Method, device, program, and recording medium for exploring medium composition and estimating cell characteristics

Abstract

One aspect of the present invention provides a method, device, and program for searching medium composition and estimating cell characteristics, which can efficiently optimize culture results even for a cell of which characteristics are unknown. In a method according to one aspect of the present invention, a processor: predicts, with respect to a subject cell, culture results for a given medium composition by using the cell characteristics or an estimated value of the subject cell; acquires the culture results of the subject cell for a portion or the entirety of the given medium composition; estimates the cell characteristics of the subject cell by using the medium composition, and the acquired culture results or the predicted culture results; evaluates, from the acquired culture results, the necessity of further estimating the cell characteristics of the subject cell; and searches and determines, on the basis of the necessity and prediction, candidates of the medium composition for which the next culture results should be acquired. The processor repeats the prediction of the culture results, the acquisition of the culture results, the estimation of the cell characteristics, the evaluation of the necessity, and the search and determination of the candidates of the medium composition until a determined termination criterion is satisfied.

Description

Method, device, and program for exploring medium composition and estimating cell characteristics

[0001] The present invention relates to a method, an apparatus, and a program for searching for a medium composition and estimating a cell characteristic, and more particularly to a method, an apparatus, and a program for searching for a medium composition that is particularly suitable for a specific cell line.

[0002] The recent spread of gene editing technology has led to an increasing diversity of cell types, including Chinese Hamster Ovary (CHO) cells for producing antibody drugs, human embryonic kidney (HEK) cells for producing viruses for gene therapy, and induced pluripotent stem (iPS) cells for regenerative medicine. While there are many cases where cell culture conditions are optimized to increase cell proliferation and product yield, there are also many cases where media are specifically optimized for specific cells. In these cases, a common solution is to start with a known medium composition and gradually change the content of one or more components, culturing cells in multiple media to maximize the target productivity, such as proliferation.

[0003] For example, Patent Document 1 describes the construction of a functional enviromics map that represents the intensity of activation or inhibition of basic cellular functions, and the use of this functional enviromics map to develop an optimized cell culture medium formulation.

[0004] Special table 2014-503220 publication

[0005] When optimizing a medium for specific cells (unique cells), it is considered effective to (1) understand the characteristics of the unknown cells and (2) specify the medium composition according to those characteristics. However, there are cases where it is necessary to search for an optimal medium for cells whose characteristics are not necessarily known. This is because the process of understanding cell characteristics itself requires cost and time, and the analysis required to understand cell characteristics can be difficult due to industrial constraints. Therefore, there is a need to search for and propose an optimal medium composition for cells with unknown characteristics within the limited information available.

[0006] Generally, in optimization problems such as (2) above, a specific good medium composition is gradually changed to optimize it. In contrast, in (1) above, in order to clarify cell characteristics, it is common to examine the response to various medium compositions, and when searching for an optimal medium composition for cells with unknown characteristics, there is a problem that it may ultimately require the same amount of time and cost as the general solution described above.

[0007] As described above, with conventional technology, it has been difficult to achieve both (1) and (2) above within a limited development period and number of experiments.

[0008] The present invention has been made in consideration of the above circumstances, and aims to provide a method, an apparatus, and a program for searching for a medium composition and estimating cell characteristics, which can efficiently optimize the culture results even for cells with unknown characteristics.

[0009] In order to achieve the above-mentioned object, a method according to a first aspect of the present invention is a method for searching for medium compositions for cells and estimating cell characteristics, which includes a processor-implemented prediction step of predicting a culture outcome for a given medium composition using the cellular characteristics or estimated values ​​of the target cell for a target cell; an acquisition step of acquiring the culture outcome for the target cell for some or all of the given medium compositions; an estimation step of estimating the cellular characteristics of the target cell using the medium composition and the acquired or predicted culture results; an evaluation step of evaluating the need for further estimation of the cellular characteristics of the target cell from the acquired culture results; a search and determination step of searching for and determining candidate medium compositions that are medium compositions for which a culture result should be obtained next in the acquisition step, based on the need and the prediction step; and a control step of repeating the prediction step, acquisition step, estimation step, evaluation step, and search and determination steps until a predetermined termination criterion is met.

[0010] According to a first aspect, the culture results for a given medium composition are predicted using the cell characteristics or estimated values ​​of the target cell, and the culture results for the target cell for some or all of the given medium compositions are obtained. Cell characteristics are estimated using the medium composition and the obtained or predicted culture results, and the need for further estimation of the cell characteristics is evaluated from the obtained culture results. Based on this need and the predicted culture results, candidate medium compositions for which the next culture results should be obtained are searched for and determined, and the above steps are repeated until a predetermined termination criterion is met. In this way, by repeating the prediction and acquisition of culture results, estimation of cell characteristics, evaluation of the need for further estimation of cell characteristics, and search and determination of candidate medium compositions, the culture results can be efficiently optimized even for cells with unknown characteristics.

[0011] In the first embodiment and each of the following embodiments, the term "given medium composition" refers to a "specific medium composition" or "any medium composition," and the amount of each component is clearly defined. This "given medium composition" is not limited to a "known" medium in the sense of being widely used or already being prepared, but may also be a "novel" medium that is about to be prepared, i.e., has not yet been prepared. However, an "unknown" medium in which the amount of components is unclear is not included in the "given medium composition."

[0012] In the acquisition step, for example, actual culture results may be acquired in a culture evaluation system (experimental system), or culture results may be acquired by simulation or the like without actually performing culture.

[0013] In the method according to the second aspect, in the first aspect, the processor performs the evaluation in the evaluation step under the assumption that there is continuity in the change in the culture result in response to a change in the medium composition. In the second aspect, the evaluation is performed under the assumption that "the cell culture result does not fluctuate greatly in response to a small change in the medium composition."

[0014] In the method according to the third aspect, in the first or second aspect, the processor performs the estimation in the estimation step by assuming that the cellular property of the target cell is similar to the cellular property of one or more cells in the known cell group. In the third aspect, the estimation is performed using the cellular property of the known cell group as "prior knowledge."

[0015] A fourth aspect of the method is any one of the first to third aspects, in which the processor performs the prediction step, estimation step, acquisition step, evaluation step, and search and determination step using a solution to a bandit problem that estimates the unknown function that best fits the observation history, regarding the culture results as an observation history, the prediction means in the prediction step as an unknown function, and the shape of the unknown function as a cell characteristic. A bandit problem is a problem of optimizing an unknown function based on the observation history in a limited number of trials, and the fourth aspect assigns specific components to the bandit problem.

[0016] In the method according to the fifth aspect, in the fourth aspect, the processor expresses the prediction means, the unknown function and / or a similar function to the unknown function in the prediction step by a cell simulation including a metabolic pathway model.

[0017] The method according to the sixth aspect is the same as that of the fifth aspect, except that the processor inputs culture data of the target cell, determines parameters of a metabolic pathway model from the culture data, and estimates the cell characteristics based on the determined parameters. The sixth aspect specifically defines the aspect of estimating the cell characteristics. The parameters of the metabolic pathway model constitute part of the cell characteristics of the present invention.

[0018] The method according to the seventh aspect is the same as that of the sixth aspect, wherein the culture data includes one or more of the total cell number of the target cells, the amount of cell secretory substances, the amount of cell-produced substances, the amount of cell metabolic substances, and the amount of medium components, and / or one or more of data on changes in the total cell number of the target cells over time, changes in the amount of cell secretory substances over time, changes in the amount of cell-produced substances over time, changes in the amount of cell metabolic substances over time, and changes in the amount of medium components over time, and the processor separates the parameters into input factors to the target cells and output factors from the target cells in the estimation step, and estimates the cell characteristics so that the input factors and output factors at any given time hold true. The seventh aspect further specifies the estimation of the cell characteristics in the sixth aspect. Note that in the seventh aspect, "so that the input factors and output factors hold true" means configuring the cell characteristics so that the output factors calculated as the input factors based on the cell characteristics are close to the actual output factors.

[0019] An eighth aspect of the present invention is a method according to any one of the first to seventh aspects, wherein in the prediction step, the processor predicts a biological behavior quantity of the target cell based on a given medium composition and culture conditions, modifies the cellular environment of the target cell based on the predicted biological behavior quantity, and repeats the prediction of the biological behavior quantity and the modification of the cellular environment based on the modified cellular environment, thereby predicting the culture outcome. The eighth aspect specifically defines one method of predicting the culture outcome.

[0020] The method according to the ninth aspect is the eighth aspect, wherein the processor predicts the change over time in the total number of cells, the change over time in the amount of a substance produced by the target cells, and the change over time in the amount of a substance contained in the culture environment of the target cells as biological behavior quantities, and changes the total number of the target cells and the amount of the substance produced by the target cells as the cellular environment. The ninth aspect further specifies the prediction according to the eighth aspect.

[0021] A tenth aspect of the present invention is a method according to any one of the first to ninth aspects, wherein the processor evaluates the degree of discrepancy between the estimated value of the cell characteristic and the true value based on the predicted culture result and the acquired culture result in the evaluation step, and searches for and determines candidate medium compositions based on the degree of discrepancy in the search and determination step. The tenth aspect specifically defines one method for evaluating the need for further characteristic estimation and for searching for and determining candidate medium compositions.

