Cultivation-related process optimization method and cultivation-related process optimization system

Abstract

The present invention provides a cultivation-related process optimization method and a cultivation-related process optimization system capable of optimizing a cultivation-related process. The cultivation-related process optimization method includes: an acquisition step, obtaining a starting execution process that specifies content related to operations performed in the cultivation-related process and serves as a search starting point; a variable parameter item determination step, determining a variable parameter item in the starting execution process that can set a variable parameter value; an execution process generation step, setting a variable parameter value for the variable parameter item based on past execution performance results and evaluation performance results, and generating an execution process; an execution result acquisition step, obtaining an execution result when the execution subject actually executes the execution process in an execution environment; an evaluation result acquisition step, obtaining an evaluation result of the execution result by an evaluation result acquisition unit; and a storage step, recording the execution process, variable parameter value, execution result, and evaluation result in correspondence.

Description

Technical Field

[0001] The present invention relates to a cultivation association process optimization method and a cultivation association process optimization system. Background Art

[0002] Patent Document 1 proposes a system for executing and managing laboratory experiments in life sciences.

[0003] Prior art literature

[0004] Patent Literature

[0005] Patent Document 1: U.S. Patent Application Publication No. 2018 / 0196913 Summary of the Invention

[0006] Problems to be solved by the invention

[0007] In particular, when performing cell production that utilizes cells to produce substances (biopharmaceuticals, useful proteins), or (cell) culture-related processes that utilize cells themselves as medicines such as regenerative medicine, there is a need to maximize the gains obtained from the culture-related processes. Patent document 1 does not disclose a method for optimizing the culture-related processes to achieve the maximum gains.

[0008] Therefore, the present invention has been completed in consideration of the above points, and its object is to provide a cultivation-related process optimization method and a cultivation-related process optimization system that can optimize the cultivation-related process.

[0009] Means for solving problems

[0010] [1] The culture-related process optimization method according to the present invention is a culture-related process optimization method in a culture-related process optimization system for optimizing a culture-related process associated with cell culture, comprising:

[0011] an acquisition step, wherein a start point execution process acquisition unit acquires a start point execution process as a start point of the search, wherein the start point execution process specifies one or more operations performed in the association cultivation process and specifies contents related to the operations;

[0012] a variable parameter item determining step, wherein the variable parameter item determining unit determines one or more variable parameter items capable of setting variable parameter values ​​in the starting point execution process;

[0013] an execution process generating step of generating an execution process by setting the variable parameter values ​​for the variable parameter items determined in the variable parameter item determining step based on past execution performance results and evaluation performance results by an execution process generating unit;

[0014] an execution result obtaining step, wherein the execution result obtaining unit obtains the execution result when the execution subject actually executes the execution process in the execution environment;

[0015] an evaluation result obtaining step, in which an evaluation result obtaining unit obtains an evaluation result of the execution result; and

[0016] The storage step records the execution process, the variable parameter value, the execution result and the evaluation result in a database in a corresponding manner.

[0017] [2] The method for optimizing a culture-related process according to [1] above, wherein, in the culture-related process, the operation includes at least one of an operation related to adjustment of a culture medium and an operation related to cell culture based on the culture medium.

[0018] [3] The method for optimizing a culture-related process according to [1] above, wherein the culture-related process is a culture medium adjustment process for adjusting a culture medium and / or a cell culture process for culturing cells.

[0019] [4] A cultivation-related process optimization method according to any one of [1] to [3] above, wherein, in the variable parameter item determination step, based on the execution performance results when the execution subject in the execution environment executes according to a past execution process that is identical to or related to the starting point execution process, one or more variable parameter items in which the variable parameter values ​​can be set in the starting point execution process are determined.

[0020] [5] The cultivation-related process optimization method according to any one of [1] to [4] above, further comprising: a template execution process generation step in which a template execution process is generated by a template execution process generation unit, wherein the search range is used to represent the range of the variable parameter values ​​that can be set in the variable parameter item; in the execution process generation step, the variable parameter value is selected from the search range specified in the template execution process, the selected variable parameter value is set in the variable parameter item, and the execution process is generated.

[0021] [6] A cultivation-related process optimization method according to any one of [1] to [5] above, wherein, in the execution process generation step, the variable parameter value is determined based on the execution performance result when the execution subject in the execution environment executes according to a past execution process that is the same as or related to the starting point execution process.

[0022] [7] A cultivation-related process optimization method according to any one of [1] to [6] above, wherein, in the execution process generation step, a regression model is generated based on the execution performance results, and the regression model is used to determine the variable parameter value set in the variable parameter item.

[0023] [8] A cultivation-related process optimization method according to any one of [1] to [7] above, wherein, in the execution process generation step, the variable parameter value is set for the purpose of obtaining the desired evaluation result.

[0024] [9] A cultivation-related process optimization method according to any one of [1] to [8] above, wherein the evaluation result is based on at least one of the cost, output, quality, execution time, their deviations, and deviations from their given target values ​​of the execution process.

[0025]

[10] The cultivation-related process optimization method according to any one of [1] to [9] above also includes: an execution plan generation step, in which an execution plan is generated by an execution plan generation unit, and the execution plan indicates how the corresponding execution subjects respectively execute the multiple operations specified in the execution process in a coordinated and chronological manner in the execution environment.

[0026]

[11] According to the cultivation-related process optimization method described in

[10] above, in the execution plan generation step, the execution process is subjected to grammatical parsing to generate a data structure that can at least parse the dependencies between the various operations specified in the execution process, the processing objects of the operations, the results obtained through the operations, and the constraints related to the operations, that is, a syntax tree, and the execution plan is generated based on the syntax tree.

[0027]

[12] The cultivation-related process optimization method according to the above-mentioned

[10] or

[11] further includes an execution environment information acquisition step, in which the execution environment information is acquired by the execution environment information acquisition unit, wherein the execution environment information represents the execution environment information represented by the execution subject that actually performs the operation specified in the execution process in the execution environment, and the execution plan generation step generates the execution plan based on the execution process and the execution environment information.

[0028]

[13] A cultivation-related process optimization method as described in any one of

[10] to

[12] above, wherein, in the execution plan generation step, the execution plan is generated taking into account the constraints set in the constraint condition items of the execution process.

[0029]

[14] The cultivation-related process optimization method according to any one of [1] to

[13] above, further comprising: an execution instruction information generation step, wherein the execution instruction information generation unit generates execution instruction information that instructs the execution subject in the execution environment to execute the operation specified in the execution process.

[0030]

[15] According to the training association process optimization method described in

[14] above, the execution instruction information generation step extracts the constraints set in the constraint condition items of the execution process and includes them in the execution instruction information.

[0031]

[16] A cultivation-related process optimization method according to any one of [1] to

[15] above, wherein, in the execution process generation step, the constraint conditions when the execution subject executes the execution process in the execution environment are set in the constraint condition item of the execution process.

[0032]

[17] According to the cultivation-related process optimization method described in

[16] above, in the execution process generation step, in the constraint condition item, a time constraint related to the operation is set as the constraint condition.

[0033]

[18] According to the cultivation association process optimization method described in

[17] above, in the execution process generation step, as the time constraint, it includes at least any one of the time constraint that stipulates the execution time spent by the execution subject when performing the operation, and the time constraint that sets a time constraint between the operations.

[0034]

[19] According to the training association process optimization method described in

[16] above, in the execution process generation step,

[0035] In the constraint condition item, a parallelism constraint that is defined regarding whether or not the plurality of operations can be performed in parallel is set as the constraint condition.

[0036]

[20] According to the cultivation-related process optimization method described in

[16] above, in the execution process generation step, in the constraint condition item, as the constraint condition, an execution condition constraint is set that stipulates that the operation must be performed within the scope of the specified conditions.

[0037]

[21] A cultivation-related process optimization method according to any one of [1] to

[20] above, wherein, in the variable parameter item determination step, an item selection simulation is performed by calculation processing, and the variable parameter item is selected based on the result of the item selection simulation, wherein, in the item selection simulation, a parameter value for item selection is selected from candidate variable parameter items that can become the variable parameter items that are presumed to be able to obtain a specified evaluation result, the selected parameter value for item selection is used as input, and the evaluation result is used as output.

[0038]

[22] According to the cultivation-related process optimization method described in any one of [1] to

[20] above, in the execution process generation step, a variable parameter value selection simulation is performed through calculation processing, and the range of the variable parameter value is limited based on the result of the variable parameter value selection simulation, wherein, in the variable parameter value selection simulation, a candidate variable parameter value that can become the variable parameter value that is inferred to be able to obtain a specified evaluation result is selected, the selected candidate variable parameter value is used as input, and the evaluation result is used as output.

