorithm. A random forest grows a large number of classification trees, each of which votes for the most popular class. The random forest algorithm establishes the prediction outcome based on the predictions of the decision trees. It predicts by taking the average or mean of the output from various trees in the forest. Increasing the number of trees in the forest, thus, increases the predictive power of the algorithm. Random forest algorithms are useful for predictive accuracy within a rich dataset and are suitable for both regression and classification tasks.

In addition to the above approaches, the process of in silico algorithm and rule generation may be enhanced by adoption of an automated machine learning approach (autoML). AutoML is the process of automating the process of applying machine learning to real-world problems, such as optimizing the value of one or more metrics in an automated liquid handling step. AutoML was proposed as an artificial intelligence-based solution to the ever-growing challenge of applying machine learning. An advantage of the high degree of automation in autoML is that it can allow non-experts to make use of machine learning models and techniques. Typically, during model training, autoML can create a number of parallel pipelines that try different algorithms (such as any of the approaches described above) and parameters, thus, iterating through ML algorithms paired with feature selections. Each iteration produces a model with an associated training score. The better the training score for the metric that is to be optimized for, the better the model is considered to ‘fit’ the data.

In an embodiment of the invention, an autoML approach is used to general in silico models for metrics that contribute to an automated liquid handling step in a biological or chemical process having at least one automated liquid handling step. Tree-Based Pipeline Optimization Tool (TPOT) is one autoML package in python that uses genetic programming concepts to optimize the machine learning pipeline. Genetic programming concepts referto the use of principles of natural selection to generate an optimized search space. The TPOT approach is described in Olson et al. Evo. Applications (2016): Applications of Evolutionary Computation pp 123-137.

In accordance with a specific embodiment of the present invention, liquid handling metrics for at least one liquid handling step in an automated chemical or biological process may be comprised of equipment factors; liquid factors; and/or protocol factors. A value is identified for each factor in order to facilitate either or both of (i) generating a prediction for the automated liquid handling step by applying one or more in silico rules or models of liquid handling; (ii) generating performance quality criteria for the automated liquid handling step following execution of the baseline execution strategy. Hence, the factor represents the process variable which contributes to a metric that may comprise a range of different values. The values of these metrics may thereby be modified for more optimal or improved overall process performance according to the methods of the invention.

Suitably, the equipment factors may be selected from the group consisting of: an automation factor; a pipetting factor; a pipette tip factor; a dispensing factor; and a containment factor. Automation factors may be selected from the group consisting of: make and/or model of liquid handling apparatus; configuration of liquid handling apparatus; and setup of liquid handling apparatus. Whilst, pipetting factor may be selected from the group consisting of: aspirate speed; dispense speed; mix speed; waiting time; excess aspirated volume; excess dispensed volume; aspirate/dispense position relative to the container/liquid being addressed (e.g. free, wet, or dry dispense); tip movement speed; blowout choice; pre-wet of tip; number of tip uses; pre-mix of liquid source; post-mix of liquid destination; liquid source volume; and liquid source depth. The pipette tip factor may be selected from the group consisting of: pipette tip size; pipette tip capacity; presence or absence of filter; fixed or removable tip; conductive properties of tip material; selection of tip material; bore size; tip make; tip coating; tip geometry; and batch number. Typically, the dispensing factor may be selected from the group consisting of: amount of liquid in destination well; type of liquid in destination well; force of dispense; choice of pulsed dispense or continuous dispense; duration of dispense; and selection of acoustic or physical dispense. Whereas the containment factor may be selected from the group consisting of: containment properties; source container geometry; destination container geometry; source container material; destination container material; and destination container capacity.

The liquid factors, that contribute to the determination of liquid class, are suitably selected from the group consisting of: liquid type; solute concentration; viscosity; surface tension; volatility; density; adherence; stability; evaporation; charge; hydrophobicity; homogeneity; rheology; vapour pressure; liquid temperature; sheer sensitivity; shear stress; hydrostatic pressure change as the liquid is aspirated/dispensed; and miscibility. More specifically, in certain embodiments the liquid type is selected from the group consisting of: aqueous solvent; non-aqueous solvent; biological medium; emulsion; particulate suspension; cell culture suspension; serum-containing medium; protein-in- solution (e.g. BSA); volatile solvent; nucleic-acid in solution; and viscous solution. The stability factor may be selected from the group consisting of: temperature lability; light lability; chemical stability; and biochemical stability. The chemical liquid factor may be selected from the group consisting of: pH; ion content; total organic carbon content; solute identity; and radioisotope content. Whilst, the biological liquid factor may be selected from the group consisting of: cell survival; cell density; cell stability; cell health; biomolecular composition; biomolecular concentration; cell health; and biopolymer integrity.

The protocol factor may be typically selected from the group consisting of: an environmental factor; an agitation factor; and a timing factor. The environmental factor is optionally selected from the group consisting of: environmental temperature; environmental humidity; barometric pressure; atmospheric circulation; atmospheric flow rate; electromagnetic radiation exposure levels; and type of electromagnetic radiation. The agitation factor may be selected from: presence or absence of agitation; type of agitation; and amount of agitation. Whilst, the timing factor may be selected from the group consisting of: timing of protocol; presence or absence of time delay between process steps; length of time delay between process steps; and number of time delays between process steps.

The above factors are not limiting and represent a collection of process parameters that may be defined as liquid handling performance metrics. It will be appreciated that further granular inspection of potential process factors (such as from process metrics) may identify previously unknown contributory factors that affect performance. In addition, it is possible that metrics may be comprised of second or even third order factors that are defined by multifactorial interactions. If a metric can be defined and a variable value ascribed to that metric that can be tested via the methods described herein then the potential for modification of that value to enable more optimal performance of the process exists.

In a specific embodiment of the invention, the method includes providing a prediction of a range of values for a liquid handling performance metric that allows for the system to set the liquid handling behaviour without the need for user input e.g. without requiring user discretion. This approach may require establishing a default maximum coefficient of variation (max % CV) threshold value for each liquid handling step and also a threshold maximum probability of failure for that step. In addition or as an alternative, setting the liquid handling performance metrics may require establishing a default maximum threshold for bias/inaccuracy. These thresholds define the range of possible transfer options available which can be evaluated by the system. The modelled solution can be prioritised in order of desirability and as soon as a solution meets the predefined thresholds that establish the quality criteria, that solution is the option is identified as the most optimal and executed accordingly. In addition, the prioritisation of a range of options may be cross referenced with another factor. For instance, it may be desirable to cross reference the range of solutions provided with a resource consumption based process metric, such as cost, reagent consumption or time taken. This additional metric can then be used to sort the various transfer options accordingly and identify the most optimal values. In circumstances where time and/or resource consumption are not priorities, a further embodiment provides that the liquid handling performance metrics selected are those which are predicted to provide the best accuracy/highest precision.