[0022] The method according to an eleventh aspect is the method according to any one of the first to ninth aspects, wherein in the evaluation step, a processor inputs pairs of medium compositions and culture results for the medium compositions to evaluate the sufficiency of the search, and in the search and determination step, searches for and determines candidate medium compositions based on the results of the evaluation of the sufficiency. The eleventh aspect specifically defines other methods for evaluating the need for further characteristic estimation and for searching for and determining candidate medium compositions.

[0023] A twelfth aspect of the method is any one of the first to eleventh aspects, wherein the processor, in the control step, determines that a termination criterion has been met and terminates the repetition when a difference between a predicted culture result and an obtained culture result and / or the number of repetitions satisfies a predetermined condition. The twelfth aspect specifically defines one aspect of the termination criterion for processing.

[0024] A thirteenth aspect of the present invention relates to the method of any one of the first to twelfth aspects, wherein the target cells are CHO cells, HEK cells, or iPS cells. The thirteenth aspect specifically defines an example of the target cells.

[0025] A fourteenth aspect of the method is any one of the first to thirteenth aspects, wherein the culture results are one or more of the amount of antibody or virus produced by the target cells, the proliferation of the target cells, and the success rate of inducing differentiation of the target cells into a specific tissue. The fourteenth aspect specifies an example of the culture results. In the present invention, it is preferable to use the culture results according to the type of target cells.

[0026] In order to achieve the above-mentioned object, an apparatus according to a fifteenth aspect of the present invention is an apparatus for searching for medium compositions for cells and estimating cellular characteristics, and includes a processor. The processor predicts a culture outcome for a given medium composition for a target cell using the cellular characteristics or estimated values ​​of the target cell, obtains the culture outcome for the target cell for some or all of the given medium compositions, estimates the cellular characteristics of the target cell using the medium composition and the obtained or predicted culture results, evaluates the need for further estimation of the cellular characteristics of the target cell from the obtained culture results, and searches for and determines candidate medium compositions that are the medium compositions for which the next culture result should be obtained based on the need and the prediction, and repeats the prediction, acquisition, estimation, evaluation, and search and determination until a predetermined termination criterion is met.

[0027] According to the fifteenth aspect, it is possible to efficiently optimize the culture results even for cells with unknown characteristics, similar to the first aspect. Note that the device according to the fifteenth aspect may have the same configuration as the second to fourteenth aspects.

[0028] In order to achieve the above-mentioned object, a sixteenth aspect of the present invention provides a program that causes a processor to search for medium compositions for cells and estimate cell characteristics, and causes the processor to execute the following steps: a prediction step of predicting a culture outcome for a given medium composition using the cellular characteristics or estimated values ​​of the target cell for a target cell; an acquisition step of acquiring the culture outcome for the target cell for some or all of the given medium compositions; an estimation step of estimating the cellular characteristics of the target cell using the medium composition and the acquired or predicted culture results; an evaluation step of evaluating the need for further estimation of the cellular characteristics of the target cell from the acquired culture results; a search and determination step of searching for and determining candidate medium compositions that are medium compositions for which a culture result should be obtained next in the acquisition step, based on the need and the prediction step; and a control step of repeating the prediction step, acquisition step, estimation step, evaluation step, and search and determination steps until a predetermined termination criterion is met.

[0029] According to the sixteenth aspect, similar to the first and fifteenth aspects, it is possible to efficiently optimize the culture results even for cells with unknown characteristics. The program according to the sixteenth aspect may have the same configuration as the second to fourteenth aspects. Non-transitory tangible recording media (e.g., various magneto-optical recording devices and semiconductor memories) on which computer-readable code of the programs of these aspects is recorded can also be cited as aspects of the present invention. The above-mentioned "non-transitory tangible recording media" does not include non-tangible recording media such as carrier signals or propagation signals themselves.

[0030] As described above, the method, device, and program for searching for medium compositions and estimating cell characteristics of the present invention make it possible to efficiently optimize culture results even for cells with unknown characteristics.

[0031] FIG. 1 is a diagram conceptually illustrating the workflow of the entire system. FIG. 2 is a block diagram illustrating the configuration of a search and estimation system according to a first embodiment. FIG. 3 is a block diagram illustrating the configuration of a processing unit. FIG. 4 is a diagram conceptually illustrating the configurations of the search and estimation system and the culture evaluation system. FIG. 5 is a flowchart illustrating the processing of medium composition search and cell characteristic estimation according to the first embodiment. FIG. 6 is a conceptual diagram for explaining the evaluation of the need for additional estimation of medium composition. FIG. 7 is a diagram illustrating a phylogenetic tree of CHO cells. FIG. 8 is a diagram illustrating the utilization of prior knowledge in search. FIG. 9 is a diagram illustrating the processing of a search based on history. FIG. 10 is a diagram illustrating an example of search results for the DG44 strain. FIG. 11 is a diagram illustrating an example of search results for the K1 strain. FIG. 12 is a diagram illustrating the configuration of a culture result prediction unit. FIG. 13 is a flowchart illustrating the processing of culture result prediction. FIG. 14 is a diagram illustrating an overview of a simulator for the physical behavior quantity prediction process. FIG. 15 is a schematic diagram illustrating a cellular metabolic pathway. FIG. 16 is a diagram illustrating an overview of the Michaelis-Menten equation. FIG. 17 is a diagram illustrating an overview of a conventional cell metabolism model. FIG. 18 is a schematic diagram illustrating the creation of a cell death model. Fig. 19 is a diagram showing the configuration of a cell property estimation unit. Fig. 20 is a flowchart showing a method for creating a mathematical cell model. Fig. 21 is a diagram explaining a feature extraction step. Fig. 22 is a flowchart showing a mathematical model creation step. Fig. 23 is a diagram explaining the mathematical model creation step.

[0032] The following describes in detail embodiments of a method, an apparatus, and a program for searching for a medium composition and estimating a cell characteristic according to the present invention. In the description, reference is made to the accompanying drawings as necessary.

[0033] [Processing Overview] The following describes an example of exploring the composition of a medium for antibody-producing CHO cells, with the productivity targeted for optimization being the antibody production volume. However, the target cells in the present invention may be CHO cells, HEK cells that produce viruses for gene therapy, or iPS cells for regenerative medicine. The target productivity may also be one or more of the target cell proliferation, antibody production volume or virus production volume, and success rate of inducing differentiation of the target cells into a specific tissue. Even when these target cells or productivity are selected, embodiments and examples similar to those described below can be constructed. Furthermore, the culture method can be appropriately selected from well plates (microtiter plates, microplates), flasks, batch, fed-batch, perfusion culture, etc.

[0034] Here, a medium is a mixture of nutrients for culturing cells, generally including, but not limited to, amino acids, vitamins, and metals. Therefore, it can be expressed as a multidimensional vector (or a point in a multidimensional space (hereinafter referred to as "medium composition space") in which each dimension corresponds to a nutrient). In the present invention, the amount and content ratio of part or all of the composition (nutrients) of such a medium can be optimized. Note that the medium composition space is generally multidimensional (the number of dimensions varies depending on the number of nutrients), but in the following explanation, it may be shown in one dimension for simplicity (such as the horizontal axis in Figures 6 and 8 described below).

[0035] [Workflow of the Present Invention] Figure 1 shows the workflow (overview of the process) envisioned in the present invention. A culture medium presenting party (e.g., a search and estimation system 10 described below) presents multiple culture medium candidates (step S1), and an evaluator (e.g., a user of the culture evaluation system 600 described below) actually cultures certain cells (target cells) in the candidate culture media, obtains the culture results, and evaluates the results (step S2). The evaluator (e.g., the culture evaluation system 600 itself) may obtain and evaluate the culture results by simulation or the like, rather than by actual culture (step S2). The evaluation results are notified to the presenting party (step S3), and this cycle is repeated.

[0036] In such a situation, the objective of the entire system is to obtain the best possible culture results while limiting the number of culture medium sets and repetitions presented, or within those constraints. For example, steps S1 to S3 constitute one round, and five culture media are presented and evaluated, with the number of rounds being approximately 3 to 5. However, other conditions may also be used. When repeating rounds, the culture media included in the presented set may be changed depending on the evaluation results.

[0037] 1 , for example, the medium refers to a medium for CHO cells (CHO: Chinese Hamster Ovary), the cells refer to certain CHO cells (including not only original CHO cells but also various lineages of CHO cells generated by mutation, modification, etc., such as CHO-DG44, CHO-K1, and CHO-S), and the culture result refers to the amount of antibody production obtained after culturing the cells for a predetermined period of time. However, as described above, the medium, cells, and culture result of the present invention are not limited to these examples.