[0039]

[23] According to the cultivation association process optimization method described in

[22] above, the simulated input and output are selected using the variable parameter value as training data to generate a trained regression model that can differentiate the output by the input, and based on the regression model, the range of the variable parameter value is limited.

[0040]

[24] According to the cultivation-related process optimization method described in

[22] above, the input and output of the simulation of the selected variable parameter value are used as training data to generate a trained regression model that extracts feature quantities corresponding to changes in the candidate variable parameter value, and the range of the variable parameter value is limited based on the regression model using the feature quantities extracted from the trained regression model.

[0041]

[25] Before the execution process generation step, it includes:

[0042] an existing condition acquisition step of acquiring, by an existing condition acquisition unit, existing conditions related to the cultivation-related process; and

[0043] An optimal range search step is performed by an optimal range search unit to search, within a predetermined search range of the variable parameter value, for an optimal range of the variable parameter value that does not satisfy at least one or more of the multiple constituent elements included in the existing conditions and that can generate the execution process outside the range of the existing conditions. In the execution process generation step, the variable parameter value is set from the optimal range of the variable parameter value searched in the optimal range search step to generate the execution process.

[0044]

[26] According to the cultivation association process optimization method described in

[25] above, the optimal range search step includes:

[0045] an optimal range logical expression generating step, wherein an optimal range logical expression generating unit generates an optimal range logical expression that logically expresses an optimal range of the variable parameter value based on an existing condition logical expression that logically expresses the existing condition and a search range logical expression that logically expresses a preset search range of the variable parameter value; and

[0046] In the analyzing step, a logic formula analyzing unit analyzes the optimal range logic formula to determine, for each variable parameter item, an optimal range of the variable parameter value that can generate a different execution flow outside the existing condition range.

[0047]

[27] The cultivation-related process optimization method according to

[25] or

[26] above, wherein the optimal range search step uses any one of a patent gazette, a published patent gazette, and technical information as the existing condition.

[0048]

[28] According to the cultivation association process optimization method described in

[27] above, when the existing condition is the patent gazette or the published patent gazette, in the existing condition acquisition step, the existing claim recorded in the patent gazette or the published patent gazette is obtained as the existing condition, and the optimal range search step includes: a patent information parsing step, in which the subordinate relationship of multiple existing claims is parsed by the patent information parsing unit, and for each of the existing claims, the mutual relationship of multiple constituent elements respectively specified in the existing claim is parsed.

[0049]

[29] A culture-related process optimization system for optimizing a culture-related process associated with cell culture, comprising: a starting point execution process acquisition unit, which acquires a starting point execution process that specifies one or more operations to be performed in the culture-related process and specifies content related to the operations, and serves as a starting point for search; a variable parameter item determination unit, which determines one or more variable parameter items in the starting point execution process for which variable parameter values ​​can be set; an execution process generation unit, which sets the variable parameter values ​​for the variable parameter items determined in the variable parameter item determination unit based on past execution performance results and their evaluation performance results, and generates an execution process; an execution result acquisition unit, which acquires an execution result when an execution subject executes according to the execution process in an execution environment; an evaluation result acquisition unit, which acquires an evaluation result of the execution result; and a database, which records the execution process, the variable parameter values, the execution result, and the evaluation result in correspondence with each other.

[0050]

[30] The cultivation association process optimization system according to

[29] above, wherein it includes:

[0051] an existing condition acquisition unit for acquiring existing conditions related to the cultivation-related process; and

[0052] An optimal range search unit searches, within a predetermined search range for the variable parameter value, for an optimal range of the variable parameter value that does not satisfy at least one or more of the multiple constituent elements included in the existing conditions and is capable of generating the execution process outside the range of the existing conditions. In the execution process generation step, the variable parameter value is set from the optimal range of the variable parameter value searched by the optimal range search unit to generate the execution process.

[0053] Effects of the Invention

[0054] According to the present invention, culture-related processes related to cell culture can be optimized. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is a block diagram showing the overall configuration of the cultivation-related process optimization system according to the present embodiment.

[0056] Figure 2 This is a schematic diagram showing an example of the configuration of the start-point execution flow.

[0057] Figure 3 This is a schematic diagram showing an example of the arrangement of the execution performance results and the evaluation performance results 1 based on the past execution flow.

[0058] Figure 4 This is a schematic diagram showing an example of the arrangement of the execution performance results and the evaluation performance results 2 based on the past execution flow.

[0059] Figure 5 This is a schematic diagram showing an example of the configuration of the template execution flow.

[0060] Figure 6 This is a schematic diagram showing an example of a configuration for executing process 1.

[0061] Figure 7 This is a schematic diagram showing an example of a configuration for executing process 2.

[0062] Figure 8 This is a schematic diagram showing an example of the structure of an execution flow abstract syntax tree.

[0063] Figure 9A Yes Figure 8 A schematic diagram of the subsequent configuration of the execution flow abstract syntax tree is shown.

[0064] Figure 9B Yes Figure 8 A schematic diagram of the subsequent configuration of the execution flow abstract syntax tree is shown.

[0065] Figure 10A This is a schematic diagram showing the configuration of execution environment information.

[0066] Figure 10B It is a diagram showing the configuration of a poset.

[0067] Figure 11 It is a schematic diagram for explaining a set A' to which execution subjects are respectively allocated in an operation.

[0068] Figure 12 This is a schematic diagram showing an example of the structure of an extended abstract syntax tree.

[0069] Figure 13 Yes Figure 12 A schematic diagram of the subsequent configuration of the extended abstract syntax tree is shown.

[0070] Figure 14 Yes Figure 12 A schematic diagram of the subsequent configuration of the extended abstract syntax tree is shown.

[0071] Figure 15 This is a schematic diagram for explaining a set A″ in which execution start time and execution end time are assigned to each element of the set A′.

[0072] Figure 16 Schematic diagram for explaining a set A″′ of combinations satisfying constraint conditions.

[0073] Figure 17A This is a flowchart showing the flow of the cultivation-related process optimization processing according to this embodiment.

[0074] Figure 17B This is a diagram used to explain how to execute multiple execution flows.

[0075] Figure 18 This is a block diagram showing the configuration of a template execution flow generation unit.

[0076] Figure 19 This is a schematic diagram showing an example of the configuration of another start point execution flow.

[0077] Figure 20 This is a schematic diagram showing an example of the configuration of the past related execution process A.

[0078] Figure 21 This is a schematic diagram showing an example of the configuration of the past related execution process B.

[0079] Figure 22This is a schematic diagram showing an example of evaluation performance results of past related execution processes.

[0080] Figure 23 This is a schematic diagram showing an example of the configuration of another template execution flow.

[0081] Figure 24 This is a schematic diagram showing an example of the configuration of another execution process 1.

[0082] Figure 25 This is a schematic diagram showing an example of the configuration of another execution flow 2.

[0083] Figure 26 This is a flowchart showing the template execution process generation process flow.

[0084] Figure 27 This is a block diagram showing the configuration of a variable parameter value setting unit.

[0085] Figure 28 This is a flowchart showing the flow of variable parameter value setting processing.

[0086] Figure 29 This is a block diagram showing the configuration of the execution plan generation unit.

[0087] Figure 30 This is a flowchart showing the flow of the execution plan generation process.

[0088] Figure 31 This is a schematic diagram showing an example of the structure (1) of an individual abstract syntax tree.

[0089] Figure 32 This is a schematic diagram showing an example of the structure (2) of an individual abstract syntax tree.

[0090] Figure 33 This is a schematic diagram showing an example of the structure (3) of an individual abstract syntax tree.

[0091] Figure 34 This is a schematic diagram showing an example of the structure (4) of an individual abstract syntax tree.

[0092] Figure 35 This is a flowchart showing the flow of processing for generating individual abstract syntax trees.

[0093] Figure 36 This is a schematic diagram for explaining the outline (1) of the individual abstract syntax tree generation process flow.

[0094] Figure 37 This is a schematic diagram for explaining the outline (2) of the individual abstract syntax tree generation process flow.

[0095] Figure 38This is a schematic diagram for explaining the outline (3) of the individual abstract syntax tree generation process flow.

[0096] Figure 39 This is a schematic diagram for explaining the outline (4) of the individual abstract syntax tree generation process flow.

[0097] Figure 40 This is a flowchart showing the execution flow of the abstract syntax tree generation process.

[0098] Figure 41 This is a schematic diagram showing the structure (1) of the individual abstract syntax trees synthesized by executing the flow abstract syntax tree generation process.