In yet a further embodiment, the method includes providing a prediction of a range of values for a liquid handling performance metric that allows for the system to set the most optimal stock concentrations, without need for user discretion or input. This may be developed further into providing that a baseline execution strategy is modified to optimize the values of one or more metrics for the automated liquid handling step to increase the number of free dispenses, CV, precision etc. This modification of values may apply to automatically adjusting reagent liquid concentration and thus the volume of liquid that is transferred in a liquid handling step. In a specific embodiment of the invention, the method provides for monitoring of the hydrostatic pressure change within a liquid handling device, suitably within the pipette tip or head, as the liquid is aspirated/dispensed. The data associated with a change in pressure in the pipette head/tip during the aspirate and dispense steps may be used to generate a pressure profile model for the liquid type being handled. This enables associated metrics to be used as model features correlated with the liquid properties. Hence, the method provides for model simulations to be used to develop pressure curves that act as signatures for particular liquid types and allow for improved predictions of accuracy, CV, precision etc. during liquid handling steps within a protocol.

Devices and apparatus may be configured to operate the methods of the invention as described herein. Such devices may be existing liquid handling systems, such as those set out above. In embodiments of the invention the device comprises or is comprised within a laboratory pipetting robot. Suitably the device comprises an automated liquid handling system selected from the group consisting of: a dispenser; an acoustic liquid handler; and an optofluidic liquid handler

In an alternative embodiment the invention provides a guided semi-automated pipetting system comprising of a tablet/computer, or other GUI incorporating device, that is capable of informing a user where and how to transfer liquid to for each step. This may consist of an electronic “connected” pipette which may set the volume and flow rate automatically for each step, or a manual pipette with the tablet/computer providing one or more guiding instructions to the user on the recommended approach of transferring the liquid being handled from a first receptacle (i.e. the origin) to the destination receptacle. Hence, in this embodiment all of the above-described methods are implemented manually under the guidance of a system.

Selection of optimal stock concentrations

In laboratory liquid handling procedures, whether executed manually or via automation, the selection of appropriate stock concentrations for reagents can have profound effects on the precise and efficient execution of experiments. These stock concentrations serve as the basis for calculating and preparing the desired volumes and concentrations of solutions required for the mixtures used as the basis for various reactions and assays. By establishing standardized stock concentrations, laboratories can optimize workflows, minimize errors, and ensure improved precision of liquid handling across experiments. Automated liquid handling systems, for example, will use the stock concentrations as input parameters to enable accurately dispensing of specific volumes of reagents, thereby facilitating the reproducibility and scalability of experiments. Additionally, maintaining stock concentrations allows for better inventory management and cost-effectiveness by reducing wastage of reagents. Overall, precise control and utilization of stock concentrations are fundamental to the success of automated laboratory liquid handling procedures, enabling high-throughput experimentation and reliable results. Techniques such as DoE are known to be used to investigate the effects of a particular composition mixture for a given application. By way of example, such mixture compositions will vary depending upon the type of assay, synthetic process or reaction. Mixtures may include a proportion of solvent, as well as various reactants, substrates, analytes, samples and also buffering agents, stabilizers etc. A typical reaction carried out on a 96 well plate will require a plurality of such mixtures, i.e. one for each well (or ‘run’), as well as for any control runs. Hence, in such experiments, processes or reactions each of these mixtures comprises various combinations of components. A starting concentration is, thus, required for each of the mixture components in order to enable the reaction, assay or process to be executed. Starting concentrations are required to be within certain operating tolerance ranges that define the ability of the mixture to attempt to execute its intended function effectively.

It is conventional for mixture components to be assembled from a range of stock solutions. The stock solutions of each of the required components can be created and added in different proportions in order to achieve a given set point for that mixture. The ‘set point’ is the value of the range of concentrations for components within a mixture around which variations within a defined range can be made. In embodiments of the invention, the defined range of variation around the set point establishes the upper and lower limits (i.e. upper and lower set points) of possible concentrations within which the assay, process or reaction for which the mixture is intended remains executable.

Executing an experiment where liquid mixtures are required can be challenging when the mixtures are composed of several components which need to be combined. Given that reaction volumes are often constrained - e.g. using 96, 384 or 1536 multiplate well formats - the more liquid components that are required in any single mixture, the less volume is available per liquid and therefore the average stock concentrations required are higher to achieve the target set points with less volume added per component. Hence, stock concentrations may be made as high as possible in order to minimise volume per component in the target. However, the trade-off comes in the form of accuracy, as with the smaller transfer volumes of higher stock concentration components during liquid handling typically comes a reduction in accuracy and/or precision. In some instances, theoretical stock concentrations may even fall below the lower dispense volume limit for automated or manual liquid handling. Additionally, there are conceivable upper thresholds for the concentration of stock solutions, dictated by factors like solute solubility or pragmatic considerations such as availability or pre-dilution from suppliers. Moreover, heightened concentrations of certain solutes can alter the solution's physical properties, notably increasing viscosity, which in turn affects pipetting accuracy and necessitates adjusted pipetting techniques, such as employing different liquid classes. This phenomenon is particularly evident with common components like proteins, cells, DNA, glucose, and glycerol. Furthermore, when conducting parallel experiments with varying target set points spanning orders of magnitude, it might be imperative to maintain multiple liquid stocks for each subcomponent. However, excessive stock diversity can prove cumbersome to manage practically, necessitating careful consideration to balance efficiency with experimental requirements.

Sub-optimal or inaccurate stock concentrations of reagents can impact reaction performance in laboratory liquid handling procedures. Deviations from the intended concentrations or in their delivery can lead to inaccuracies in the final solution compositions, affecting the reliability and reproducibility of experimental results. Suboptimal concentrations or delivery may result in incomplete reactions, skewed reaction kinetics, or erroneous data interpretation, ultimately compromising the validity of experimental findings. Moreover, variations in stock concentrations from run-to-run can introduce inconsistencies in experimental outcomes across different batches or laboratories. Such discrepancies can impede the identification of trends or patterns, hindering progress and leading to wasted resources. Therefore, maintaining precise and optimal stock concentrations is paramount in liquid handling reactions to ensure the integrity and robustness of scientific investigations.

Hence according to an embodiment of the present invention, a method is provided to set at least one stock concentration for at least one liquid reagent comprised within a liquid handling protocol for a biological or chemical process. It is of considerable advantage for the user or operator of the liquid handling protocol to (a) to reduce complexity; (b) improve accuracy, speed and reduction of consumption of disposables; and/or (c) reduce the number of liquid reagent stocks required to as few as possible. Suitably liquid handling protocol comprises at least one automated liquid handling step, optionally substantially all of the liquid handling protocol comprises automated liquid handling steps. In one specific embodiment, all stock concentrations for reagents comprised within a liquid handling protocol for a biological or chemical process are defined according to the process of the invention. In a specific embodiment, the most concentrated stock is selected which is predicted to result in acceptable pipetting behaviour. Acceptable pipetting behaviour may be determined by comparison to a baseline execution protocol, using factors such as:

• Lower % coefficient of variation (%CV)

• Higher number of free dispenses

• Lower probability of failure - e.g. pipetting failure

• Fewer number of stocks required

• Less diluent required

In the context of establishing appropriate operating parameters or selecting threshold values for a laboratory processes, a so-called ‘heuristic’ refers to a practical, experience-based approach that relies on guidelines or ‘rules of thumb’ rather than strict mathematical or theoretical principles. Heuristics in this context are used to help the user or operator to make decisions about key threshold values by considering factors such as past experience, empirical observations, and practical constraints. Thus, by selecting heuristics a user can enable systems to provide quick, efficient solutions that are often sufficient for achieving desired outcomes without requiring exhaustive computational analysis or over-optimization. Heuristics can streamline decision-making processes in the laboratory, offering practical guidelines for setting threshold values that balance simplicity, feasibility, and effectiveness in achieving desired experimental goals.