[0038] [First Embodiment] FIG. 2 is a block diagram showing the configuration of a search and estimation system 10 and related systems according to a first embodiment of the present invention. The search and estimation system 10 is a system (search and estimation system) that searches for medium compositions for cells and estimates cell characteristics, and can be implemented using a computer. As shown in FIG. 2, the search and estimation system 10 includes a processing unit 100 (processor), a storage unit 200, a display unit 300, and an operation unit 400, which are connected to each other to transmit and receive necessary information. These components can be installed in various ways, and each component may be installed in a single location (e.g., in a single cabinet or room) or in separate locations and connected via a network. The search and estimation system 10 can also be connected to an external server 500 and an external database 510 via a network NW such as the Internet to acquire information on cell characteristics, medium compositions, cell lines, and the like as needed.

[0039] The culture evaluation system 600 is a system that acquires, evaluates, and manages cell culture data, and is connected to the search and estimation system 10 via a network NW. As will be described later, the search and estimation system 10 presents a plurality of candidate culture media, and a user of the culture evaluation system 600 actually cultures cells (target cells) using the candidate culture media to obtain cell culture results. The culture evaluation system 600 may also obtain cell culture results by performing simulations or the like that do not involve actual culture. The search and estimation system 10 obtains cell culture results for a given medium composition from the culture evaluation system 600.

[0040] [Configuration of Processing Unit] Fig. 3 is a block diagram showing the configuration of the processing unit 100 (processor). The processing unit 100 includes a culture result prediction unit 102, a cell characteristic estimation unit 104, a search and estimation unit 110, an input / output control unit 120, a communication control unit 122, a ROM 130 (Read Only Memory), and a RAM 140 (Random Access Memory). The search and estimation unit 110 includes a culture result acquisition unit 112, an additional estimation evaluation unit 114, a medium composition presentation unit 116, and a repetition control unit 118. The input / output control unit 120 and the communication control unit 122 control input / output and communication with an external server 500, an external database 510, a culture evaluation system 600, etc. Search and estimation processes using these units of the processing unit 100 will be described in detail below.

[0041] The functions of each unit of the processing unit 100 described above can be realized using various processors. The various processors include, for example, a CPU (Central Processing Unit), which is a general-purpose processor that executes software (programs) to realize various functions. The various processors also include a programmable logic device (PLD), such as an FPGA (Field Programmable Gate Array), whose circuit configuration can be changed after manufacturing. Furthermore, the various processors also include dedicated electrical circuits, such as an ASIC (Application Specific Integrated Circuit), which is a processor with a circuit configuration designed specifically to execute specific processing.

[0042] The functions of each unit may be realized by a single processor or by a combination of multiple processors. Furthermore, multiple functions may be realized by a single processor. Examples of multiple functions configured by a single processor include: a first configuration, as typified by client and server computers, in which a single processor is configured by combining one or more CPUs and software, and this processor realizes multiple functions; a second configuration, as typified by system-on-chip (SoC), in which a processor is used to realize the functions of the entire system on a single IC (Integrated Circuit) chip; and various functions are thus configured as a hardware structure using one or more of the various processors described above. Furthermore, the hardware structure of these various processors is, more specifically, an electrical circuit (circuitry) that combines circuit elements such as semiconductor devices.

[0043] When the processor or electrical circuit described above executes software (programs), processor-readable code for the software to be executed is stored in a non-transitory, tangible recording medium such as ROM 130, and the processor references the software. The software stored in the non-transitory, tangible recording medium includes a program (a program for searching for a medium composition for cells and estimating cell characteristics; hereinafter referred to as the "search and estimation program") for executing the method of the present invention (a method for searching for a medium composition for cells and estimating cell characteristics; hereinafter referred to as the "search and estimation method"). The code may be recorded in a non-transitory, tangible recording medium such as a magneto-optical recording device or semiconductor memory, instead of ROM 130. When processing using the software, for example, RAM 140 is used as a temporary storage area, and data stored in, for example, an EEPROM (Electronically Erasable and Programmable Read Only Memory), not shown, may also be referenced.

[0044] It should be noted that the above-mentioned "non-transitory and tangible recording medium" does not include non-tangible recording media such as carrier signals or propagation signals themselves.

[0045] [Configuration of the storage unit] The storage unit 200 is composed of various types of magneto-optical recording media, non-transitory and tangible recording media such as semiconductor memory, and its input / output control unit, and stores information such as medium composition, cell characteristics, and culture results.

[0046] [Configuration of Display Unit and Operation Unit] The display unit 300 is equipped with a monitor 310 (display device), and can display input information, information stored in the storage unit 200, results of processing by the processing unit 100, etc. The operation unit 400 includes a keyboard 410 and a mouse 420 as input devices and / or pointing devices, and a user can perform operations required to execute the search and estimation method and search and estimation program according to the present invention via these devices and the screen of the monitor 310.

[0047] [Processing of the Search and Estimation Method] Fig. 4 is a diagram conceptually showing the configurations of the search and estimation system 10 and the culture evaluation system 600, and Fig. 5 is a flowchart showing the processing of the search and estimation method according to the first embodiment. Hereinafter, the processing of the search and estimation method and search and estimation program according to the first embodiment will be described with reference to these figures. Note that, to simplify the description, illustration of some elements of the search and estimation system 10 will be omitted as necessary.

[0048] In the first embodiment, antibody-producing CHO cells are the subject, and an example will be described in which the composition of a medium is explored when the productivity to be optimized is the amount of antibody produced.

[0049] Here, the medium composition is X i Let C be the cell characteristic, and let Y be the antibody production amount. i (The subscript i indicates that there may be multiple cell types, and it is assumed that there is only one type of cell type in the workflow.) X i is freely configurable, C is unknown, and Y i The true value of Y can only be obtained through experiments. i Optimize (maximize or make as large as possible) X i The challenge is to find the answer, and to solve this problem, we will build a system that combines the following elements.

[0050] In the first embodiment, the target of optimization is Y i is the amount of antibody produced, while Y i Other experimental data may be appropriately attached to the cell characteristics estimation unit 104 (processing unit 100; processor) (for example, other associated data such as the time series of cell counts and antibody production levels, changes in culture medium composition, etc.) Here, "experimental data" refers to, for example, the time series of cell counts and antibody production levels, changes in culture medium composition, etc., and the cell characteristics estimation unit 104 (processing unit 100; processor) can use this data to estimate cell characteristics. In other words, the cell characteristics estimation unit 104 can estimate cell characteristics using not only the culture results but also intermediate experimental data (see "Specific Configuration and Processing Examples of the Cell Characteristics Estimation Unit" described below).

[0051] [Prediction of Culture Results] The culture result prediction unit 102 (processor) predicts the culture results of the target cells for a given medium composition (step S100: prediction step). i ＾ =F(X i , C), that is, the culture result prediction unit 102 predicts the culture medium composition X i and cell characteristics C, the antibody production amount Y i ＾ Here, the superscript symbol "^ (hat)" indicates that it is a predicted value (same below). However, in practice, the true cell characteristic C is unknown, so instead, the estimated value of the cell characteristic C ^ Enter.

[0052] The above-mentioned F may be a function, or may be a predictor obtained by machine learning, or a model-based simulator. When a simulator is used, a simple simulation capable of high-speed calculation may be performed. An example of the specific configuration and processing of such a culture result prediction unit 102 will be described in detail later (see "Example of specific configuration and processing of the culture result prediction unit"). Note that the true value "Y i " is also considered a type of F (which includes the true C) and is called F0.

[0053] [Presentation of Medium Composition] The medium composition presentation unit 116 (processor) presents medium composition X i is presented to the culture evaluation system 600, and the user of the culture evaluation system 600 selects the presented medium composition X i The culture evaluation system 600 may actually culture the target cells and obtain the culture results. Alternatively, the culture evaluation system 600 may obtain the culture results by simulation (e.g., high-precision simulation). The culture evaluation system 600 provides the culture results obtained by actual culture or simulation to the search and estimation system 10.

[0054] [Acquisition of Culture Results] The culture result acquisition unit 112 (processor) acquires the culture results of the target cells for some or all of the given medium compositions (Step S110: Acquisition step). i and antibody production Y ipair ({X in Figure 4) i , Y i}; hereafter also referred to as "history R").

[0055] [Estimation of Cellular Properties] The cellular property estimation unit 104 (processor) estimates the cellular properties of the target cells using the medium composition and the culture results predicted in step S100 (step S110: estimation step). Here, while "cellular properties" can be defined in various biological or cytological ways, in the present invention, "cellular properties" are defined as "a mechanism that determines the productivity of a cell (such as the cell proliferation rate and antibody production amount described above) when a medium composition is given." Specifically, in this embodiment, the "cellular properties" are a mechanism that determines the antibody production amount Y for a medium composition X, and can be expressed, for example, by a function, a machine learning model, or a mathematical cellular model. Therefore, for example, the coefficients of a function, the parameters of a machine learning model, or the parameters of a metabolic pathway model in a mathematical cellular model can be considered part of the cellular properties in the present invention.