[0099] Figure 42 This is a schematic diagram showing the structure (2) of the individual abstract syntax trees synthesized by executing the flow abstract syntax tree generation process.

[0100] Figure 43 It means comprehensive Figure 41 The individual abstract syntax trees shown and Figure 42 Schematic diagram of the structure of the intermediate abstract syntax tree of the individual abstract syntax trees shown.

[0101] Figure 44 This is a flowchart showing the flow of the extended abstract syntax tree generation process.

[0102] Figure 45 This is a schematic diagram for explaining the outline of partial order.

[0103] Figure 46 It is a block diagram showing the configuration of an execution instruction information generating unit.

[0104] Figure 47 This is a flowchart showing the flow of execution instruction information generation processing.

[0105] Figure 48 It is a block diagram showing the configuration of a template execution flow generation unit according to the second embodiment.

[0106] Figure 49 This is a flowchart showing the flow of template execution flow generation processing according to the second embodiment.

[0107] Figure 50 is a block diagram showing the configuration of a variable parameter value setting section according to the second embodiment.

[0108] Figure 51 This is a flowchart showing the flow of variable parameter value setting processing according to the second embodiment.

[0109] Figure 52This is a schematic diagram for explaining how to limit the range of a variable parameter value in the second embodiment.

[0110] Figure 53 It is a block diagram showing the overall configuration of a cultivation-related process optimization system according to a fourth embodiment.

[0111] Figure 54 It is a block diagram showing the configuration of the optimal range search unit.

[0112] Figure 55 This is a flowchart showing the flow of the optimal range search process.

[0113] Figure 56 This is a schematic diagram showing an example of a claim syntax tree generated based on conventional claims 1 to 4.

[0114] Figure 57 This is a schematic diagram showing an example of a constituent element syntax tree generated based on the content of conventional claim 1.

[0115] Figure 58 This is a schematic diagram showing a logical expression obtained by associating the dependency relationships of the existing conditional logical expressions of the existing claims 1 to 4. DETAILED DESCRIPTION

[0116] Hereinafter, one embodiment of the present invention will be described in detail with reference to the accompanying drawings. In the following description, the same components are given the same reference numerals and repeated descriptions are omitted.

[0117] (1) Cultivation-related process optimization method according to the first embodiment

[0118] (1-1) Overview of the Cultivation-Related Process Optimization Method According to the First Embodiment

[0119] First, an overview of the culture-associated process optimization method according to the present embodiment is described. The culture-associated process mentioned here is various processes associated with the culture of cells, which perform one or more operations necessary in the process of culturing various cells. In the following embodiments, a series of processes from the process of adjusting the culture medium for cell culture (hereinafter also referred to as the culture medium adjustment process) to the process of using the culture medium for cell culture (hereinafter also referred to as the cell culture process) can be used as the culture-associated process. In addition, only the culture medium adjustment process can be used as the culture-associated process, or only the cell culture process can be used as the culture-associated process. The culture-associated process includes at least a portion of the following operations: selection of basal culture medium, adjustment of culture medium, selection of components used in the adjustment of culture medium, culture of cells performed by the adjusted culture medium, selection of execution entities such as culture conditions and devices for performing these operations, and determination of the processing flow of the execution entity.

[0120] Figure 1 1 is a block diagram showing the overall configuration of a cultivation-related process optimization system 1 according to the present embodiment that executes the cultivation-related process optimization method. Figure 1 As shown, the cultivation-related process optimization system 1 has a structure in which a cultivation-related process optimization device 2 and a plurality of communication devices 3a, 3b, 3c, and 3d are connected to a network 4 such as the Internet.

[0121] In the culture-related process optimization system 1, for example, in various culture-related processes, such as culture-related processes related to regenerative medicine, in which human iPS cells are cultured and induced to differentiate into specific cells, culture-related processes related to the production of biopharmaceuticals (antibody production), and culture-related processes related to the production of useful proteins, in which Escherichia coli X is cultured and the expression of enzyme R is induced to produce compounds, an execution flow that maximizes the gain by trial and error is determined.

[0122] Here, the term "optimal evaluation result" refers to the application of at least one of the following: cost, yield, quality, execution time, their deviations, and deviations from their predetermined target values, related to the execution flow of a culture-related process. For example, an example of an evaluation result for a culture-related process involving culturing cells in multiple culture media P and Q and inducing differentiation from human iPS cells, etc., into specific cells would include the cost, cell yield, cell quality, execution time until cell production, their deviations, and their predetermined target values.

[0123] Here, the so-called execution process refers to data that records the following: a series of operations performed in the culture-related process and evaluation process (for example, adjustment and cultivation of the culture medium, evaluation of marker gene expression, etc.), the processing objects processed by each operation (for example, human iPS cells, etc.), the results obtained after the processing objects are processed by each operation (for example, cells induced by differentiation from human iPS cells, etc.), various execution parameter values ​​when the processing objects are processed by the operation (for example, basal culture medium concentration, BMP4 concentration, VEGF concentration, glucose concentration, culture period, etc.) and constraints related to the operation (for example, "after more than 5 minutes have passed after the device is started", etc.).

[0124] For example, as execution processes, there can be cited: an experimental process that represents a series of operations when multiple experimenters perform life science experiments in a laboratory, a culture medium adjustment and culture process that represents a series of operations from adjusting the culture medium to culturing cells with the culture medium using multiple devices, etc.

[0125] Here, the production conditions within the execution flow of the culture-related process that maximize the gain obtained from a certain culture-related process (for example, a combination of various execution parameter values ​​such as basal culture medium concentration and glucose concentration) are not obvious, and in simple methods, it is necessary to repeatedly perform multiple experiments and conduct trial and error. The more complex the product, the more complex and large its culture-related process becomes, and the more difficult it becomes to search for various production conditions to determine the optimal production conditions. In addition, in the production of a small number of multiple varieties of culture-related processes, pharmaceuticals, etc., robots and the like are also flexibly used in the culture-related processes using organisms to achieve automation. However, due to the differences in the varieties of the production objects and the individual differences of the organisms, it is necessary to find the optimal production conditions each time in each case. As a result, the total cost of searching for production conditions increases.

[0126] Generally speaking, when searching for production conditions, if the types and number of execution entities such as mechanical devices, robots, and people (operators) that perform various operations in the process of cultivating associations are increased, it is necessary to generate appropriate execution instruction information for each search condition of the production conditions in turn for each execution entity, which requires a lot of labor.

[0127] In the cultivation-related process optimization system 1 according to this embodiment, taking the above points into consideration, when the execution subject actually executes the cultivation-related process according to the execution process in the execution environment 100 and searches for the optimal production conditions that can obtain the greatest possible gain, it can obtain the optimal production conditions with a greater gain with the least possible number of experiments, thereby reducing the total cost and labor required for searching for production conditions, and at the same time being able to obtain the optimal cultivation-related process with a greater gain.

[0128] Here, as an example of a culture-related process optimized by the culture-related process optimization system 1 according to this embodiment, a culture-related process is described in which, for example, differentiation-inducing medium A (hereinafter, also referred to as "medium A") and differentiation-inducing medium B (hereinafter, also referred to as "medium B"), which is different from medium A, are used to induce differentiation of human iPS cells and produce specific cells. The following is an overview of the culture-related process optimization system 1 according to this embodiment. Furthermore, as an example of an evaluation process for evaluating cells induced to differentiate from human iPS cells obtained using the culture-related process, the case of evaluating the expression positivity rate of marker genes is described.

[0129] like Figure 1 As shown, the cultivation-related process optimization device 2 included in the cultivation-related process optimization system 1 includes a processing unit 7, a database 8, a display unit 9, an operation unit 10, and a transmitting / receiving unit 11. The transmitting / receiving unit 11 serves as an execution result acquisition unit and an evaluation result acquisition unit.

[0130] The arithmetic processing unit 7 has a microcomputer structure composed of a CPU (Central Processing Unit), RAM (Random Access Memory) and ROM (Read Only Memory) (not shown), and is connected to a database 8, a display unit 9, an operating unit 10 and a sending / receiving unit 11.

[0131] When various operation commands are received from the administrator (also called the user) of the cultivation-related process optimization device 2 via the operation unit 10, the operation processing unit 7 appropriately reads the cultivation-related process optimization program, template execution flow generation processing program, variable parameter value setting processing program, execution plan generation processing program, execution instruction information generation processing program, etc. pre-stored in the ROM based on the operation commands, and deploys them to the RAM, thereby controlling each loop unit according to the cultivation-related process optimization program, etc.

[0132] The calculation processing unit 7 generates a template execution flow, an execution plan, execution instruction information, etc. by, for example, executing the training-related process optimization process, and stores these calculation processing results in the database 8.