The choice of stock solutions to be made for a given experiment is determined by the target solutions required to be made in the said experiment. It will be understood that the concentration of a stock solution for a liquid reagent must be greater than the highest set point for that component in all target solutions - i.e. the stock solution must have a concentration equal to or greater than the downstream solutions comprising the reagent that are utilised within the experiment. A Concentration Factor (CF) may be defined as how much more concentrated a stock solution should be compared to the concentration of that component required in a given target solution. For example, if a stock solution is 10 g/L and the target concentration of that component required within a reaction in the experiment is 1 g/L then the CF is 10.

In a specific embodiment of the invention a method is provided for determining stock concentrations of a plurality of liquid reagents intended for use within a plurality of parallel or sequential reactions comprised within a fully or partially automated biological or chemical process. The method comprises: identifying the number of liquid reagents (N) within each of the plurality of reactions; identifying a first volumetric constraint that defines a total volume (TVol), wherein the TVol is taken as the lowest total reaction volume for the plurality of reactions; and identifying a second volumetric constraint that defines a minimum volume (MinVol) for viable liquid handling for each of the plurality of liquid reagents in the plurality of reactions; wherein the stock concentrations for the plurality of liquid reagents are set according to one or more heuristics.

Setting of a stock concentration for a given liquid reagent may comprise establishing a defined absolute concentration for the liquid reagent; or may comprise establishing a concentration factor (CF) that represents a concentration multiple of a standard accepted absolute working concentration for the liquid reagent.

In an embodiment, the one or more heuristics are selected from, but not limited to: a) setting a stock concentration factor (CF) for all reagents in the process to conform to the reaction within the plurality of parallel or sequential reactions which has the highest value of N; or b) setting a fixed CF for all reactants within all reactions; or c) setting a stock CF as high possible by dividing the TVol by a MinVol (TVol/MinVol); or d) specifying a fixed proportion of diluent for all reactions within the plurality of parallel or sequential reactions to be specified and then setting a stock CF as low as possible based on a mean value of N; or e) specifying a fixed proportion of diluent for all reactions within the plurality of parallel or sequential reactions to be specified and then setting the stock concentration (or stock CF) as low as possible based or the reaction within the plurality of parallel or sequential reactions which has the highest value of N.

According to the embodiments described, considerations for each of the heuristics are summarised below: a) Allows for higher accuracy in pipetting and makes free dispense more likely and reduces need for diluent addition, thus potentially reducing the number of pipetting steps. If consistent diluent proportion is important for performance of the reaction and the diluent is not common across all runs, then this strategy is impractical (although it can be practical if there is a single diluent by diluting all the stocks in the diluent), it is also more likely to result in multiple stocks being required. b) Simple, allows for consistency for user/operator between runs, but does not allow for full potential for optimisation. c) Less likely to require multiple stock solutions so saves on resources; provides lower accuracy by pipetting at the edge of the minimum volume; less chance of unsolvable runs occurring which cannot satisfy all component setpoints from the given stocks without exceeding the total volume. d) and e) Allows for more accurate pipetting with free dispense more likely, diluent addition is also supported and can be more consistent. More likely to result in multiple stocks being required.

Hence, through the use of heuristic principles, the methods of the invention are able to provide a novel approach to setting of stock concentrations that can improve flexibility, precision and reduce consumption of resources in biological or chemical protocols. Devices and apparatus may be configured to operate the methods of the invention as described herein. Such devices may be existing liquid handling systems, such as those set out above. In embodiments of the invention the device comprises or is comprised within a laboratory pipetting robot. Suitably the device comprises an automated liquid handling system selected from the group consisting of: a dispenser; an acoustic liquid handler; and an optofluidic liquid handler.

The invention is further illustrated by the following non-limiting examples.

EXAMPLES

Example 1 - General approach to predictive model generation The data required to build models for predicting accuracy, precision and probability of failure may be gathered by running accuracy tests.

Accuracy tests are performed on the device for which a predictive model is desired. For particular compositions or liquid properties of interest, liquids are set up to systematically vary the composition or properties of interest. These are then pipetted at various volumes with sufficient replication to allow measurement of precision and probability of failure. Various different means of handling the test liquids may be included in order to generate data on alternative pipetting strategies to later choose between.

Three different liquid compositions (Aqueous, 30% Glycerol, 70% Glycerol) are tested with four different handling strategies (glycerol liquid policy with free dispense, glycerol liquid policy with wet dispense, water liquid policy with free dispense, water liquid policy with wet dispense) at three different test volumes (1 l, 2pl, 4pl). Eight replicates of each condition are run in order to generate sufficient data to calculate coefficient of variance (% CV) and probability of failure.

After performing this experiment, a suitable assay is used to estimate the measured volume of each test transfer. For example, using the Artel Multichannel verification system (MVS). The assay output is then converted to the key metrics of interest (1) % systematic error, (2) coefficient of variance, (3) probability of failure. Models can then be generated which correlate the change in input variables tested to the resulting metrics of interest.

Methods of calculating each metric of interest are as follows:

1. % Systematic error = 100% x (Measured Volume - Target Volume) I Target Volume

2. % Coefficient Of Variance = 100% x Standard Deviation I Mean

3. % Probability of failure = 100% x Condition Failures* I Condition failures* and successes

*A failure is defined as a run which failed to meet given success criteria. For example, any run with a systematic error of below -75% may be considered a failure. A success is a run which is not a failure. A condition is a row in the Table 1 design below.

Table 1 shows an example design forgathering accuracy test data which can be used to build models correlating viscosity with liquid handling quality for two base liquid policies when dispensed (DSPREFERENCE) at either the well bottom (into the destination liquid i.e. wet dispense) or well top (i.e. free dispense above the destination liquid). The two BasePolicies, water and glycerol, represent collections of liquid handling parameters designed for either glycerol-like, or viscous liquids, and water-like liquids respectively. The Testvolume is given in microlitres (pl). CAN_MULTI is a setting which enables the transfers to be performed using multichannel pipetting. Free dispenses (DSPREFERENCE = well_top) are not permitted to use multi channel pipetting because it is preferred that the replicates re-use the pipette tip in order to give an indication as to how robust the condition/liquid combination is to facilitate tip re-use.

Table 1

In this example, either the glycerol volume fraction itself may be included as an input or the glycerol volume fraction may be first converted into estimated viscosity using the published empirically determined relationship between glycerol volume fraction, temperature and viscosity (Volk, A., Kahler, C.J. Density model for aqueous glycerol solutions. Exp Fluids 59, 75 (2018)). Following the running of the accuracy test design above, the data is plotted and models are generated for each of the metrics as shown in Figures 2 and 3.