[0056] The estimation of the cell characteristic by the cell characteristic estimation unit 104 in step S110 is ^ =G({X i , Y i}) That is, the cell characteristic estimation unit 104 can be expressed as one or more "medium composition X i and the antibody production amount Y i Given a pair of Y (the "history R" mentioned above), cell properties are estimated. i As mentioned above, G may be accompanied by experimental data. Here, G is a function F, for example, depending on the cell characteristics mentioned above. ^ =F_C ^ It can also be a function that returns (X) ^ (It can be considered that the cell property estimation unit 104 returns the parameter value of the cell property estimation unit 104.) Furthermore, the cell property estimation unit 104 may be a machine learning parameter determination unit, a model-based model estimator, or the like. Examples of the specific configuration and processing of the cell property estimation unit 104 will be described in detail below (see "Examples of specific configuration and processing of the cell property estimation unit").

[0057] Here, finally, when some history R is obtained, the history R is first input to G (cell characteristic estimation unit 104) to obtain the cell characteristic C. ^ Then, F (culture result prediction unit 102) is calculated by adding a plurality of medium composition candidates {X i}, the medium composition X that returns the maximum antibody production amount Y is the optimal medium. However, if we assume repetition, the next medium composition that should obtain the culture results is the medium composition candidate {X i Therefore, in the first embodiment, the following element is further introduced into the search and estimation system 10:

[0058] [Evaluation of the necessity of additional estimation] The additional estimation evaluation unit 114 (processor) evaluates the necessity of further estimating the cell property (step S130: evaluation step). The additional estimation evaluation unit 114 can evaluate the necessity of the additional property by, for example, the following method.

[0059] (Evaluation method 1) Evaluation method 1 is L = V1({Y i ^ , Y i That is, the additional estimation evaluation unit 114 calculates the predicted value Y i ^ (Predicted culture result) and the true value Y of antibody production by culture i By inputting pairs of Y (which indicate the difference between the predicted culture results and the obtained culture results; hereafter also referred to as "actual difference"), the degree of discrepancy between the estimated value and the true value of the cell characteristics can be evaluated. If Y i ^ Y i If it is close to C ^ It would be reasonable to judge that C is also close to C. ^ A function F that incorporates ^ The degree of deviation between the true unknown function F0 and the calculated function F0 may be evaluated.

[0060] The additional estimation evaluation unit 114 can evaluate that "the need for additional characteristics is high" when the evaluated deviation is large (exceeding a predetermined deviation), and can evaluate that "the need for additional characteristics is low" when the evaluated deviation is small (equal to or less than the predetermined deviation). If the evaluation is that "the need for additional characteristics is low," the repetition control unit 118 can determine in step S140 that "the termination criterion is satisfied" and terminate the repetition.

[0061] (Evaluation Method 2) Evaluation Method 2 is L = V2({X i , Y i}) That is, the additional estimation evaluation unit 114 can be expressed as the history R (medium composition X i and medium composition X i Culture results in Y i By entering a value (e.g., a pair with antibody production), you can evaluate the sufficiency of the search. i} already sufficiently covers the distribution of possible candidates for the medium composition X in the medium composition space described above, the additional estimation evaluation unit 114 may evaluate that it is difficult to bring the estimated value of the cell characteristic closer to the true value any further.

[0062] Furthermore, in evaluation method 2, the additional estimation evaluation unit 114 can perform evaluation on the assumption that "the change in the culture result in response to the change in the medium composition is continuous." This is because, in most cases, it can be assumed that the cell culture result does not fluctuate greatly in response to a small change in the medium composition. Therefore, the additional estimation evaluation unit 114 can perform evaluation on the assumption that "in the above-mentioned medium composition space, X i For changes in Y i Areas where the change in X is small are evaluated as having little need for additional estimation, and conversely, areas where the change in X is small are evaluated as having little need for additional estimation. i For changes in Y i For example, the additional estimation evaluation unit 114 may evaluate that the area around "High" in FIG. 6 (a conceptual diagram for explaining the need for additional estimation; the horizontal axis indicates the medium composition space, and the vertical axis indicates the culture result) is "X i For changes in Y iThe area marked "low" can be evaluated as "the change in Yi is small relative to the change in Xi, so the need for additional estimation is low."

[0063] [Searching for and Determining the Next Candidate Medium Composition] The repetitive control unit 118 (processor) causes the medium composition presentation unit 116 (processor) to search for and determine the candidate medium composition, which is the medium composition from which the next culture result should be obtained, based on the necessity for additional estimation evaluated in step S130 and the prediction results in the prediction process (steps S140, S150: control process, search and determination process).

[0064] Specifically, if no further estimation is necessary (YES in step S140: control step), the termination criterion is met, and the repetitive control unit 118 (processor) terminates the process. If further estimation is necessary, the termination criterion is not met, and the repetitive control unit 118 causes the medium composition presenting unit 116 (processor) to search for and determine a candidate medium composition that is the medium composition from which the next culture result should be obtained (NO in step S140: control step, search and determination step).

[0065] In step S140, the repetition control unit 118 may determine that the "termination criteria have been met" and terminate the repetition if the number of repetitions exceeds a predetermined number.

[0066] The medium composition presentation unit 116 can search for and determine a medium composition candidate as follows, depending on the evaluation method used by the additional estimation evaluation unit 114. Specifically, when the additional estimation evaluation unit 114 uses evaluation method 1, the medium composition presentation unit 116 prioritizes and sets the medium composition candidate X in a range (region) in the medium composition space that has not yet been sufficiently searched if the degree of deviation is large, and prioritizes and sets the medium composition candidate X that is expected to produce a larger culture result Y from the current estimation result if the degree of deviation is small. Furthermore, when the additional estimation evaluation unit 114 uses evaluation method 2, the medium composition presentation unit 116 prioritizes and sets the medium composition candidate X in that range (X i For changes in Y iConversely, if the need for additional estimation is low, priority is given to setting X where Y is expected to be large based on the current estimation results.

[0067] [Effects of the First Embodiment] According to the search and estimation system 10 (search and estimation method, search and estimation system, search and estimation program) of the first embodiment, by repeating the processes of predicting and obtaining culture results, estimating cell characteristics, evaluating the need for additional estimation of cell characteristics, and searching for and determining candidate medium compositions, it is possible to efficiently optimize culture results even for cells with unknown characteristics.

[0068] [Additional Embodiments Related to the Method for Predicting Cellular Properties] An additional embodiment related to the method for predicting cell properties will be described. For example, even in the case of CHO cells, it is known that various CHO cell lines are not "equally spaced" from one another, and that relationships such as a phylogenetic tree can be modeled based on derivation relationships, etc. Specifically, as shown in Figure 7, there are multiple CHO cell lines with different properties, such as "CHO-K1," "CHO-S," and "CHO-DG44," relative to the original CHO cell (CHO-ori).

[0069] The phylogenetic tree in FIG. 7 is described in Non-Patent Document 1 below.

[0070] [Non-Patent Document 1] "Wurm, FM; Wurm, MJ Cloning of CHO Cells, Productivity and Genetic Stability A Discussion. Processes 2017, 5, 20." [Retrieved March 1, 2022], Internet (https: / / www.researchgate.net / publication / 316314849_Cloning_of_CHO_Cells_Productivity_and_Genetic_Stability-A_Discussion / link / 59410f0045851566e1beae32 / download). The above paper is cited in accordance with http: / / creativecommons.org / licenses / by / 4.0 / .

[0071] The above-described phylogenetic tree-like relationship is not limited to CHO cells; HEK cells also have phylogenetic trees, and such relationships may also emerge with iPS cells through long-term culture. In the present invention, "cells" include, for example, host cells containing a certain gene, generally established by culture or modification, and antibody-producing cells into which an antibody-producing gene has been inserted (sometimes referred to as "cell lines"). Such lineages and cell lines may be defined by genes or by reference to commonly known lineages (e.g., "CHO-K1," "CHO-S," "CHO-DG44," etc., in the case of CHO cells). While various definitions of "cell line" exist, it can be defined, for example, as "cells isolated from a living organism or cells that have been modified in some way, such as genetically, so that they can be stably grown and cultured for long periods of time while maintaining certain properties."

[0072] Based on this definition of cell lineage, we can grasp the cellular characteristics of various available CHO cell lines by some means and construct a set of cellular characteristics Cs such as {C1, C2, C3, ...} in advance. Then, even if a cell has an unknown cellular characteristic C, it is expected that the cell belongs to one of the cell lineages and therefore has characteristics similar to those of a representative cell of one of the cell lineages. In other words, if the cellular characteristics of the CHO cell line are known, we can assume that the true cellular characteristic C of the unknown cell will likely be similar to one of the cellular characteristics included in the set Cs (this assumption is called "prior knowledge"). Furthermore, especially if the unknown cell is known to be a derivative of a specific cell lineage, we can assume that C is similar to C_k (the known information that C belongs to a specific cell lineage is called "prior knowledge").

[0073] Here, the notation using an underscore, such as "C_k," means that a cell characteristic is set using prior knowledge "k" for a certain piece of information. In the above example, the prior knowledge is that "a certain unknown cell is close to a specific cell characteristic Ck among the cell characteristics included in the set Cs."