[0133] In addition to storing the calculation results of the calculation processing unit 7, the database 8 also stores execution environment information (described later) received from the outside by the sending / receiving unit 11. Furthermore, the database 8 stores various data, such as the execution results of past executions of the execution flow of the training-related process by the execution subject in the execution environment 100 (hereinafter, the past execution results are also referred to as the execution performance results), and the past evaluation results of the execution performance results (hereinafter, the evaluation results of the past execution results are also referred to as the evaluation performance results).

[0134] The display unit 9 displays the calculation processing results such as the template execution flow and the execution flow generated by the calculation processing unit 7, so that the administrator who manages the training-related process optimization device 2 can grasp these processing results.

[0135] According to the present embodiment, the culture-related process optimization device 2 generates, through culture-related process optimization processing, the following: an execution flow of the culture-related process and the evaluation process for producing specific cells by using culture media A and B to differentiate and induce human iPS cells; an execution plan, which is data indicating at what timing each execution subject should coordinately perform each operation within the execution flow; and execution instruction information, which is data for instructing the execution subject of the execution environment 100 to perform corresponding operations respectively according to the execution plan.

[0136] The training-related process optimization device 2 sends the execution plan and execution instruction information to the communication devices 3 a , 3 b , 3 c , and 3 d of the corresponding execution entities in the execution environment 100 via the network 4 .

[0137] here, Figure 1 Execution environment 100 depicts an environment, such as a factory or laboratory, where the execution flow of the cultivation-related and evaluation processes is actually executed. Execution entities within execution environment 100 perform the operations specified within the execution flow of the cultivation-related and evaluation processes. These entities vary depending on the type of cultivation-related and evaluation processes and may include, for example, mechanical devices, robotic arms, humans, incubators, measuring equipment such as cameras, laboratory automation equipment, and computers.

[0138] In this embodiment, the culture-related process of producing specific cells by inducing differentiation of human iPS cells (hereinafter also referred to as "cells") using culture media A and B, and the evaluation process of evaluating the expression of marker genes in the produced cells by executing the execution flow of the culture-related process are explained as an example. However, in this case, the culture medium mixing device P for adjusting and mixing culture media A and B, the cell culture devices X and Y for culturing cells, and the flow cytometer Q for evaluating the expression of marker genes can become the execution entities.

[0139] In the culture-related process optimization system 1 according to this embodiment, in order to prompt the execution plan and execution instruction information generated by the culture-related process optimization device 2 to the culture medium mixing device P, cell culture devices X, Y and flow cytometer Q as the execution subject, the communication devices 3a, 3b, 3c, and 3d of the execution subject are sent via the network 4.

[0140] For example, the communication device 3a can also be configured to be connected to or mounted on the culture medium mixing device P, prompting the culture medium mixing device P with the setting information contained in the execution instruction information generated for the use of the culture medium mixing device P, and automatically operating the culture medium mixing device P according to the execution plan based on the setting information.

[0141] Alternatively, the communication device 3a may be, for example, an electronic device such as a personal computer or a smartphone owned by an operator who performs operations for adjusting culture media A and B, and mixing culture media A and B as needed. In this case, the communication device 3a transmits the execution plan received from the culture-related process optimization device 2 and the execution instruction information generated for the use of the culture media mixing device P to the operator operating the culture media mixing device P, thereby causing the operator to perform adjustment operations using the culture media mixing device P.

[0142] For example, the communication device 3b may be connected to or mounted on cell culture apparatus X, and the setting information included in the execution instruction information generated for the use of cell culture apparatus X may be presented to cell culture apparatus X, so that the cell culture apparatus X automatically operates according to the execution plan based on the setting information. Alternatively, the communication device 3c may be connected to or mounted on cell culture apparatus Y, and the setting information included in the execution instruction information generated for the use of cell culture apparatus Y may be presented to cell culture apparatus Y, so that the cell culture apparatus Y automatically operates according to the execution plan based on the setting information.

[0143] Alternatively, communication devices 3b and 3c may be electronic devices such as personal computers and smartphones owned by operators performing cultivation operations on culture medium A and culture medium B, respectively. In this case, communication devices 3b and 3c present the execution plan received from cultivation-related process optimization device 2 and the execution instruction information generated for the use of cell culture apparatuses X and Y to the operators operating cell culture apparatuses X and Y, thereby enabling each operator to perform cultivation operations using cell culture apparatuses X and Y.

[0144] Flow cytometer Q is a cell analysis device that can evaluate the expression of marker genes in cells cultured using culture media A and B. It can also be configured so that the communication device 3d is connected to or mounted on the flow cytometer Q, and the setting information contained in the execution instruction information generated for the use of the flow cytometer Q is prompted to the flow cytometer Q, and the flow cytometer Q is automatically operated according to the execution plan based on the setting information. In addition, the communication device 3d can also be, for example, an electronic device such as a personal computer or a smart phone owned by an operator who performs the marker gene expression evaluation operation. In this case, the communication device 3d prompts the operator who operates the flow cytometer Q with the execution plan received from the culture-related process optimization device 2 and the execution instruction information generated for the use of the flow cytometer Q, so that the operator uses the flow cytometer Q to perform the marker gene expression evaluation operation, etc.

[0145] Furthermore, these communication devices 3a, 3b, 3c, and 3d appropriately send the execution results and evaluation results obtained when the corresponding execution subjects execute according to the execution plan and execution instruction information to the training-related process optimization device 2 via the network 4.

[0146] Next, the calculation processing unit 7 of the training-related process optimization device 2 will be described. The calculation processing unit 7 includes a template execution process generation unit 15, a variable parameter value setting unit 16, an execution process generation unit 17, a determination unit 18, an execution plan generation unit 19, and an execution instruction information generation unit 20. The calculation processing unit 7 generates the template execution process, execution process, execution plan, and execution instruction information described later.

[0147] Here, the template execution flow is created by rewriting the execution flow that is the starting point of a certain search (hereinafter referred to as the starting point execution flow). Figure 2 This is a schematic diagram showing an example of the configuration of the start execution flow for the culture-related process for cells cultured in culture media A and B and the evaluation process.

[0148] If the culture-related processes (for example, adjustment of culture media A and B, cell culture based on culture media A, culture based on culture media B for cells in the process of differentiation cultured in culture media A) and evaluation processes (marker gene expression evaluation) that the manager wants to optimize are selected or input through the operation unit 10, the template execution process generation unit 15 selects the execution process corresponding to these culture-related processes and evaluation processes from the multiple execution processes stored in the database 8 as the starting execution process.

[0149] In this case, the template execution process generation unit 15, such as Figure 2 As shown, the execution process that specifies the following operations is selected as the starting point execution process, wherein, as operations of the culture-related process, four operations are specified, namely, adjustment of culture medium A, cell culture based on culture medium A (already adjusted), adjustment of culture medium B, and culture based on culture medium B (already adjusted) for cells in the middle of differentiation cultured in culture medium A, and one operation, namely, marker gene expression evaluation, is specified as an operation of the evaluation process.

[0150] For each operation in the starting execution process, the operation summary columns C1, C2, C3, and C4 specify the details of the operation performed during the training-related process and the evaluation process. For example, the operation summary columns C1, C2, C3, and C4 of the starting execution process according to this embodiment include: an operation item 26a specifying the details of the operation; an input item 26b specifying the processing target of the operation; an output item 26c specifying the output obtained by the operation; an execution parameter item 26d specifying the numerical value associated with the operation; and a constraint item 26e specifying the constraints related to the operation.

[0151] For example, in the operation summary column C1, "Adjustment of differentiation induction medium A" is specified in the operation item 26a, "basal culture medium, BMP4, VEGF" as the processing objects are specified in the input item 26b, "differentiation induction medium A" as the result is specified in the output item 26c, "basal culture medium concentration", "BMP4 concentration", "VEGF concentration" and "glucose concentration" are specified in the execution parameter item 26d, and "None" is specified in the constraint condition item 26e.

[0152] In addition, in the "Basal culture medium concentration" specified in the execution parameter item 26d, "10 mM" is specified as the parameter value indicating the basal culture medium concentration, in the "BMP4 concentration", "2 μM" is specified as the parameter value indicating the concentration of BMP4 (Bone Morphogenetic Protein 4), in the "VEGF concentration", "5 μM" is specified as the parameter value indicating the concentration of VEGF (Vascular Endothelial Growth Factor), and in the "Glucose concentration", "4 μM" is specified as the parameter value indicating the glucose concentration.