As well as models to predict each metric as a function of the other inputs, it is desirable to establish the volume threshold where free dispense with acceptable quality can be achieved:

Accurate (Systemic error close to 0)

Precise (low Coefficient of Variance (CV))

Low probability of failure (failure = Systemic error below -75%)

As fast dispense strategy as possible which matches the quality criteria

Avoid tip changes where possible

Facilitate building of models to help decision making

Figure 2 shows the data plots and model for determining a volume threshold for pipetting above which no failures are predicted to occur for free dispenses as well as accepting an acceptable bias level. Figure 3 shows models for predicting bias in execution of pipetting strategies.

Example 2 - Linear Regression Model (RidqeCV) for eliminating bias in automated free dispenses of solutions comprising water and glycerol

• Choose features / Response

Linear modelling techniques such as ridge regression may be used to fit the experimental data and build predictive models. They are particularly useful for fitting the bias (systematic error) model. The data is split into features and a response. If the data is in a tabular format then typically the features and responses will be column headers. Features are chosen as properties which might be used to predict the response. These may be either categorical (e.g. device, dispense mode, liquid policy, liquid) or, preferably, numerical (e.g. target transfer volume, viscosity, subcomponent concentration, dispense flow rate, evaporation potential, aspirate tip depth).

A) Example Features: a. Dispense Mode (Dry, Wet, Free) b. Policy (glycerol, water) c. Transfer Volume (1 pl, 2pl, 4pl, 8pl) d. Viscosity (0.001 , 0.002, 0.05 N/m².s)

B) Response: % systematic error

Split data into training/test data or cross validation groups The data is then split into training and test data. Typically 80% of the data is used as training data and 20% as test data. Training data is used to train a model; test data is used to validate the performance of the trained model on data which the model has not been exposed to (i.e. during training). If the accuracy of the model at predicting responses in the training data is much better than predicting responses using the test data then the model is overfitting to the training data.

• Pre-process categorical features using one-hot encoding

Categorical features are typically not supported in most machine learning modelling strategies so they are converted to numerical values prior to model fitting. A safe and robust way to do this is by one-hot encoding. Whereby each categorical value is converted into a new feature with a 1 for rows where that categorical value is present in the original row and zero where it is not. Table 2 shows the output of this process for the first four rows of a data set including two categorical features (Dispense Mode with values Free and Wet and Policy with values glycerol and water).

Table 2

• Pre-process numerical values to add transformed versions of each.

If linear regression techniques are to be used (for example Ridge regression) then transformations are added to support non linear relationships between features and the response. In this case additional columns would be added representing the transformed feature.

E.g. shown as: Transfer_Volume_pl ², Transfer_Volume_pl ³. Additionally, interactions between may be added (e.g. Transfer_Volume_pl x Viscosity_N/m².s).

• Regression model approach

RidgeCV is an approach which can be used to prevent overfitting.

Ridge regression addresses some of the problems of Ordinary Least Squares by imposing a penalty on the size of the coefficients. The ridge coefficients minimize a penalized residual sum of squares: minw||Xw-y||22+a||w||22

The complexity parameter a>0 controls the amount of shrinkage: the larger the value of a, the greater the amount of shrinkage and thus the coefficients become more robust to collinearity.

Grid search to find model tuning parameters which fit the data best In this example, this includes the alpha applied as a penalty for addition of new terms in the Ridge regression approach and the polynomial to use in preprocessing of numerical features.

Example python code using the scikit learn library (sklearn) is shown in Figure 4.

An exemplary output from the bias model using linear (RidgeCV) polynomial order 3 is shown in Figure 5.

Usage of the model when planning liquid handling instructions:

The model can then be stored, for example as a file, and deployed by a server which may be accessed via software responsible for planning liquid handling. In this case the model was deployed using FASTAPI and the model then queried by the planner using an API. At which point, when a liquid is to be transferred, the transfer properties will be queried via the API and a prediction of the expected bias returned for the particular liquid policy, volume and viscosity queried. At this point, alternative variants may be queried in order to choose between alternative pipetting strategies, e.g. different liquid policy or dispense modes; the best may be chosen, or the fastest which meets some specified success criteria (i.e. an acceptable level of bias); alternatively the model may be used to correct anticipated bias by adjusting the target volume.

An example of how model code might be deployed is shown using FASTAPI in Figure 6. An example of how this can be queried by liquid handling planner software is shown in Figure 7. Example criteria which may be used to choose a suitable liquid policy if the criteria are predicted to be achieved by querying the models with a given set of candidate conditions (see Figure 8).

Example transfer options which might be queried from the model generated in this example are shown in Figure 9.

Transfer options can be evaluated in order of desirability and as soon as a solution meets the quality criteria, that option is used by the system. For example, Free Dispense with Proxy liquid Aqueous is evaluated first as of the 4 options, this is the fastest to execute.

Alternatively, a model for time cost may be added which is used to sort the various transfer options.

Another option is to prioritise other metrics rather than time, for example to always choose the option which results in the best pipetting performance or to choose the option with the lowest liquid wastage.

The models may also be used to choose the stock concentration of each reagent required in the experiment. The example code in Figure 10 shows how this may be performed by iteratively querying the model for transfer of increasing volume; when a prediction is returned which passes success criteria, the stock concentration may be set to the value which will always result in achieving the success criteria; additionally the transfer options provided may be restricted to only include free dispense options in order to ensure the stock concentration is set to a value which always results in free dispenses.

Example 3 - Generation of a neural net ML model for eliminating bias in automated free dispenses of solutions comprising water and glycerol

The model generated in Example 2 may alternatively be generated using a neural net based approach. In this example, non linear relationships can be found without transforming the numerical features using a polynomial transformer. The example below uses the multilayer perceptron regressor from the scikit learn library to fit a model to the data.

A grid search of hyperparameters in the pipeline is also performed in order to fit the model with highest accuracy.

An example of python code is shown in Figure 11 .

The model output is shown in Figure 12 and compares experimental data (in left panel) with predictions made by the model (in right panel).

Example 4 - Generation of an autoML model (TPOT) for eliminating bias in automated free dispenses of solutions comprising water and glycerol

An autoML based approach may also be used whereby choice of model as well as hyperparameters is performed by the algorithm using a genetic programming approach.

An example of python code is shown in Figure 13.

The model output is shown in Figure 14 comparing experimental data (in left hand panel) with predictions made by the model (in right hand panel).

Example 5 - Generation of an autoML model (TPOT) for predicting probability of failure in free versus wet dispenses

A probability of failure model can be generated in a similar fashion; either using a regression model to predict the calculated probability of failure as described earlier or using a classification model to predict whether or not any given datapoint is a failure and using the model confidence in that prediction as the probability of failure. This example uses TPOT to predict the former, using the same code as shown in Example 4 (see Figure 13). The example model output is shown in shown comparing experimental data (in left panel) with predictions made by the model (in right panel). The model provides a prediction for values which define the volume threshold above which no pipetting failures are expected when performing free dispenses (R² = 0.80).