[0074] In this case, for example, the initial value of C can be set as C = C_k. Alternatively, in a system that incorporates C, F may be initially set as F = F_k. For example, in FIG. 7, even if it is unknown that a certain cell is CHO-Sc, if it is known that it is a CHO-S0 derived cell line, by initially setting "C = C_S0" or "F = F_S0", it is expected that the search will be more efficient than if the initial setting is "C = C_ori". The same method as in the first embodiment described above can be applied to the subsequent processing.

[0075] In this way, the cell property estimation unit 104 (processor) can perform search and estimation by assuming that the cell property of the target cell is similar to the cell property of one or more cells in the known cell population (i.e., by utilizing prior knowledge). Such a search utilizing prior knowledge is shown in Figure 8. In Figure 8, an unknown function is estimated. The gray-shaded area (its width in the vertical axis direction) indicates the estimated range of the cell property for a certain medium composition (a point on the horizontal axis). Note that while the medium composition actually corresponds to a point in the medium composition space, which is a multidimensional space, in Figure 8 it is shown as a point in a one-dimensional space (the horizontal axis of Figure 8) for convenience. When estimating the cell property, the fact that the known function is close to the unknown function (prior knowledge) is utilized.

[0076] [Aspects Applying a Solution to Bandit Problems] A specific embodiment will now be described by applying a bandit problem. A bandit problem is a problem of optimizing an unknown function based on an observation history within a limited number of trials. In the present invention, specific components such as cells, cell characteristics, and medium composition are assigned to such generalized problems, and domain-specific elements (e.g., evaluation of the need for additional estimation assuming continuity of results, utilization of prior knowledge regarding similarities in characteristics, etc.) are also introduced. Specifically, in this aspect, the culture results are considered as the observation history, the prediction means in the prediction step as the unknown function, and the shape of the unknown function as the cell characteristics, and the unknown function that best fits the observation history is estimated.

[0077] Specifically, for example, using GP-UCB (Gaussian Process Upper Confidence Bound), Fgp_t(X, C ^) = {μ_t(X) + σ_t(X), μ_t(X) - σ_t(X)} (see Figure 9). Here, Fgp is a function that configures the width (interval) of F, and μ is the response prediction value of F (i.e., F ^ ), σ indicates the uncertainty of the forecast, and F ^ are updated sequentially, so they are different at the time t of a trial. Also, t(X), σ_t(X), t(X, C ^ ) are the uncertainties of the medium composition X, the cell characteristic C, and the estimated value C of the cell characteristic for the medium composition X in the t-th trial, respectively. ^ is.

[0078] Among various medium compositions X, the medium composition X that maximizes Fupper_t(X_C) = μ_t(X) + σ_t(X) (i.e., the upper limit value of the interval) is one of the medium composition candidates to be presented by the medium composition presentation unit 116. The medium composition presentation unit 116 can select the medium composition X that maximizes Fupper in each of a plurality of intervals that are set randomly or appropriately, for example, as a presentation set (medium composition candidates to be presented). In this way, the medium composition presentation unit 116 can be configured. In GP-UCB, a certain history {X i , Y i}, based on F(X i , C) = μ_t(X i ) = Y i Fix and then X i The surrounding σ_t is reduced. This allows the cell property estimation unit 104 and the additional estimation evaluation unit 114 to be configured. That is, Fgp is updated, and the candidates for the range to be searched are also updated.

[0079] Furthermore, for example, if the function for a known cell is set as F(X, C_k) = μ_k(X), then F_0(X, C) = μ_k(X) can be initially set, and the assumption of similarity in cell properties mentioned above can also be incorporated.

[0080] Figure 9 illustrates a search using GP-UCB (a diagram showing the prediction state after a certain number of searches have been performed). The horizontal axis of the figure corresponds to the medium composition, and the vertical axis corresponds to the estimated value of the cell characteristic. In the figure, the solid line indicates the predicted value of the cell characteristic for a certain medium composition, and the gray area indicates the uncertainty of the predicted value. Crosses indicate the observation results up to that point, and circles indicate the next search estimation candidate. Part (a) of Figure 9 illustrates a state in which the medium composition corresponding to the circle has been determined as the search estimation candidate. The medium composition presentation unit 116 selects the point at which the cell characteristic is maximized or maximized as the search estimation candidate. Part (b) of Figure 9 illustrates a state in which the observation value corresponding to the circle in part (a) has been obtained, and the uncertainty in that part (the dotted circle) has converged. In this state, the medium composition presentation unit 116 can further determine the medium composition corresponding to the circle (the point at which the cell characteristic is maximized or maximized) as the next search estimation candidate.

[0081] A cell simulator or its analogous function may be applied to the above-described culture result prediction unit 102. When a cell simulator is applied, a cell simulation including a metabolic pathway model can be used to represent a prediction means in the prediction step, an unknown function, and / or an analogous function to the unknown function (see "Examples of specific configurations and processing of the culture result prediction unit" described later).

[0082] [Example] The search effect of the present invention is demonstrated by simulation. Here, a cell simulator (culture outcome prediction unit 102) was configured using the method described later in "Example of specific configuration and processing of culture outcome prediction unit." In the following example, DG44 is the original DG44 cell line of CHO cells, X is a cell line derived from DG44, and K is the original K cell line (see the phylogenetic diagram in Figure 7 for the relationship between these CHO cell lines).

[0083] In this example, we investigated the effectiveness of the search method using the fourth aspect of the present invention (a method using a bandit problem solution) for the following cases: (1) when the unknown cell is DG44 (the same strain as X) and (2) when the unknown cell is K. Furthermore, we constructed functions from a simulator for each of the following cases: (a) with prior knowledge X and (b) without prior knowledge X. We then investigated whether a medium composition superior to the reference medium could be discovered in five short trials, with five media presented at a time. The results are shown in Figures 10 and 11. Figure 10 shows the results when the unknown cell is DG44, and Figure 11 shows the results when the unknown cell is K. In Figures 10 and 11, the horizontal axis represents the number of search trials, and the vertical axis represents the amount of antibody production (an example of a cell characteristic).

[0084] As a result, in both cases, "when prior knowledge X is used" and "when prior knowledge X is not used," as the number of trials increased, it was possible to successfully discover a medium composition that could produce antibody amounts exceeding those of the standard medium through limited exploration, and it was confirmed that utilizing prior knowledge was particularly effective for cell lines of the same lineage.

[0085] [Specific Configuration and Processing Example of Culture Result Prediction Unit] A specific configuration and processing example of the above-mentioned culture result prediction unit 102 (processor) will be described. Fig. 12 is a diagram showing a configuration example of the culture result prediction unit 102. In the example shown in the figure, the culture result prediction unit 102 includes a culture environment input unit 102A, a biological behavior amount prediction unit 102B, a cellular environment change unit 102C, a repeat unit 102D, and an output unit 102E. Each of these units can be configured using various processors and memories (ROM, RAM, etc.) in the same manner as described above for the processing unit 100.

[0086] <<Method for Predicting Cell Cultivation Outcome>> Figure 13 is a flowchart showing a method for predicting the culture outcome of a target cell using the culture outcome prediction unit 102 configured as described above. This method includes a culture environment input step of inputting a culture environment consisting of a medium composition and culture conditions for culturing the cells, a biological behavior amount prediction step of predicting a biological behavior amount based on the culture environment input in the culture environment input step, and a cellular environment modification step of modifying the cellular environment based on the biological behavior amount predicted by the biological behavior amount prediction step. The method also includes a repeating step of repeating the biological behavior amount prediction step of predicting a biological behavior amount based on the cellular environment modified by the cellular environment modification step and the cellular environment modification step until a predetermined time, which is the culture time, has elapsed. The method also includes an output step of outputting the cellular environment modified by the cellular environment modification step after the predetermined time has elapsed.

[0087] Each step will be described below.

[0088] <<Culture Environment Input Step (Step S12)>> The culture environment input unit 102A performs the culture environment input step (step S12). The culture environment input step is a step of inputting a culture environment consisting of a medium composition and culture conditions for culturing cells. The culture environment is input by a user and processed by the culture environment input unit 102A.

[0089] The medium composition includes the medium components and the amounts of the medium components in the medium in which the cells are cultured. The culture conditions are set conditions for optimizing the cell culture process, and include the culture method, the size and type of culture vessel, oxygen addition, supplementation of medium and nutrient sources, addition of pH adjusters and carbon dioxide, removal of medium containing growth-inhibitory by-products, and harvesting of the target product.

[0090] Examples of the culture method include whether or not sterilization is performed, whether the culture medium is liquid or solid, the culture temperature, etc. Culture conditions can be set not only to the conditions at the start of culture, but also to include whether or not oxygen is added during the culture period and the amount of oxygen added, whether or not medium and nutrient sources are replenished and the amount of replenishment, addition of a pH adjuster and carbon dioxide, whether or not medium containing growth-inhibiting by-products is removed and the amount of replenishment, and whether or not the target product is harvested and the amount of harvest.