[0153] "None" specified in constraint item 26e indicates that no constraints are specified for the operation "Adjustment of Differentiation Induction Medium A." Constraints specify conditions that constrain operations, such as time constraints, parallelism constraints, and execution condition constraints. These constraints specify that operations are subject to time, environment, temperature, humidity, cleanliness, and other restrictions.

[0154] Examples of time constraints include time constraints that specify the time required for an execution subject to perform operations in an execution flow, and time constraints that impose time constraints between operations in an execution flow. Examples of execution condition constraints include those that specify that operations in an execution flow must be performed within a specified range of conditions. For example, a constraint such as "within 60 minutes of completion of preparation of differentiation induction medium B" is shown in constraint condition item 26e in the operation summary column C4.

[0155] In the starting execution flow of the culture-related process and evaluation process according to this embodiment, in addition to the aforementioned operation summary column C1 related to "Adjustment of Differentiation Induction Medium A," there are also: Operation Summary Column C2 for "Cell Culture Using Differentiation Induction Medium A (Target Cell Induction)"; Operation Summary Column C3 for "Adjustment of Differentiation Induction Medium B"; Operation Summary Column C4 for "Cell Culture Using Differentiation Induction Medium B (Target Cell Differentiation)"; and Operation Summary Column C5 for "Marker Gene Expression Evaluation." Furthermore, the "SCF Concentration" specified in the execution parameter item 26d of Operation Summary Column C3 specifies a parameter value representing the concentration of SCF (Stem Cell Factor). Furthermore, in the operation summary column C5 for "Marker Gene Expression Evaluation," an indicator ranging from 0 to 100% is set in the output item 26c, representing the probability of the marker gene's expression-positive rate.

[0156] In the operation summary column C2 regarding "Cell culture based on differentiation induction medium A (target cell induction)", "Differentiation induction medium A (adjusted), human iPS cells (HPS0003 strain)" is specified in input item 26b. According to the provisions of "Differentiation induction medium A (adjusted)", it can be seen that it is specified as an operation to be performed after the operation specified in the operation summary column C1 is completed. In addition, in the operation summary column C4 regarding "Cell culture based on differentiation induction medium B (target cell differentiation)", "Differentiation induction medium B (adjusted), human iPS cells (HPS0003 strain) (in the process of differentiation cultured in differentiation induction medium A)" is specified in input item 26b. According to the specification of "Differentiation induction medium B (adjusted)", it can be seen that it is specified as an operation to be performed after the operation specified in the operation summary column C3 is completed. In addition, according to the specification of "human iPS cells (HPS0003 strain) (in the process of differentiation cultured in differentiation induction medium A)", it can be seen that it is specified as an operation to be performed after the operation specified in the operation summary column C2 is completed.

[0157] The template execution process generation unit 15 generates a template execution process, which is used to select the execution parameter item 26d whose execution parameter value can be changed in the starting execution process as a variable parameter item based on the execution performance results and evaluation performance results of the past execution processes stored in the database 8, and then determine what execution parameter value should be used in the execution environment 100 to execute the training association process and the evaluation process and evaluate its execution results.

[0158] Figure 3 and Figure 4 Indicates based on Figure 2 The contents of the operation summary columns C1, C2, C3, and C4 of the starting execution flow shown in the figure are examples of the arrangement of execution performance results and evaluation performance results of past execution flows, retrieved from the database 8. The template execution flow generation unit 15 compares the multiple execution performance results and evaluation performance results retrieved from the database 8 and identifies, among the execution parameter items 26d in the operation summary columns C1, C2, C3, and C4, those that can be variable parameters with changeable execution parameter values. The determination of the variable parameter items 26d can be based on, for example, the trend of changes in the numerical execution parameter values ​​in the execution performance results and evaluation performance results.

[0159] The template execution flow generation unit 15 compares, for example, Figure 3 The execution performance results and evaluation performance results shown in 1 are the same as Figure 4The execution performance results and evaluation performance results 2 shown in the figure are based on the fact that the execution parameter values ​​of the execution parameter item 26d "BMP4 concentration", "VEGF concentration" and "glucose concentration" in the operation summary column C1 of "Adjustment of differentiation induction medium A" are different, and the difference between the two results is caused by the difference. Figure 3 and Figure 4 For example, if the evaluation performance results shown in the output item 26c of the operation summary column C5 are different, it can be estimated that these execution parameter values ​​can be items (variable parameter items) with variable parameter values.

[0160] Alternatively, the execution parameter values ​​of the starting execution flow can be used to identify the search range set in the previous execution flow for which the execution performance results were obtained, and the intersection of the search range and the union of the previous search range can be set as the new search range. Furthermore, even if the same execution parameter item is changed in multiple execution performance results, if the fluctuation range of each evaluation performance result is below a specified value, then the impact of the execution parameter item on the evaluation performance result is minimal, and a rule can be set to exclude the execution parameter item from being a variable parameter.

[0161] In this embodiment, for simplicity of description, the case where variable parameter items are determined from among the execution parameter items with set numerical values ​​is described. However, operation items 26a, for example, which specify other than numerical values ​​for medium adjustment and cell cultivation, can also be selected as variable parameter items. In other words, in the starting execution process, all operations that can be changed in the execution environment 100 and the conditions related to these operations can be selected as variable parameter items. Furthermore, examples of selecting operation items 26a, for example, which specify other than numerical values, as variable parameter items will be described in detail in other embodiments described later.

[0162] The template execution flow generation unit 15 compares Figure 3 The execution performance results and evaluation performance results shown in 1 are the same as Figure 4 The execution performance results and evaluation performance results 2 shown in the figure set the numerical range of the variable parameter value for the variable parameter item, that is, the search range. In this case, the template execution process generation unit 15 can estimate the range of the variable parameter value (search range) that can obtain the desired marker gene expression positive rate based on the different "marker gene expression positive rate" of the output item 26c in the operation summary column C5 of the "marker gene expression evaluation" of the evaluation process and the execution parameter value of the execution parameter item 26d as the variable parameter item, such as Figure 5 As shown, a template execution flow in which the variable parameter item 26g and its search range are set can be generated.

[0163] For example, as an example, in Figure 3In the execution performance results and evaluation performance results 1 shown, the "BMP4 concentration" of the execution parameter item 26d in the operation summary column C1 of "Adjustment of differentiation induction medium A" is 8 μM, which is higher than Figure 4 The "BMP4 concentration" and "VEGF concentration" were 12 μM, which were also higher than Figure 4 The VEGF concentration was 4 μM, and the glucose concentration was 4 μM, which was lower than Figure 4 The glucose concentration is Figure 3 The expression positive rates of the marker genes shown were compared Figure 4 The main reason for the higher marker gene expression positive rate of "75%" is shown. Next, it is inferred that the execution parameter item 26d in the "Adjustment of Differentiation Induction Medium A" operation summary column C1 affects the marker gene expression positive rate, and this is used as the search range for the variable parameter value. The search range for the variable parameter value is inferred using this feature extraction rule.

[0164] In this way, the template execution flow generation unit 15 estimates one or more execution parameter items and their search ranges that may affect the evaluation performance results based on one or more past execution performance results and evaluation performance results and on the basis of the assigned feature extraction rules.

[0165] In addition, the description herein describes a case where a past execution flow consisting of the same operations as the starting execution flow exists, and the variable parameter item 26g and its search range are set in the starting execution flow based on the execution performance results and evaluation performance results of the past execution flow. Even when the starting execution flow consists of operations different from those of the past execution flow, the template execution flow generation unit 15 has the function of selecting a past execution flow associated with the starting execution flow (hereinafter referred to as an associated execution flow) and setting the variable parameter item 26g and its search range in the starting execution flow based on the execution performance results and evaluation performance results of the associated execution flow.

[0166] For example, if the operations of the starting execution flow include unknown operations not included in the previous execution flow, the types and values ​​of the execution parameter values ​​of the two processes will generally be different. However, the template execution flow generation unit 15, for example, compares and analyzes the execution performance results and evaluation performance results of multiple related execution flows to infer the execution parameter items and their search ranges that will affect the evaluation results in the operations during the evaluation process of the starting execution flow.

[0167] The template execution flow generation unit 15 achieves a correspondence between the execution parameter values ​​of the starting execution flow and the execution performance results and evaluation performance results of the plurality of associated execution flows by designing a feature conversion function that combines at least one matrix operation and at least one linear or nonlinear transformation. Furthermore, the feature conversion function itself can be modified through sequential optimization.

[0168] Here, the feature conversion function refers to a function used to extract a low-dimensional space that has an effect on the target variable based on past execution and evaluation results, even when the search space used to estimate the search range is high-dimensional, if the dimensions actually contributing to the target variable are low-dimensional. By searching for variable parameter values ​​in the low-dimensional search space generated by this feature conversion function, search efficiency can be improved.