Example 6 - Generation of a linear regression model for predicting evaporation of liquids by combining literature models with experimental data

Models for other metrics (besides accuracy/precision) may also be generated from experiments and used for predictions in future runs. This example shows this applied to generating a model to predict evaporative loss. An empirical evaporation model may be generated by either using established evaporation models from literature or setting up and performing experiments to measure evaporation in the relevant context. This example demonstrates the latter case.

An experiment is set up whereby wells of liquid are set up varying the following parameters:

• Surface Area of liquid exposed to the air (varied by use of plate types with different well geometry, i.e. 96 and 384 well plate types etc)

• Initial volume in well (liquid height)

• Total Moles of Solute (1 , 0.1 , 0.01)

• Solute composition (mixtures of DMSO, NaCI & Glycerol)

Periodically (every hour for 96 hours), the plates are moved from the liquid handling deck to the plate reader and absorbance readings are made which allow the measurement of liquid height based on the Beer Lambert law. This was achieved by using absorbance of 977 nm, at which water produces a peak and absorbance at a reference wavelength of 900nm, where there is no peak. The following equation is used to measure liquid height:

Liquid height in cm = (Absorbance at 977 nm - Absorbance at 900 nm) I K factor

Where, K factor = Absorbance at 977 nm at path length of 1 cm - Absorbance at 900 nm at path length of 1 cm.

Sensors are used to measure temperature, humidity and pressure during experimentation; these readings are used to calculate the expected evaporation rate according to a literature formula which predicts pure water evaporation independent of solute composition;

Gs = 0 A (Xs - x) / 3600 Where g_s is the amount of evaporated water per second (kg/s)

0 is (25 + 19 v) = evaporation coefficient (kg/m²h) v is the velocity of air above the water surface (m/s)

A is the water surface area (m²)

Xs is the maximum humidity ratio of saturated air at the same temperature as the water surface (kg/kg) (kg H2O in kg dry air) x is the humidity ratio of air (kg/kg) (kg H2O in kg dry air)

(from https://www.engineeringtoolbox.com/evaporation-water-surface-d_690.html)

Once the data is collected and aligned by time; linear regression using ordinary least squares or ridge regression can be used in a similar approach as in Examples 1 and 2 using similar code shown in Example 2 (see Figures 6 to 9), in this case, to predict the rate of evaporation based on the features: (i) initial liquid height in the well (ii) expected evaporation rate calculated from the literature formula (iii) well surface area (iv) volume or molar fraction of water and other solutes in the liquid.

The results are shown in the plots of the experimental data set out in Figure 16.

Predicted versus actual results generated from the Linear regression model were fitted to the data (Figure 17 top panel)and a chart showing magnitude of each factor toward achieving accuracy was also generated (Figure 17 bottom panel). The Linear regression can be used to adapt the published literature model to fit to the empirical data generated by the apparatus and protocol being evaluated. This approach also allows for the incorporation of additional parameters such as water volume within the well (v/v) and liquid height thereby further extending the capabilities of reported literature models. The approach allow for highly accurate predictive models of evaporation (R² > 0.92).

Example 8 - Generation of a model to predict homogeneity of a solution based on mixing conditions:

This example shows how models to predict homogeneity of solutions based on mixing conditions, liquid composition and diffusion may also be experimentally generated. To test forthis, the Artel MVS accuracy test may be used. Homogeneity is assessed by transferring first to an intermediate plate, applying any test mixing conditions, and then transferring half of the contents of each intermediate well (by aspirating from the top of the liquid) and then transferring to a second plate. If intermediate liquid was homogenous at the point at which the transfer to the second plate occurs then the dye concentration measured in the intermediate well will be the same as the second well. If the intermediate well was completely unhomogenous then the dye concentration measured in one of the wells will be zero. This is shown pictorially in Figure 18(a). Variables investigated are shown in Figure 18(b) as: Proportion of total volume mixed (0, 0.5, 1 , 2), mix volume (5, 12.5, 25, 50 pl), Z offset (0.5 mm), mix flow rate (75 pl/s), viscosity (0.001 N/m².s, 0.0023 N/m².s, 0.012 N/m².s).

Proportion of Total Volume Mixed is calculated as shown in Figure 18 (c), and as below:

Mix volume x Cycles I Total liquid volume = Proportion of total volume mixed

20 pl x 1 / 100pl = 0.2

20 pl x 5 / 100 pl = 1

20 pl x 10 / 100 pl = 2 50 p/ x 4 / 100 pl = 2

The results show the volume of dye measured in the intermediate well (remaining) and second well (sampled). 4ul was measured for each for cases where the intermediate was homogenous at the point at which it is sampled and the results are set out in Figure 19.

The two measured volumes (or concentrations) can be converted into a number (between -1 and 1) representing the concentration gradient. A concentration gradient of 0 is the target; this represents homogeneity. The results following conversion are shown in Figure 20.

An accurate linear model may be generated using the same code as shown in Example 2. The output of the model is shown in Figure 21 and demonstrates highly accurate predictive capability for determination of concentration gradient (R² > 0.95).

Example 9 - Generation of a model to predict bias based on pressure curve profiles:

In this example, an equivalent experiment is set up to that shown in Example 1 but with an additional assay performed on the samples to measure the change in pressure in the pipette head during the aspirate and dispense steps; particularly if this capability is available directly on the liquid handling device for which the models are being generated; in which case the metrics can be captured at the same time as the experiment is set up. If not, the pressure profiles may be generated on a compatible device (for example a Hamilton Star) and used as a “signature” for that liquid, for which the associated metrics may be used as model features correlated with liquid properties. In this example, once pressure profile signatures exist for a given liquid/liquid policy combination, pressure profile metrics may be used to query the model in order to make predictions for the likely quality of the candidate transfer. Examplary accuracy data was obtained from Artel MVS for several liquid solutions and is shown in Figure 22 - from left to right: bovine serum albumin (BSA), glucose, glycerol, sodium chloride (NaCI) and water.

Exemplary pressure profiles obtained during pipette aspirate and dispense phases of water and BSA (bovine serum albumin) solutions are shown in Figure 23.

A model output was generated for predicting transfer volume based on metrics derived from the pressure profiles obtained during aspiration (see Figure 24). The model showed highly accurate predictive capability for determination of transfer volumes for water and BSA containing solution (R² > 0.99).

Example 10 - Generation of a model to predict precision (CV) for semi-automated pipetting using an electronic “connected pipette” which is guided by the Gilson Pipette Pilot system:

The same approach can be used when generating liquid handling plans for semi-automated pipetting systems such as the Gilson PipettePilot. In this case, liquid addition order, flow rates, stock concentrations, volumes, pre-wetting, multi-dispense and pre and post mixing can be set by the model-driven scheduling system. Additionally, guidance displayed to the user, via a GUI, advising on the correct dispense position (free dispense, side dispense on to the wall of the well or a wet dispense into liquid).