[0091] <<Biological Behavior Amount Prediction Step (Step S14)>> The biological behavior amount prediction unit 102B performs the biological behavior amount prediction step (step S14). The biological behavior amount prediction step is a step of predicting a biological behavior amount based on the culture environment input in the culture environment input step. The biological behavior amount is the change over time in the total number of cells, the change over time in the substances and amounts produced by the cells, and the change over time in the substances and amounts contained in the culture environment. The substances produced by the cells include antibodies produced by the cells, etc. The substances contained in the culture environment include medium components and by-products, etc.

[0092] Fig. 14 is a diagram showing an outline of a simulator for the biological behavior quantity prediction step. As shown in Fig. 14, the biological behavior quantity prediction step is performed using an action mechanism model (biological action mechanism model) and a machine learning model. The action mechanism model uses separate mathematical models for the inside and outside of the cell. For the inside of the cell, the absorption rate F Ａ is used to calculate the cell growth rate F ｇ , and the antibody production rate F ｐ A cell metabolism model is used to calculate the change in the concentration of each component of the culture medium. For the outside of the cell, a culture medium model is used to calculate the change in the concentration of each component of the culture medium. For cell death, a model derived by machine learning is used.

[0093] First, we will explain the cellular metabolism model that operates inside cells. The cellular metabolism model can calculate the cell growth rate and antibody production rate by using a cell culture simulation method that reproduces the mechanism and bioprocess of the cultured cells. The cell culture simulation method can be performed using a modeling approach that includes metabolic flux analysis (MFA) using a genome-scale metabolic model or flux balance analysis (FBA).

[0094] "Flux" or "metabolic flux" refers to the rate at which molecules pass through a pathway or reaction of interest. Factors that control flux include the catalytic rate of the enzyme in the pathway, the availability (durability) of the substrate, the concentration of the enzyme in the cell, and the proximity of the enzyme in the pathway. "Metabolic flux analysis" is a method for determining the amount of molecules moving based on these factors. "Flux balance analysis" is an analytical method that focuses on stoichiometry and metabolic flow. Even when metabolic constants cannot be fully measured, it analyzes the behavior range and characteristics of the target metabolic circuit based on the structure of metabolic reactions and basic constraints such as the law of conservation of mass.

[0095] Next, intracellular metabolism will be explained using Figure 15. Figure 15 is a schematic diagram showing an example of a metabolic pathway of a cell. For the cell metabolism model, first, the substrates used for metabolism in the culture medium and the uptake constraint conditions, which are the upper and lower limits for uptake of nutrients, etc., are obtained. In Figure 15, substances produced by metabolism are indicated by A, B, C, .... １ , F ２ , F ３ , ... are functions that indicate the time change in the concentration of each substance, for example, F １ is the flux of substance A taken up, and F ２ is the flux of substance B metabolized.

[0096] In the example shown in Figure 15, substance A is converted to substance B in equal amounts due to the law of conservation of mass. Similarly, the amount of substance B is equal to the sum of the amounts converted to substances C and E. In this way, the amount of substance converted is determined by the amount of substrate and nutrients initially taken up by the cells, etc. In addition, reaction rate is a parameter that limits the conversion of substances. For example, as described above, substance A is converted to substance B in equal amounts, but due to reaction rate limitations in a certain unit time, there is an upper limit to the conversion to substance B, and it may be that not all of substance A is converted to substance B. As an uptake constraint, a parameter that limits the conversion of this substance is obtained, and this uptake constraint is obtained using an equation that mathematically models the mechanism by which cells take up medium components, and is applied when calculating the culture results.

[0097] As a mathematical model for calculating this reaction rate, for example, the Michaelis-Menten equation can be used. A schematic diagram of the Michaelis-Menten equation is shown in FIG. 16. The Michaelis-Menten equation is an equation relating to the enzyme reaction rate V. When the substrate concentration S is low, the reaction rate V is proportional to the substrate concentration S, and when the substrate concentration S is high, the reaction rate V converges to a maximum rate Vmax regardless of the substrate concentration S. Another mathematical model that can be used is Fick's law. Fick's law is an equation for calculating the flux, which is the amount passing through a unit area per unit time, and is proportional to the diffusion coefficient D and the substrate concentration gradient on both sides of the membrane. In this embodiment, by using the Michaelis-Menten equation or Fick's law as a constraint condition for taking in the medium, simulations can be performed without calculating conditions for a reaction rate that is faster than the reaction rate and flux calculated by the mathematical model but is within the range of the conversion amount according to the law of conservation of mass. This allows for more accurate culture predictions.

[0098] The simulation calculations may use multiple mathematical models that mimic the state of cells in parallel. For example, a mathematical model for cell proliferation and a mathematical model for bioproduction without cell proliferation can be used in parallel, and the ratio between the multiple models can be changed depending on the culture state, such as the medium concentration.

[0099] Next, we will explain the culture medium model that operates outside the cell. The culture medium model is a model that calculates the concentration changes in the culture medium surrounding the cell when metabolism occurs under the conditions of the above-mentioned cell metabolic model, and a cell signaling model expressed as a differential equation can be used. The concentration changes of each component can be calculated by solving ordinary differential equations over time t using the Runge-Kutta method.

[0100] Here, a conventional cell metabolism model including a cell death model will be described. FIG. 17 is a diagram showing an outline of the conventional cell metabolism model. The main cell culture activities over the culture period can be divided into cell proliferation, antibody production, and cell death (cessation of activity). As shown in FIG. 17 , these activities include a proliferation phase in the early stage of culture, in which cells proliferate; an antibody production phase in the middle stage, in which the cell proliferation rate decreases and the antibody production rate increases; and a cell death phase in which cell activity ceases and cell death occurs. Conventional culture prediction simulations use a cell metabolism model to calculate these three culture activities, with a proliferation model for cell proliferation, a production model for antibody production, and a cell death model for cell death. However, the cell death model was implemented using a simplified model with constants, which did not achieve sufficient accuracy. Therefore, in this embodiment, as shown in FIG. 14 , a simulation is performed using a trained cell death model obtained by machine learning, thereby calculating culture prediction results with high accuracy.

[0101] FIG. 18 is a schematic diagram of creating a cell death model using machine learning. A random forest method can be used to create a trained model using machine learning. When constructing a cell death model using machine learning, the number of live cells, the concentrations of 20 amino acids, and the ammonia concentration are input as input data. Furthermore, the cell viability is input as output data. The data used is obtained by actually culturing cells, and, for example, 80% of the measured data is used for learning, and 20% is used for testing. This allows the creation of a cell death model.

[0102] Note that the creation of trained models using machine learning is not limited to cell death models. For data items that are difficult to model based on mechanisms, such as cell death, creating trained models using machine learning allows for more accurate calculation of culture prediction results. Examples of such data items include growth inhibition and antibody production suppression. When creating a growth inhibition model using machine learning, the cell growth amount (growth ratio) can be used as the output data. When creating an antibody production suppression model, the antibody production amount (production ratio) can be used as the output data.

[0103] In the biological behavior quantity prediction step of this embodiment, a trained model of cell death created by machine learning is used, which allows the biological behavior quantity to be measured with higher accuracy, and therefore allows the predicted results of culture to be calculated with high accuracy.

[0104] The biological behavior quantity prediction step (step S14) can be performed using a CHO cell simulator (a device or system that simulates the biological behavior of CHO cells using a mechanism of action model and a trained model of cell death created by machine learning) that includes these models, based on the culture conditions (medium component concentrations) and uptake constraint conditions acquired in the culture condition input step (step S12).

[0105] <Cellular environment changing step (step S16)> The cellular environment changing unit 102C performs the cellular environment changing step (step S16). The cellular environment changing step changes the cellular environment based on the biological behavior amount predicted by the biological behavior amount prediction step. The cellular environment refers to the total number of cells and the substances and amounts produced by the cells.

[0106] The biological behavior amount prediction step calculates the change over time in the number of cells after the culture time t has elapsed, the change over time in the substances and amounts produced by the cells, and the change over time in the substances and amounts contained in the culture environment. The cellular environment change step reflects the change in the number of cells after the culture time t has elapsed and the changes in the components in the culture medium in the cellular environment.

[0107] <<Repetition Step>> The repetition unit 102D controls the execution of the repetition step, which is a step of repeating a biological behavior amount prediction step (step S14) and a cellular environment modification step (step S16).

[0108] After the cell environment changing step (step S16) is completed, the repeating unit 102D determines whether a predetermined culture time has elapsed (step S18). If the predetermined culture time (e.g., 14 days) for the calculation has elapsed (if YES), the calculation is terminated and the process proceeds to the output step (step S24). If the predetermined culture time has not elapsed (if NO), the process determines whether to replenish the culture medium, nutrients, etc. (step S20). If the culture medium, nutrients, etc. will not be replenished (if NO), the process returns to the biological behavior quantity prediction step (step S14). If the culture medium, nutrients, etc. will be replenished (if YES), the culture medium concentration changing step (step S22) is performed.

[0109] If the culture environment input step (step S12) specifies that the culture medium and nutrients be replenished after a predetermined time has elapsed, a culture medium concentration change step (step S22) is performed. The replenishment of the culture medium and nutrients includes the replenishment of a pH adjuster and carbon dioxide in addition to the culture medium and nutrients. The culture medium concentration change step changes the culture medium concentration by adding the replenished culture medium and nutrients to the cellular environment at the time the cellular environment change step is completed.