[0169] Here, for example, a machine learning model based on a neural network or the like, which is a type of regression model, can be defined in a form that combines at least one matrix operation and at least one linear or nonlinear transformation.

[0170] The following example illustrates the design of a feature conversion function that combines at least one matrix operation and at least one linear or nonlinear transformation. For example, assume that there is a certain relationship between cell culture temperature and culture time. When conditions (culture temperature and culture time) are assigned while satisfying this relationship, it is assumed that a higher evaluation value (a desired evaluation result) is achieved. In this case, by using the feature conversion function to learn the relationship between culture time and culture temperature that results in a higher evaluation value, the feature conversion function is sequentially optimized within a specified space, enabling efficient search.

[0171] Furthermore, the template execution process generation unit 15 regularizes prior knowledge, obtained by a manager or other person, regarding which factors, among the execution performance results and evaluation performance results of the associated execution processes, contribute most to improving the evaluation results. This allows for efficient setting of variable parameter items and their search ranges in the starting execution process based on the execution performance results and evaluation performance results of the associated execution processes. A detailed explanation of how variable parameter items and their search ranges are set based on the execution performance results and evaluation performance results of the associated execution processes will be provided later.

[0172] Figure 5 This shows an example of a configuration of a template execution flow in which a variable parameter item 26g is set in the start point execution flow and a search range is set in the variable parameter item 26g, respectively, with reference to the execution performance results and the evaluation performance results.

[0173] In this example, with reference to the execution performance results and the evaluation performance results, "BMP4 concentration", "VEGF concentration" and "glucose concentration" in the operation summary column C1 of "adjustment of differentiation induction medium A" are set as variable parameter items 26g.

[0174] The template execution flow generation unit 15 sends the generated template execution flow to the variable parameter value setting unit 16 and the execution flow generation unit 17. The variable parameter value setting unit 16 determines, based on past execution performance results and evaluation performance results, which variable parameter values ​​within the search range should be used to execute the "execution flow" in the execution environment 100, and sends the multiple variable parameter values ​​selected from the search range to the execution flow generation unit 17.

[0175] Here, the variable parameter value setting unit 16 generates a plurality of variable parameter values ​​within the search range of the template execution flow based on one or more past execution performance results and evaluation performance results in accordance with a predetermined procedure (e.g., Bayesian optimization, orthogonal array, Latin hypercube method, etc.). Alternatively, the variable parameter values ​​may be set by projecting the execution performance results onto a search space representing the search range of the variable parameter item 26g.

[0176] Furthermore, in this embodiment, even if the execution performance results and evaluation performance results do not exist in the database 8 in the past, variable parameter values ​​can be set from the search range according to rules specified by the manager, such as the orthogonal method.

[0177] In the variable parameter value setting unit 16 according to this embodiment, for example, a regression model (response surface) is generated based on one or more past execution performance results and evaluation performance results, and the regression model is used to select multiple variable parameter values ​​from the search range, for example, through Bayesian optimization or multi-task Bayesian optimization.

[0178] The execution flow generator 17 writes the variable parameter values ​​selected by the variable parameter value setting unit 16 into the variable parameter items 26g of the template execution flow, thereby generating a plurality of execution flows with different variable parameter values.

[0179] Figure 6 and Figure 7 An example of two execution flows in which the variable parameter values ​​set in the variable parameter item 26g are different is shown. Figure 6 As an example of an execution process, the execution process is shown in which "5 μM" is set as the variable parameter value of "BMP4 concentration", "10 μM" is set as the variable parameter value of "VEGF concentration", and "11 μM" is set as the variable parameter value of "glucose concentration" in the variable parameter item 26g in the operation summary column C1 of "Adjustment of differentiation induction medium A".

[0180] in addition, Figure 7 As an example of another execution process, the execution process is shown in which, in the variable parameter item 26g in the operation summary column C1 of "Adjustment of differentiation induction medium A", "6 μM" is set as the variable parameter value of "BMP4 concentration", "8 μM" is set as the variable parameter value of "VEGF concentration", and further, "6 μM" is set as the variable parameter value of "glucose concentration".

[0181] The execution flow generation unit 17 generates a list of such a plurality of execution flows, and transmits the list of these execution flows to the execution plan generation unit 19 .

[0182] The execution plan generation unit 19 selects execution processes from the list of execution processes in sequence and generates an execution plan for each execution process. Figure 6 The following describes the overview of generating an execution plan based on the execution flow shown.

[0183] In addition, here, in order to simplify the explanation, the situation of selecting execution processes in sequence from the list of execution processes and generating an execution plan for the execution processes is explained, but it is also possible to generate separate execution plans for multiple execution processes in the list at one time or an execution plan that represents the relationship between the progress status of multiple execution processes.

[0184] When an execution subject is prompted with multiple execution plans at once, for example, based on the contents of the execution plans, it can simultaneously execute the execution processes according to the multiple execution plans, or it can sequentially select any execution plan from the multiple execution plans and execute the execution processes according to each execution plan. Preferably, each time the execution subject obtains the execution results and evaluation results of the execution processes, it sends them to the training-related process optimization device 2.

[0185] While the training-related process optimization device 2 presents an execution plan to the execution subject in the execution environment 100 and causes the execution subject to execute the "execution process" according to the execution plan, it is preferable that, for example, upon receiving execution results and evaluation results from the execution subject, the execution process is regenerated to reflect the contents of these execution results and evaluation results, and an execution plan is generated for each regenerated execution process. This allows the generation of an optimal execution process that reflects the actual execution results and evaluation results obtained by the execution subject in the execution environment 100, thereby optimizing the training-related process.

[0186] In addition, as described above, when multiple execution plans are presented to the execution subject, the training-related process optimization device 2 receives the execution results and evaluation results of multiple execution processes from each execution subject, and the processing of the training-related process optimization device 2 is described later. Figure 17BNarrate.

[0187] The execution plan generation unit 19 is based on Figure 6 The execution flow shown in Figure 8 、 Figure 9A as well as Figure 9B The execution flow abstract syntax tree t is shown. The execution flow abstract syntax tree is a data structure that performs grammatical analysis of the execution flow and can analyze the dependencies between the various operations specified in the execution flow, the processing objects of the operations, the results obtained by the operations, and the constraints related to the operations, that is, a syntax tree.

[0188] As a syntax tree, the execution process abstract syntax tree t has a tree-structured data structure, in which the contents of the operation items 26a, input items 26b, output items 26c, execution parameter items 26d, variable parameter items 26g and constraint items 26e specified in the operation summary columns C1, C2, C3, and C4 in the execution process are used as nodes, and these nodes are connected by edges to specify the dependencies between operations.

[0189] In this case, the execution plan generation unit 19 generates individual abstract syntax trees for the operation summary columns C1, C2, C3 and C4 of each execution process, respectively. In the individual abstract syntax trees, the contents of the operation item 26a, input item 26b, output item 26c, execution parameter item 26d, variable parameter item 26g and constraint item 26e are used as nodes, and these nodes are connected by edges and dependencies are specified; according to the order in which the operations of the execution process are performed, these individual abstract syntax trees are associated with each other to generate an execution process abstract syntax tree t.

[0190] Note that details of the individual abstract syntax tree generation process and the execution-flow abstract syntax tree generation process for generating the individual abstract syntax tree and the execution-flow abstract syntax tree t from the execution flow will be described later.

[0191] Next, the execution plan generation unit 19 generates an extended abstract syntax tree (described later) that establishes an association between the execution subject that executes each operation in the "execution process" and the execution process abstract syntax tree t based on the execution environment information (described later) received from the execution environment 100 via the sending / receiving unit 11 and the above-mentioned execution process abstract syntax tree t.

[0192] As an extension of the syntax tree, the abstract syntax tree performs grammatical analysis on the execution process. It can analyze the dependencies between the various operations specified in the execution process, the processing objects of the operations, the results obtained through the operations, the constraints related to the operations, and the data structure of the execution subjects of the operations in the "execution process", namely the syntax tree.

[0193] Here, Figure 10A An example of the structure of the execution environment information E is shown. Figure 10A As shown, the execution environment information E is information representing the following: candidates for the execution subject that can actually execute each operation specified in the "execution process" in the execution environment 100; the execution time when the execution subject executes the operation; and the non-executable time (usage status) when the execution subject cannot be used in the operation.