This example shows development of a model generated by setting up a similar experiment to that shown in Example 1 albeit with the following design:

Test Solution: Aqueous, 24% Glycerol, 40% Glycerol, 70% Glycerol

Volumes: 1 ,2, 4, 8, 18 pl

LiquidPolicy: water, glycerol

Note: the liquid policies in this example differ in flow rate; (“glycerol” using a slower relative flow rate than “water”)

The resulting predictive model was generated using the Ridge modelling methodology from Example 1 (R² = 0.95) and is shown in Figure 25. Upon scheduling liquid handling plans on the Pipette Pilot this model can be queried to ensure the policy with the lowest predicted CV is chosen for a given volume and liquid viscosity. Furthermore, as in the other examples, the stock concentration of reagents can be set appropriately to ensure all volumes that are to be transferred are high enough such that acceptable precision, as predicted by the model, is likely to be achieved. Example 11 - Generation of a model to predict precision (CV) for a liquid dispenser based on grouping of liquids by clustering bye, like behaviours:

When liquids display similar properties, results for these liquids can be clustered into groups in order to generate models from greater amounts of data with greater diversity. In this example, accuracy tests are run for a suite of liquids and then the results for the liquids are clustered into four clusters prior to generating models: 1. Aqueous-like (composed of Water, 10g/L NaCI solution and 50 g/L Glucose solution). 2. BSA-Like (composed of 2g/L BSA, 20g/L BSA and 100 g/L BSA) 3. Quite viscous (24% Glycerol and 40% Glycerol in water) 4. Very Viscous (70% Glycerol in water). The means of clustering was based on the similarity of pressure curves observed previously (as in example 9).

In this example, a liquid dispenser (SPT Dragonfly) is used which will only perform free dispenses, has no liquid classes/liquid policies and has limited control of liquid handling parameters. In this case, the model is used solely to select stock concentrations of the liquids in order to ensure transfers for each liquid are likely to be sufficient to ensure the CV for each transfer is acceptable. For example, if the acceptable CV threshold was set to 6% then liquids labelled into the BSA-like cluster would need to ensure the stock concentrations were set to ensure minimum volume of 7pl is pipetted; whereas liquids put into the very-viscous cluster can be set to a stock concentration which results in transfers of around 2pl or above. The resulting predictive model for predicting CV by volume forthese liquid clusters was generated using the Ridge modelling methodology from Example 1 (R² = 0.83) and is shown in Figure 26.

Example 12 - Generating stock solutions for optimal liquid handling - setting concentration factors (CFs) for a range of stock solutions to the maximum possible

A concentrated stock solution, through a series of dilutions, can create multiple target solutions of different concentrations. To produce target solutions, several liquid handling procedures are involved such as: aspirating one or more stock solutions, dispensing the solutions into a target reaction mixture, and adding appropriate amounts of diluent.

One or more heuristics of baseline execution strategies for determining concentration of the stock solution may be chosen to reduce complexity, to improve accuracy, speed, and reduction of consumption of disposables, and/or to reduce the number of stock solutions required for a given laboratory process.

One baseline execution strategy is to set concentration of the stock solution to reach an upper limit of the concentration of each of the components. The upper limit is often introduced due to the solubility limit or pre-made solution availability from the vendors.

If more components are required in the target reaction solutions, the less volume is available per each component as the total volume (Tvol) of each target concentration is limited. The stock concentration can be then set to each upper limit of the concentration of each component. The benefit of making the stock concentrations as high as possible includes a reduced need to set up multiple stock solutions having different CFs for the same liquid reagent. In addition, higher stock concentrations also reduce the likelihood of failing to satisfy some target solution combinations due to not fitting in all transfers without overshooting the TVol restriction. Furthermore, by minimizing the volumes of liquid reagent components, any potential effects of cosolvents/diluents contained in the stocks on the target reaction solutions can be minimized (e.g. by adjusting pH).

The concentration factor (CF) of a stock solution can be set by calculating ^TmL . wherein Tvol is the Minvol total volume of the lowest volume target solution in an experiment and MinVol is the minimum volume for acceptable pipetting behaviour. The acceptable pipetting behaviour may entail the minimal volume that can be dispensed with a known accuracy, such as the minimal volume limit of a pipetting device, either manual or comprised within a liquid dispensing handler apparatus.

MinVol value for the baseline execution strategy can be identified from the specification and setting of the pipetting volume of the liquid handler, which may depend on factors such as the liquid type, liquid class or dispense mode of the liquid handling apparatus.

MinVol is one of the parameter values related to the liquid handling step that can be determined experimentally. The experiment can involve measurement of CV (Coefficient of Variance) against each transfer volume, as illustrated in Figure 26. The experiment can be repeated for liquids with different composition or properties such as viscosity. The experiment can be repeated for the type of liquid handling step (e.g. free dispense of liquid from the pipette tip without contacting the destination solution, or wet dispense of liquid from the pipette tip submerged in the destination solution).

One of the performance quality criteria that can be used to optimise Min Vol is CV. The coefficient of variance, when plotted against each transfer volume, provides MinVol in a form of threshold minimal transfer volume that allows pipetting behaviour with acceptable quality. The quality in this case is determined by reference to precision, which is derived from the coefficient of variance value.

Example 1 (see above) illustrates further inputs or performance quality criteria that may help to assess an acceptable MinVol: such as accuracy, precision, low probability of failure, and speed of dispense.

To further determine an acceptable MinVol, it may be possible to alter the liquid dispensing strategy, at another threshold that is greater than MinVol. For example, at a threshold volume which is higher than MinVol but in which free dispenses may start to become less precise, the users may be directed to wet dispense, or pipette directly into the destination liquid instead. The appropriate threshold may be determined by plotting the CV against the transfer volume while performing a wet dispense. However, the wet dispense will require more time for each step of liquid handling. This represents an informed trade-off between an advantageous higher stock concentration but longer time and increased consumption of disposables such as pipette tips. Example 13 - Optimising to maximise transfer volumes and minimise diluent addition

In a different heuristic, the stock concentration factor can be alternatively set to according to the number of components (N) in the target solution/mixture having the most liquid reagent components. Doing so can avoid the need to add diluent and therefore increase the speed of liquid handling. However, this approach puts a limit on how dilute the target concentrations can be. For example, if the mixture includes six components of equal volume, the target concentration will not be more diluted than six times from the stock solutions of the components, unless a diluent is introduced.

This baseline execution strategy can be modified to include a dilution step when the use of diluent is critical, for instance, for making a growth media or a solution that requires buffer. Dilute liquids tend to be less viscous and therefore more favourable for accurate pipetting with standardised liquid handling conditions. As shown in other examples, the higher volumes also tend to be transferred more accurately and are more likely to result in allowing the faster free dispenses. This strategy therefore typically results in both faster execution and higher quality pipetting - e.g. lower %CV. The downside is the increased likelihood that multiple stock solutions will be required for a given liquid reagent component.

The examples of the liquid handling metrics related to the diluting step are given below.

Table 3

The concentration factor may be calculated in a way which takes in to consideration a required diluent proportion (F) by using the equation CF = ^N , wherein CF = concentration factor, N = number of (l-F) components in the solution, and F = The fraction of diluent volume relative to the total solution volume.