[0110] Furthermore, the medium concentration changing step can not only replenish the medium and nutrients, etc., but also remove the medium containing growth-inhibiting by-products and change the medium concentration by harvesting the target product. If the removal of the medium containing growth-inhibiting by-products and the harvesting of the target product are set as culture conditions in the culture environment input step (step S12), the medium concentration is changed by removing the removed growth-inhibiting by-products and the harvested target product from the cellular environment at the time the cellular environment changing step is completed.

[0111] When the repeating step is performed, the uptake constraint conditions are acquired based on the cellular environment changed in the cellular environment changing step (step S16) or the medium concentration changed in the medium concentration changing step (step S22), and the biological behavior amount prediction step (step S14) and the cellular environment changing step (step S16) are performed. Thereafter, the biological behavior amount prediction step (step S14) and the cellular environment changing step (step S16) are repeated until a predetermined culture time has elapsed (step S18).

[0112] After the predetermined culture period has elapsed (YES in step S18), the calculation is terminated. As a result, the number of cells and the amount of bioproduction or ammonia produced during the culture period can be determined. Furthermore, by plotting the amount of biological behavior at each point in time during the repeated process, it is possible to determine the cell growth curve and the progress of antibody production.

[0113] <<Output Step (Step S22)>> The output unit 102E performs the output step (Step S22), which is a step of outputting the cellular environment changed in the cellular environment changing step (Step S16).

[0114] The results output by the output step can include the number of cells grown during the culture period, and the amount of bioproduction produced, or the amount of bioproduction of antibodies such as ammonia. Also, a cell growth curve, progress of antibody production, etc. can be output. The results to be output may be the results after a predetermined time has elapsed, or the results may be output after any period of time has elapsed during the culture.

[0115] [Specific Configuration and Processing Example of Cell Property Estimation Unit] A specific configuration and processing example of the above-mentioned cell property estimation unit 104 (processor) will be described. Fig. 19 is a diagram showing an example configuration of the cell property estimation unit 104. In the example shown in the figure, the cell property estimation unit 104 includes a culture data input unit 104A, a feature extraction unit 104B, a mathematical model creation unit 104C, and an output unit 104D. Each of these units can be configured using various processors and memories (ROM, RAM, etc.), similar to the above-mentioned processing unit 100.

[0116] As will be described below, the cell property estimation unit 104 can create a mathematical cell model, and can estimate cell properties using the created mathematical cell model.

[0117] 20 is a flowchart showing the method for creating a mathematical cell model according to this embodiment. The method according to this embodiment includes a culture data input step (step S13) for inputting cell culture data, a feature extraction step (step S15) for extracting features of cell activity from the culture data input in the culture data input step, a mathematical model creation step (step S17) for creating a mathematical cell model from the features of cell activity extracted in the feature extraction step, and an output step (step S19) for outputting the mathematical model created in the mathematical model creation step.

[0118] Each step will be described below.

[0119] <<Culture Data Input Step (Step S13)>> The culture data input unit 104A performs the culture data input step (step S13). The culture data input step is a step of inputting cell culture data. The culture data can be input in response to an input instruction operation by the user.

[0120] The culture data may include one or more of the total cell count, the amount of cell-secreted substances, the amount of cell-produced substances, the amount of cell-metabolized substances, and the amount of medium components, and / or one or more of the data on the change over time of the total cell count, the change over time of the amount of cell-secreted substances, the change over time of the amount of cell-produced substances, the change over time of the amount of cell-metabolized substances, and the change over time of the amount of medium components. Here, cell-secreted substances refer to substances that are produced by cells and released outside the cells but are not the target substance, examples of which include ammonia and by-products. Cell-produced substances refer to substances that are produced by cells and released outside the cells but are the target substance, examples of which include antibodies. Cell-metabolized substances refer to substances that are produced by cells and are present inside the cells. These culture data can be culture data obtained by actual culture.

[0121] <<Feature Extraction Step (Step S15)>> The feature extraction unit 104B performs the feature extraction step (Step S15), in which features of cell activity are extracted from the culture data input in the culture data input step.

[0122] The features of cellular activity are the concentrations of medium components, the amounts or rates of consumption of medium components, the amounts or rates of production of cell secretory substances, the amounts or rates of production of cell-produced substances, and the amounts or rates of production of cell metabolic substances. In the feature extraction step, the features of the cellular activity are extracted from the culture data input in the culture data input step.

[0123] FIG. 21 is a diagram illustrating the feature extraction process. FIG. 21 shows time-series culture data of medium component concentrations (component X1 is shown as an example) and time-series culture data of antibody production (Y). In the feature extraction process, for example, concentration vectors of medium components X1 to Xn at culture time t are extracted from the time-series culture data of medium component concentrations as shown in FIG. 21, as indicated by arrow A. Furthermore, consumption rates ΔX1,t to ΔXn,t of medium components X1 to Xn are extracted as indicated by arrow B. The consumption rate ΔX1,t of medium component X1 at time t is calculated by dividing the consumption amount ΔX1 of medium component X1 from time t to time (t+1) by the culture time. Consumption rates ΔX2,t to ΔXn,t of other medium components can be calculated in a similar manner. Similarly, the production rate ΔYt of antibody Y is extracted from the time-series culture data of antibody production, as indicated by arrow C. The production rate ΔYt of antibody Y at time t is calculated by dividing the increase ΔY in antibody Y from time t to time (t+1) by the culture time.

[0124] Regarding the concentration of the medium components, it is preferable to calculate the theoretical upper limit of the amount of the medium component that can be absorbed by the cells or the consumption rate of the medium component by the cells, and extract the concentration vector of the medium component. An example of a metabolic pathway of a cell is described above with reference to FIG.

[0125] In the example of the cellular metabolic pathway described above with reference to Figure 15, the conversion of substance A to substance B occurs in equal amounts due to the law of conservation of mass. Similarly, the amount of substance B is equal to the sum of the amounts converted to substances C and E. Thus, the amount of substance converted is determined by the amount of substrate and nutrients initially taken up by the cell. Furthermore, reaction rate is a parameter that limits the conversion of substances. For example, as described above, the conversion of substance A to substance B occurs in equal amounts, but due to reaction rate limitations, there is an upper limit to the conversion to substance B within a certain unit time, and not all of substance A may be converted to substance B. Therefore, when the concentration of a medium component is high, there may be a difference between the actual concentration of the medium component and the concentration of the medium component that can be consumed by culture. Therefore, if a mathematical cell model is created using the actual concentration of the medium component, there is a concern that the resulting mathematical cell model may not be based on cellular characteristics. Therefore, by calculating the theoretical upper limit of the amount of medium that can be absorbed by cells or the medium consumption rate and using this as a characteristic quantity of cellular activity, it becomes possible to create a mathematical cell model that is more based on cellular characteristics.

[0126] To calculate the theoretical upper limit of the amount of material absorbed by cells or the medium consumption rate, for example, the Michaelis-Menten equation can be used. The Michaelis-Menten equation is outlined above with reference to FIG. 16 . The Michaelis-Menten equation relates to the reaction rate V of an enzyme. When the substrate concentration S is low, the reaction rate V is proportional to the substrate concentration S, whereas when the substrate concentration S is high, the reaction rate V converges to a maximum rate Vmax regardless of the substrate concentration S. Another mathematical model that can be used is Fick's law. Fick's law is a formula for calculating the flux, which is the amount passing through a unit area per unit time, and is proportional to the diffusion coefficient D and the substrate concentration gradient on both sides of the membrane. Calculating the medium concentration using the Michaelis-Menten equation or Fick's law makes it possible to create a mathematical cell model that is more based on cell characteristics.

[0127] <<Mathematical Model Creation Step (Step S17)>> The mathematical model creation unit 104C performs the mathematical model creation step (step S17). The mathematical model creation step is a step of creating a cellular mathematical model from the feature amounts of cellular activity extracted in the feature amount extraction step.

[0128] Fig. 22 is a flowchart showing the mathematical model creation process, and Fig. 23 is a diagram for explaining the mathematical model creation process.

[0129] The cell mathematical model creation step separates the features of cellular activity extracted in the feature extraction step into input factors to the target cell and output factors from the target cell (separation step: step S32). In the separation step, the concentration of a medium component is selected as the input factor. The output factors selected include at least one of the amount or rate of consumption of the medium component, the amount or rate of production of a cell-secreted substance, the amount or rate of production of a cell-produced substance, and the amount or rate of production of a cell-metabolized substance.

[0130] 23, a mathematical cellular model is created (searched) by searching for F(C) that satisfies the input factors and output factors at any time (creation step: step S34). Here, the concentrations of medium components X1 to Xn at any time t are used as input factors, and the consumption rates ΔX1,t to ΔXn,t of medium components X1 to Xn at time t and the production rate ΔYt of antibody Y at time t are used as output factors to create a mathematical cellular model. The mathematical cellular model is created by creating a mathematical cellular model that reproduces the characteristic quantities of cellular activity, where the input factors and output factors are satisfied at any time points t1, t2, t3, t4, ...