[0194] exist Figure 10A The execution environment information E shown specifies that the "adjustment" operation shown in the operation summary columns C1 and C3 of the execution process can be executed by the culture medium mixing device P, but cannot be executed by the cell culture devices X, Y and flow cytometer Q. Furthermore, the execution time when the culture medium mixing device P executes the "adjustment" is specified (the adjustment of culture medium A is 30 minutes, and the adjustment of culture medium B is 30 minutes).

[0195] Furthermore, the execution environment information E may specify, for example, the appropriate temperature and humidity for achieving the target state of the culture media A and B in the device serving as the execution subject (here, the culture medium mixing device P). Furthermore, the execution environment information E may specify information related to the functions and performance of the device serving as the execution subject, such as the device's operating accuracy and standby time.

[0196] Furthermore, the execution environment information E specifies that the "Cultivation" operation shown in the Operation Summary column C2 of the execution flow can be performed by cell culture apparatuses X and Y, but cannot be performed by the culture medium mixing apparatus P and the flow cytometer Q. Furthermore, the execution time for "Cell Culture" by cell culture apparatuses X and Y is specified (cell culture apparatus X requires six days for culturing using culture media A and B. Cell culture apparatus Y requires six days for culturing using culture media A and B).

[0197] Furthermore, the execution environment information E specifies that the "marker gene expression evaluation" operation shown in the operation summary column C5 of the execution flow can only be performed by the flow cytometer Q, and cannot be performed by the culture medium mixing device P or the cell culture devices X and Y. Furthermore, the execution time for the flow cytometer Q to execute the "marker gene expression evaluation" is specified (the flow cytometer Q requires 60 minutes to perform the marker gene expression evaluation).

[0198] In the execution environment information E, as the usage status, a non-executable time is specified, indicating that the cell culture device X cannot perform the "Execute Process" operation from 17:00 on January 3, 2020 to 17:00 on January 6, 2020; and a non-executable time is specified, indicating that the cell culture device Y cannot perform the "Execute Process" operation from 17:00 on January 12, 2020 to 17:00 on January 15, 2020.

[0199] In addition, in the execution environment information E according to this embodiment, as a usage condition indicating whether the execution subject can operate, a case is described in which a non-executable time during which the execution subject cannot perform the operation is specified, but the present invention is not limited to this. For example, as a usage condition indicating whether the execution subject can operate, execution environment information E specifying an executable time during which the execution subject can perform the operation (for example, from 9:00 to 16:00 on January 2, 2020) can be used. In addition, these non-executable times and executable times are simply referred to as usage conditions.

[0200] In this embodiment, it can also be configured that in the execution environment 100, a predetermined information processing device ( Figure 1 The execution environment information E is generated by the information processing device on the execution environment 100 side, and is received by the training-related process optimization device 2. Alternatively, each execution subject in the execution environment 100 may send its own individual information to the training-related process optimization device 2 via the communication devices 3a, 3b, 3c, and 3d, and the individual information may be aggregated in the calculation processing unit 7 of the training-related process optimization device 2 to generate the execution environment information E.

[0201] The execution plan generation unit 19 determines the execution subject node ( Figure 8 、 Figure 9A and Figure 9B The execution subject nodes (marked as "actuator") of the operation nodes in the execution process abstract syntax tree t are assigned the execution subject specified by the execution environment information E, without considering the constraints, from the obtained execution environment information E and the execution process abstract syntax tree t. Figure 11 As shown, a set A′ is obtained in which execution subjects are assigned to each operation.

[0202] Here, the set A′ assigns all execution subjects that can execute the operation to the execution subject nodes in the execution flow abstract syntax tree t without considering the constraints, and is a set representing a combination pattern.

[0203] The execution plan generation unit 19 reflects the result of assigning the execution subject set A' to each operation to the execution subject node in the execution flow abstract syntax tree t, and generates Figure 12 、 Figure 13 as well as Figure 14 The extended abstract syntax tree t′ is shown.

[0204] Figure 12 、 Figure 13 as well as Figure 14 Indicates Figure 6 The execution flow shown is assigned Figure 10A This example of an extended abstract syntax tree t′, a tree-structured data structure generated by the execution subject of the execution environment information E, associates the execution subject, execution time, and constraints for each of the operations from adjusting culture media A and B to cell culture and evaluating marker gene expression. The generation of this extended abstract syntax tree t′ will be described later.

[0205] Next, the execution plan generation unit 19 expands the abstract syntax tree t′, as shown in FIG. Figure 10B As shown, in a series of operations executed in sequence, a partially ordered set representing operations that can be executed serially and operations that can be executed in parallel is generated. The execution plan generation unit 19 determines the start time of the execution process (here set to 8:30 on January 2, 2020). For a series of operations in the execution process, based on the partially ordered set that can resolve the parallelism constraints, the execution start time and execution end time are assigned to each element of the set A' according to the order in which the operations are executed, and the result is Figure 15 The set A″ shown.

[0206] Furthermore, in an execution flow where multiple operations can be executed simultaneously, parallelism constraints are defined between operations to determine which operations can or cannot be executed simultaneously. These parallelism constraints can be reflected in the extended abstract syntax tree t′ or partially ordered sets. Furthermore, the parallelism constraints specify whether the overall time efficiency of the culture-related process can be improved by parallelizing only the adjustment of culture medium B with the cell culture using culture medium A.

[0207] here, Figure 15 The set A″ indicates that, starting from the specified start time (8:30 on January 2, 2020), the execution subject is assigned a combination of execution start time and execution end time to each operation in the order of the operations executed in the execution process, regardless of the constraints within the execution process.

[0208] For example, in Figure 15In "No. 2" of the set A", the following time allocation plan is shown for the "culture based on differentiation induction medium A" and "adjustment of differentiation induction medium B" without parallelism constraints: "culture based on differentiation induction medium A" is performed in the "cell culture device X" until 17:00 on January 13, 2020, and from 16:30 on January 15, 2020, after several days, the next "adjustment of differentiation induction medium B" is performed using the "culture medium mixing device P", and then, the "culture based on differentiation induction medium B" performed using the "cell culture device Y" and the "marker gene expression evaluation" performed using the "flow cytometer Q" are performed in sequence.

[0209] In addition, Figure 15 In "No. 4" of the set A", regarding the "culture based on differentiation induction medium A" and "adjustment of differentiation induction medium B" without parallelism constraints, the following time allocation plan is expressed: the "culture based on differentiation induction medium A" using the "cell culture device Y" (from 9:00 on January 2, 2020 to 9:00 on January 8, 2020) and the "adjustment of differentiation induction medium B" using the "culture medium mixing device P" (from 8:30 on January 8, 2020 to 9:00 on January 8, 2020) are executed in parallel, and then immediately thereafter, the "culture based on differentiation induction medium B" using the "cell culture device X" and the "marker gene expression evaluation" are performed using the "flow cytometer Q".

[0210] In addition, Figure 15 In "No.3" and "No.5" of the set A", after the "adjustment of the differentiation induction medium B" by the "medium mixing device P" is completed, after more than 60 minutes, the "culture based on the differentiation induction medium B" is carried out by the "cell culture device Y". Figure 6 In the operation summary column C4 regarding "Cell culture using differentiation induction medium B (target cell differentiation)" in the execution flow shown, the constraint condition of "within 60 minutes after completion of adjustment of differentiation induction medium B" specified in constraint condition item 26e is not satisfied.

[0211] In this way, the execution plan generation unit 19 generates all combinations of time allocation plans that have the possibility that the execution subject can execute the operation of the execution process without considering the constraints in the execution process, and defines these combination patterns as a set A".

[0212] Next, the execution plan generation unit 19 reads the constraints of each operation (the conditions specified in the constraint item 26e) from the execution process, the execution process abstract syntax tree t or the extended abstract syntax tree t′, and extracts all the time allocation plans that satisfy the constraints such as the execution time specified in the constraints from the set A″ as candidate plans, and obtains the set A″′ consisting of the candidate plans.

[0213] Figure 16 An example of a set A'' that satisfies the constraints specified in the constraint condition item 26e is shown. For example, Figure 6 In the execution process of , in the operation summary column C4 of "Cell culture with differentiation induction medium B", the constraint condition "within 60 minutes after the adjustment of differentiation induction medium B is completed" is specified in the constraint condition item 26e. Figure 16 As shown, the execution plan generation unit 19 extracts "No. 1", "No. 2" and "No. 4" that satisfy the above-mentioned constraints as candidate plans, and obtains a set A"' consisting of these candidate plans. In addition, "No. 3" and "No. 5" are not included in the set A"' because they do not satisfy the constraint condition of "within 60 minutes after the adjustment of the differentiation induction medium B" stipulated in the constraint condition 26e of "culture based on differentiation induction medium B".