Example 14 - Optimising the number of stock solutions - using MinVol

For any baseline execution strategy, the concentration as well as the number of stock solutions required can be determined by defining the desired concentration of each component in the target solution, the number of liquid reagents (N) within the plurality of reactions, and the total volume for the plurality of reactions (TVol). In the practical example of the target solutions in Table 4, parallel experiments containing multiple target solutions with different target set points are described.

Table 4

According to the embodiments of the present invention, the user can minimise the time and resources by generating a single stock solution for each component. For instance, stock solutions of 100g/L Glucose and 50 g/L Yeast Extract could be made to satisfy the design and combined to generate all four target solutions outlined above (see Table 5). Table 5

It may be necessary to have more than one stock for any given subcomponent.

Some additional metrics of the liquid handling step, such as a minimum volume (MinVol) for the plurality of reactions, can be further introduced to redefine the concentration of a stock solution, or introduce additional stock solutions.

For example, if MinVol is set to 2 pl, an additional glucose stock concentration lower than 100g/L should be made to ensure the accuracy of pipetting. As outlined in other examples, the MinVol may also be set differently for different stocks and may be determined by a property of the liquid (e.g. viscosity). Example 15 - Optimising the number of stock solutions - using fixed concentration for all reagents

An alternative heuristic for a baseline execution strategy may be to keep a fixed stock concentration factor of, say, 10 times (10x) for all reactants used within mixtures for all liguid handling steps. Using a fixed concentration factor allows a simple heuristic of maintaining a level of consistency for the user between each run without additional optimisation. This might be optimal where the same, or similar, reactions are performed repeatedly on a liguid handling apparatus.

To generate the target solutions of Table 6, stock solutions of 100g/L Glucose, 10g/L Glucose and and 50 g/L Yeast Extract can be made and combined to generate the desired target solutions.

Table 6

Example 16 - Selection of improved process/assay conditions and liguid reagent formulations using a predictive modelling approach

As described previously, in silico models can be used to predict improvements in wider process metrics that can be associated with the performance of liguid handling steps within a biological or chemical process. Optimisation of the process metrics can conseguently improve execution and overall performance of the process as a whole.

The data reguired to build models for predicting the relevant process metric may be obtained by performing a baseline execution strategy for the process or a proxy for a given process metric within that process. Models may be used to help predict assay guality, e.g. by determining Z-prime values for particular assays.

Process metrics may apply to execution of functional assays, binding assays and activity assays.

Some examples of process metrics which may be usefully predicted in biological processes, such as bio-assays, may include cell density (g/L), live cell density (cells/ml), recombinant protein expression titre (g/L), rate of protein expression (g/L/h), specific rate of protein expression (g/g/h), activity (U/ml) etc. For example, cell growth may be predicted by performing well established equivalents such as optical density measurements and estimating the cell density in g/L or cells/ml. Protein expression related metrics may be gathered by directly or indirectly measuring the activity of a protein (for example fluorescence of a fluorescent protein, enzyme activity of an enzyme, binding of a ligand), or measuring the quantity of the protein (His tag binding, western blot, dot blot, HPLC, SDS PAGE, capillary electrophoresis).

This example shows generation of models for prediction of molecules of equivalent fluorescein (MEFL) and maximum rate of MEFL production (MEFL/min).

Liquid assay reaction mixtures of microbial growth media were set up to systematically determine the effect of these components on the expression of recombinant green fluorescent protein (GFP).

Additionally, the effect of process conditions (induction time) and proxy process conditions (Total volume modified as a proxy for oxygen transfer rate) were measured. Each run was set up in microwells of a 384 well plate and incubated in a plate reader and fluorescence and optical density measured every 30 minutes and converted to MEFL by cross referencing to the fluorescence per cm readings of a standard curve of samples containing fluorescein at known concentrations. A model was then fit to the data to predict MEFL/min using a linear modelling approach (Ridge). One or more iterations can be performed, each time a model can be fit to data from all iterations.

Table 7 : Experimental Factors to investigate optimization of GFP expression

The model is generated and may be queried for future use cases in which a user has specified the use case to be similar and or the use case is algorithmically determined to be similar to the use case for which the original model was created. The model may be queried in a similar fashion to the liquid handling optimization examples at the point at which the “similar” new run is performed. Quality requirements are specified by the user for the metrics of interest (in this case MEFL/min). A combinatorial array of conditions are generated within the ranges set by the original experimental conditions which generated the model and the conditions predicted to perform best can be used in a modified execution strategy. If some specific setpoints are chosen by a user the model can alternatively be used to offer predictions for those conditions and optionally judge if they are predicted to achieve the success threshold specified by the user.

Figure 27 shows a model in the form of a graph in which all candidate conditions considered for use and the predicted responses of each of those conditions are provided for the responses of cost per run and activity. The points highlighted as triangles are those which are deemed to meet the required user threshold of an activity of greater than 65 (MEFL activity) and an reagent cost per run of less than 7.

Hence, the predictive model may be exploited to identify process conditions (e.g. by specifying particular metrics) which can achieve multiple objectives.

Although particular embodiments of the invention have been disclosed herein in detail, this has been done by way of example and for the purposes of illustration only. The aforementioned embodiments are not intended to be limiting with respect to the scope of the appended claims, which follow. It is contemplated by the inventors that various substitutions, alterations, and modifications may be made to the invention without departing from the spirit and scope of the invention as defined by the claims.

Claims

WHAT IS CLAIMED IS:

1 . A method for implementing a biological or chemical process, wherein the process comprises at least one liquid handling step, the method comprising: establishing a baseline execution strategy for the process; identifying at least one liquid handling step that may be used within the baseline execution strategy; identifying a value for a metric associated with the liquid handling step; and modifying the baseline execution strategy to generate a modified execution strategy for implementing the process by optimizing the value for the liquid handling step; wherein identification of the value for the metric comprises either or both of: i. generating a prediction for the liquid handling step by applying one or more in silico rules or models of liquid handling; and/or ii. generating performance quality criteria for the liquid handling step following execution of the baseline execution strategy.

2. The method of claim 1 , wherein the modified execution strategy comprises an improvement versus the baseline execution strategy in terms of one or more of the group consisting of: increased liquid transfer accuracy; reduced consumption of resources; reduced probability of process failure; improved process quality; reduced execution time; improved precision; improved accuracy; and improved process yield.

3. The method of any previous claim wherein the process comprises a plurality of automated liquid handling steps.

4. The method of claim 3, wherein a plurality of values are established for a plurality of metrics for each of the plurality of liquid handling steps.

5. The method of claim 4, wherein the plurality of values are propagated across the baseline execution strategy to yield summarised metric values for the baseline execution strategy.

6. The method of claim 5, wherein the summarised metric values are presented to a user.

7. The method of any one of claims 3 to 6, wherein the plurality of values are propagated across the modified execution strategy to yield summarised metric values for the modified execution strategy.

8. The method of claim 7, wherein the summarised metric values for the modified execution strategy are presented to a user as a comparison with the summarised metric values for the baseline execution strategy.

9. The method of any one of claims 3 to 8, wherein the method generates more than one modified execution strategy.

10. The method of any one of claims 1 to 9, wherein the method is a computer implemented method and wherein the at least one liquid handling step is performed by using a liquid handling apparatus under the operative control of a processor.