[0131] Furthermore, the cellular mathematical model is based on a cellular mathematical model that reproduces the characteristics of cellular activity. The mathematical model creation unit 115 can generate multiple cellular mathematical models by modifying this reference cellular mathematical model (generation process: step S36). The cellular mathematical model can be modified by randomly modifying the reference cellular mathematical model, as shown in FIG. 22 . For example, as shown in FIG. 23 , the central cellular mathematical model has modified values ​​in two rows and four columns compared to the reference cellular mathematical model (left). Furthermore, the cellular mathematical model on the right has modified values ​​in three rows and two columns compared to the reference cellular mathematical model. Furthermore, random modification can be created using a genetic algorithm. A genetic algorithm is a method of creating a cellular mathematical model by repeatedly performing operations such as mutation or crossover. Mutation in a genetic algorithm is an operation that changes part of a cellular mathematical model. Crossover is a method of selecting a crossover point between two cellular mathematical models and swapping the parts after that point to create a different cellular mathematical model.

[0132] The mathematical model creation step (step S17) allows creation of a cellular mathematical model that reproduces the features of cellular activity. Furthermore, by creating a cellular mathematical model that is based on this mathematical model and that is partially modified, even if the culture data estimated using the cellular mathematical model that reproduces the features of cellular activity does not reproduce the actual culture data, it is possible to easily search for other cellular mathematical models that can reproduce the culture data.

[0133] <<Output Step (Step S19)>> The output unit 104D performs the output step (step S19). The output step is a step of outputting the mathematical model of the cell created in the mathematical model creation step (step S17).

[0134] In the output step, the plurality of mathematical cell models created in the mathematical model creation step can be output.

[0135] According to the above-described aspects of the cell characteristic estimation unit 104 (method for creating a mathematical cell model, program for creating a mathematical cell model, and device for creating a mathematical cell model), a mathematical cell model can be created using culture data from actual cultures, and therefore a mathematical cell model can be created from culture data without requiring information at the gene level.

[0136] Although the embodiment and other examples of the present invention have been described above, the present invention is not limited to the above-described aspects and various modifications are possible.

[0137] 10 Search and estimation system 100 Processing unit 102 Cultivation result prediction unit 102A Cultivation environment input unit 102B Biological behavior quantity prediction unit 102C Cell environment change unit 102D Repeat unit 102E Output unit 104 Cell characteristic estimation unit 104A Cultivation data input unit 104B Feature extraction unit 104C Mathematical model creation unit 104D Output unit 110 Search and estimation unit 112 Cultivation result acquisition unit 114 Additional estimation evaluation unit 115 Mathematical model creation unit 116 Culture medium composition presentation unit 118 Control unit 120 Input / output control unit 122 Communication control unit 130 ROM 140 RAM 200 Memory unit 300 Display unit 310 Monitor 400 Operation unit 410 Keyboard 420 Mouse 500 External server 510 External database 600 Culture evaluation system S1 to S3 Each step of the workflow S12, S14, S16, S18, S20, S22 Each step of the culture result prediction method S13, S15, S17, S19 Each step of the cell mathematical model creation method S32, S34, S36 Each step of the mathematical model creation process S100 to S150 Each step of the search and estimation method

Claims

1. A method for exploring medium compositions for cells and estimating cell characteristics, comprising: A prediction step of predicting a culture outcome for a given medium composition for a target cell using a cellular characteristic or an estimated value of the target cell; acquiring the culture results of the target cell for some or all of the given medium compositions; an estimation step of estimating a cellular characteristic of the target cell using the medium composition and the acquired culture result or the predicted culture result; An evaluation step of evaluating the need to further estimate the cellular characteristics of the target cells from the obtained culture results; a search and determination step of searching for and determining a candidate medium composition, which is a medium composition for obtaining the culture result in the acquisition step, based on the necessity and the prediction in the prediction step; a control step of repeating the predicting step, the obtaining step, the estimating step, the evaluating step, and the searching and determining step until a determined termination criterion is met; A method for executing the above by a processor.

2. The method according to claim 1 , wherein the processor performs the evaluation in the evaluation step by assuming that there is continuity in changes in culture results in response to changes in medium composition.

3. The method according to claim 1 or 2, wherein the processor performs the estimation in the estimation step by assuming that the cell characteristics of the target cell are similar to the cell characteristics of one or more cells in a known cell group.

4. The processor, The culture result is regarded as an observation history, the prediction means in the prediction step is regarded as an unknown function, and the shape of the unknown function is regarded as the cell characteristic, The method according to claim 1 or 2, wherein the prediction step, the estimation step, the acquisition step, the evaluation step, and the search and determination step are performed by a solution to a bandit problem that estimates an unknown function that best fits an observation history.

5. The method of claim 4 , wherein the processor represents the means of prediction, the unknown function and / or a similarity function to the unknown function in the prediction step by a cell simulation including a metabolic pathway model.

6. The processor, Inputting the culture data of the target cells; determining parameters of the metabolic pathway model from the culture data; performing an estimation of the cell characteristic based on the determined parameters. The method according to claim 5.

7. The culture data includes one or more of the total number of cells of the target cells, the amount of cell secretion substances, the amount of cell production substances, the amount of cell metabolic substances, and the amount of medium components, and / or one or more of the change over time of the total number of cells of the target cells, the change over time of the amount of cell secretion substances, the change over time of the amount of cell production substances, the change over time of the amount of cell metabolic substances, and the change over time of the amount of medium components, The processor, in the estimating step, Separating the parameters into input factors to the target cell and output factors from the target cell; Estimating the cell characteristics so that the input factors and output factors at any time hold; The method according to claim 6.

8. The processor, in the prediction step, predicting the amount of biological behavior of the target cell based on the given medium composition and culture conditions; modifying a cellular environment of the target cell based on the predicted biological behavior amount; repeating the prediction of the biological behavior quantity and the modification of the cellular environment based on the modified cellular environment; The method according to claim 1 or 2, wherein the culture outcome is predicted by

9. The processor, The prediction is performed using the biological behavior quantities of the total number of cells, the substances produced by the target cells and the amounts of the substances, and the substances and the amounts of the substances contained in the culture environment of the target cells as the biological behavior quantities; The method according to claim 8 , wherein the change is made to the total number of the target cells and the substances and amounts of the substances produced by the target cells as the cellular environment.

10. The processor, In the evaluation step, a degree of discrepancy between the estimated value and a true value of the cell characteristic is evaluated based on the predicted culture result and the acquired culture result; The method according to claim 1 or 2, wherein in the searching and determining step, the candidate medium composition is searched for and determined based on the degree of deviation.

11. The processor, In the evaluation step, a pair of the medium composition and the culture result in the medium composition is input to evaluate the sufficiency of the search; The method according to claim 1 or 2, wherein in the searching and determining step, the candidate medium composition is searched for and determined based on the result of the sufficiency evaluation.

12. The method according to claim 1 or 2, wherein the processor, in the control step, determines that the termination criterion is met and terminates the repetitions when the difference between the predicted culture result and the obtained culture result and / or the number of repetitions satisfies a predetermined condition.

13. The method according to claim 1 or 2, wherein the target cells are CHO cells, HEK cells, or iPS cells.

14. The method according to claim 1 or 2, wherein the culture results are one or more of the amount of antibody or virus produced by the target cells, the proliferation of the target cells, and the success rate of inducing differentiation of the target cells into a specific tissue.

15. An apparatus for searching for a medium composition for a cell and estimating a cell characteristic, comprising: a processor; The processor, For a cell of interest, predicting a culture outcome for a given medium composition using a cellular characteristic or estimated value of the cell of interest; Obtaining the culture results of the subject cells for some or all of the given medium compositions; Using the medium composition and the obtained or predicted culture results, a cellular characteristic of the target cell is estimated; Evaluating the need to further estimate the cellular characteristics of the target cells from the obtained culture results; Searching and determining a candidate medium composition that is a medium composition for obtaining the next culture result based on the need and the prediction; The apparatus repeats the predicting, obtaining, estimating, evaluating, and searching and determining until a determined termination criterion is met.

16. A program for causing a processor to search for a medium composition for a cell and estimate a cell characteristic, The processor, A prediction step of predicting a culture outcome for a given medium composition for a target cell using a cellular characteristic or an estimated value of the target cell; acquiring the culture results of the target cell for some or all of the given medium compositions; an estimation step of estimating a cellular characteristic of the target cell using the medium composition and the acquired culture result or the predicted culture result; An evaluation step of evaluating the need to further estimate the cellular characteristics of the target cells from the obtained culture results; a search and determination step of searching for and determining a candidate medium composition, which is a medium composition for obtaining the culture result in the acquisition step, based on the necessity and the prediction in the prediction step; a control step of repeating the predicting step, the obtaining step, the estimating step, the evaluating step, and the searching and determining step until a determined termination criterion is met; A program that executes the following.

17. A non-transitory computer-readable recording medium having the program according to claim 16 recorded thereon.

Images