11. The method of claim 10, wherein the processor evaluates and selects the most optimal modified execution strategy based upon predetermined criteria established by a user.

12. The method of any one of claims 1 to 11 , wherein the at least one liquid handling step comprises transfer of a volume of liquid of less than 10,000 pl, optionally less than 1000 pl.

13. The method of claim 12, wherein the predetermined criteria are selected from one of more of the group consisting of: increased liquid transfer accuracy; reduced consumption of resources; reduced probability of process failure; improved process quality; and improved process yield.

14. The method of any one of claims 1 to 13, wherein the baseline execution strategy is defined by a user.

15. The method of claim 14, wherein the baseline execution strategy is derived from a text source, optionally wherein the text source is comprised within one or more of the group consisting of: a literature reference; a patent publication; a text book reference; a written protocol; a web page; an on line document; an automation script; and a reference manual.

16. The method of any one of claims 1 to 15, wherein the step (i) comprises assigning a quantifiable liquid property to a liquid that is subject to a liquid handling step.

17. The method of claim 16, wherein the liquid property is selected from one or more of the group consisting of: liquid type; solute concentration; viscosity; surface tension; volatility; density; adherence; stability; homogeneity; evaporation; charge; hydrophobicity; rheology; liquid temperature; sheer sensitivity; shear stress; and miscibility.

18. The method of claim 17, wherein the liquid type is selected from the group consisting of: aqueous solvent; non-aqueous solvent; biological medium; emulsion; particulate suspension; cell culture suspension; serum-containing medium; protein-in-solution; volatile solvent; nucleic-acid in solution; and viscous solution.

19. The method of any one of claims 1 to 18, wherein the step (i) comprises assigning a quantifiable hardware property to hardware used in the automated liquid handling step.

20. The method of claim 19, wherein the hardware property is selected from one or more of the group consisting of: automated or semi-automated liquid handling apparatus type and/or capability; pipetting type; pipetting tip type; pipetting tip availability; receptacle type; and receptacle availability.

21. The method of any one of claims 1 to 20, wherein the step (i) comprises assigning a quantifiable environmental property to the automated liquid handling step.

22. The method of claim 21 , wherein the environmental property is selected from one or more of the group consisting of: temperature; humidity; atmospheric composition; and atmospheric pressure.

23. The method of any one of claims 1 to 22, wherein optimizing the value for the automated liquid handling step comprises modifying the baseline execution strategy to increase the number of free dispenses and/or to ensure a better coefficient of variation and/or to reduce probability of failure.

24. The method of claim 23, wherein the optimization is computer implemented.

25. The method of claim 23 or claim 24, wherein the optimisation is determined via an in silico model in which a minimum volume of liquid for successful execution of a free dispense is established for a given liquid handling property, and wherein the minimum volume is compared to the proposed volume transferred in the baseline execution strategy, and wherein is the proposed volume is greater than or equal to the minimum volume a successful free dispense is predicted, and where the proposed volume is less than the minimum volume an unsuccessful free dispense is predicted.

26. The method of any of claims 23 to 25, wherein the baseline execution strategy is modified by changing a liquid property of a liquid that is subject to a liquid handling step, optionally wherein the concentration and/or the volume of the liquid is changed.

27. The method of claim 26, wherein the liquid comprises one or more stock solutions and wherein the baseline execution strategy is modified to optimise a concentration factor (CF) for the one or more stock solutions.

28. A liquid handling apparatus configured to implement a biological or chemical process according to a method as defined in any one of claims 1 to 27.

29. A method for determining one or more stock concentrations for a plurality of liquid reagents intended for use within a plurality of parallel or sequential reactions comprised within a fully or partially automated biological or chemical process, the method comprising: identifying the number of liquid reagents (N) within the plurality of reactions; identifying a first volumetric constraint that defines a total volume (TVol) for each of the plurality of reactions; and identifying a second volumetric constraint that defines a minimum volume (MinVol) for each liquid reagent comprised within the plurality of reactions; wherein the stock concentrations for the plurality of liquid reagents are set according to one or more heuristics that may be determined with reference to at least one of N, TVol and/or MinVol.

30. A method for implementing a biological or chemical process, wherein the process comprises at least one liquid handling step, the method comprising: establishing a baseline execution strategy for the process; identifying at least one process step that may be used within the baseline execution strategy; determining a value for a process metric; and modifying the baseline execution strategy to generate a modified execution strategy for implementing the process by optimizing the value for the process metric; wherein determination of the value for the process metric comprises either or both of: a. generating a prediction for the process metric by applying one or more in silico rules or models for the process metric; and/or b. specifying performance quality criteria for the process metric prior to or following execution of the baseline execution strategy.

31 . The method of claim 30, wherein the process metric is selected from at least one of: product yield; conversion efficiency; throughput; cycle time; product quality; product purity; product concentration; product activity; intermediate quality; intermediate purity; intermediate concentration; and intermediate activity.

32. The method of claim 30 or 31 , wherein the modified execution strategy comprises an improvement versus the baseline execution strategy in terms of one or more of the group consisting of: reduced consumption of resources; reduced probability of process failure; improved process quality; reduced execution time; and improved process yield.

33. The method of any one of claims 30 to 32, wherein the process comprises a plurality of automated liquid handling steps.

34. The method of any one of claims 30 to 33, wherein a plurality of values are established for a plurality of process metrics.

35. The method of claim 34, wherein the plurality of values are propagated across the baseline execution strategy to yield summarised process metric values for the baseline execution strategy.

36. The method of claim 35, wherein the summarised process metric values are presented to a user.

37. The method of any one of claims 34 to 36, wherein the plurality of values are propagated across the modified execution strategy to yield summarised process metric values for the modified execution strategy.

38. The method of claim 37, wherein the summarised process metric values for the modified execution strategy are presented to a user as a comparison with the summarised metric values for the baseline execution strategy.

39. The method of any one of claims 34 to 38, wherein the method generates more than one modified execution strategy.

40. The method of any one of claims 30 to 39, wherein the method is a computer implemented method and wherein the at least one liquid handling step is performed by using a liquid handling apparatus under the operative control of a processor.

41 . The method of claim 40, wherein the processor evaluates and selects the most optimal modified execution strategy based upon predetermined criteria established by a user.

42. The method of any one of claims 30 to 41 , wherein the at least one liquid handling step comprises transfer of a volume of liquid of less than 10,000 pl, optionally less than 1000 pl.

43. The method of claims 41 or 42, wherein the predetermined criteria are selected from one of more of the group consisting of: reduced consumption of resources; reduced probability of process failure; improved process quality; reduced process execution time; and improved process yield.

44. The method of any one of claims 30 to 43, wherein the baseline execution strategy is defined by a user.

45. The method of claim 44, wherein the baseline execution strategy is derived from a text source, optionally wherein the text source is comprised within one or more of the group consisting of: a literature reference; a patent publication; a text book reference; a written protocol; a web page; an on line document; an automation script; and a reference manual.

46. A liquid handling apparatus configured to implement a biological or chemical process according to a method as defined in any one of claims 30 to 45.