[0214] Here, in the execution plan generation unit 19, the administrator pre-sets selection conditions, such as selecting the candidate plan with the earliest execution completion time for "marker gene expression evaluation," the final operation of the execution process. Thus, the execution plan generation unit 19 selects a candidate plan from the set A"' that meets the selection conditions as the final execution plan.

[0215] exist Figure 16 In the set A″′ shown, the candidate plan with the earliest execution completion time is "No. 4". In addition, in "No. 4", after the "adjustment of differentiation induction medium A", the "culture based on differentiation induction medium A" and the "adjustment of differentiation induction medium B" without parallelism constraints are executed in parallel. After the "culture based on differentiation induction medium A" and the "adjustment of differentiation induction medium B" are completed at the same time, the "culture based on differentiation induction medium B" is immediately executed, thus becoming the plan that achieves the greatest time reduction among "No. 1", "No. 2" and "No. 4". In addition, since parallelism constraints are not considered in "No. 2", it becomes the plan with the longest execution time.

[0216] Furthermore, when there are multiple candidate plans with the earliest end times, for example, an indicator for determining the superiority or inferiority is set in addition to the execution end time, and a plan with a smaller number (No) as an identifier of the candidate plan is selected as the execution plan.

[0217] The execution plan generation unit 19 sends the selected execution plan to the execution instruction information generation unit 20. The execution instruction information generation unit 20 refers to the execution environment information E and generates execution instruction information for each execution subject that executes each operation, instructing the execution to proceed according to the execution plan.

[0218] Here, execution instruction information is data that describes information sufficient to cause the execution subject to execute each operation of the "execution flow" in the execution environment 100 according to the execution plan. Typically, this information includes setting information such as a program for controlling the operation of the culture medium mixer P, cell culture devices X and Y, and flow cytometer Q.

[0219] Based on the list of execution processes, the operation processing unit 7 sends the execution plans and execution instruction information generated for each execution process to the sending / receiving unit 11, and then sends them from the sending / receiving unit 11 to the communication devices 3a, 3b, 3c, and 3d of the corresponding execution entities via the network 4.

[0220] The communication devices 3a, 3b, 3c, and 3d that receive the execution plan and execution instruction information respectively prompt the culture medium mixing device P, cell culture devices X, Y, and flow cytometer Q as the corresponding execution subjects to perform various operations in the execution environment 100.

[0221] As a result, in the cultivation-related process optimization system 1, within the execution environment 100, based on the execution plan and execution instruction information, each execution subject executes the "execution process" operation respectively. As a result, when the execution results and evaluation results of the execution process are obtained, these execution results and evaluation results are sent from the communication devices 3a, 3b, 3c, and 3d to the cultivation-related process optimization device 2 each time.

[0222] When the cultivation association process optimization device 2 receives execution results and evaluation results from the communication devices 3a, 3b, 3c, and 3d of the execution environment 100, the execution process, the variable parameter values ​​set in the execution process, the execution results, and the evaluation results are recorded correspondingly in the database 8 each time, and these execution results and evaluation results are analyzed by the judgment unit 18 of the operation processing unit 7.

[0223] At this time, the re-planning judgment unit 23 of the judgment unit 18 regenerates the execution plan and execution instruction information for executing the "execution process" according to the execution plan when (i) the actual execution process in the execution environment 100 is different from the execution plan and it is judged that the "execution process" is not executed according to the execution plan, or (ii) there is an impact on the unexecuted part in the execution plan and it is judged that the execution plan needs to be changed.

[0224] For example, culture based on culture medium A is performed in cell culture apparatus X, and then culture based on culture medium B is performed similarly. According to the execution plan executed in cell culture apparatus X, when cell culture apparatus X executes the "execution process", cell culture apparatus X requires longer than the scheduled time to perform culture based on culture medium A. When an execution result is received from cell culture apparatus X indicating that the subsequent execution of culture based on culture medium B using cell culture apparatus X is delayed, determination unit 18 determines that cell culture apparatus X has not executed the "execution process" according to the execution plan, and regenerates the execution plan and execution instruction information.

[0225] On the other hand, for example, in accordance with an execution plan in which culture using culture medium A is performed in cell culture apparatus X and culture medium B is adjusted in parallel in culture medium mixing apparatus P, when cell culture apparatus X and culture medium mixing apparatus P execute an "execution flow," even if culture using culture medium A in cell culture apparatus X requires more than a predetermined time, this does not affect the adjustment of culture medium B by culture medium mixing apparatus P, and operations such as adjustment of culture medium B or subsequent cell culture using culture medium B using cell culture apparatuses X and Y may be performed according to the execution plan. In this case, even if determination unit 18 receives an execution result from cell culture apparatus X indicating that culture using culture medium A requires more time and cannot be performed according to the execution plan, this does not affect the unexecuted portions of the execution plan (e.g., adjustment of culture medium B, cell culture using culture medium B using cell culture apparatuses X and Y, marker gene expression evaluation, etc.), and thus determines that the execution plan does not need to be changed.

[0226] That is, the judgment unit 18 not only simply determines whether all execution entities have performed operations according to the execution plan, but also even if a part of the execution entities in the execution environment 100 does not perform according to the execution plan, the operation of this part of the execution entity will not affect the operation of other execution entities. The other execution entities perform operations according to the execution plan, and finally determine whether the execution process is completed by the end date and time suggested by the execution plan.

[0227] In this example, even if some execution entities in the execution environment 100 do not execute according to the execution plan, if the execution process ends by the end date and time indicated in the final execution plan, the determination unit 18 determines that the execution can be executed according to the execution plan and there is no need to change the execution plan.

[0228] The culture-related process optimization device (2) regenerates the execution plan and execution instruction information, and sends the regenerated execution plan and execution instruction information to the communication devices 3a, 3b, 3c, and 3d of the corresponding execution entities again, prompting the corresponding execution entities, namely the culture medium mixing device P, cell culture device X, Y, or flow cytometer Q, with the new execution plan and execution instruction information so that they can perform various operations in the execution environment 100.

[0229] In addition, when receiving the execution result and the evaluation result from the communication devices 3a, 3b, 3c, and 3d of the execution environment 100, the continuation determination unit 22 of the determination unit 18 determines whether the variable parameter value setting unit 16 should search again for a new variable parameter value reflecting the execution result and the evaluation result based on whether a continuation command is provided from the administrator via the operation unit 10, whether the evaluation result is the desired evaluation result, whether the evaluation result is obtained a specified number of times, etc., according to the execution result and the evaluation result received from the communication devices 3a, 3b, 3c, and 3d of the execution environment 100.

[0230] When a resume command is received from an administrator via the operation unit 10, for example, the continuation determination unit 22 causes the variable parameter value setting unit 16 to regenerate a regression model (response surface) that includes the newly obtained execution results and evaluation results. Consequently, the calculation processing unit 7, through the variable parameter value setting unit 16, uses this regression model to reselect multiple variable parameter values ​​from the search range through Bayesian optimization or multi-task Bayesian optimization, and generates a list of multiple execution flows with different or identical variable parameter values ​​through the execution flow generation unit 17.

[0231] Furthermore, when the execution flow generation unit 17 generates a list of execution flows with the same variable parameter values, although the execution subject of the execution environment 100 executes the same execution flow multiple times, this is effective from the perspectives of, for example, verifying the reliability of the execution flow and whether the same evaluation results can be obtained. More specifically, in cases where it is foreseeable that the execution results and evaluation results of the training-related process will be highly noisy (even if the evaluation result obtained with a specified variable parameter value exceeds 5% of the previous maximum value, it is unclear from a single observation whether this is due to noise or whether an improvement has actually occurred), obtaining statistics such as the standard deviation or the average allows for more accurate evaluation.

[0232] In this way, the culture-related process optimization device 2 regenerates the execution plan and execution instructions corresponding to the newly generated execution process, and resends them to the communication devices 3a, 3b, 3c, and 3d of the execution subject in the execution environment 100, prompting the execution plan and execution instruction information to the corresponding execution subjects in the execution environment 100 (i.e., the culture medium mixing device P, cell culture devices X, Y, and flow cytometer Q), and performing various operations in the culture medium mixing device P, cell culture devices X, Y, and flow cytometer Q in the execution environment 100.

[0233] In this way, the culture-related process optimization system 1 repeatedly performs: generation of execution plans and execution instruction information; prompting of the execution subjects (culture medium mixing device P, cell culture devices X, Y and flow cytometer Q) of the execution plans and execution instruction information; acquisition of execution results and evaluation results from the execution subjects based on this; and setting of new variable parameter values ​​reflecting the obtained execution results and evaluation results.

[0234] Therefore, in the training-related process optimization system 1, the execution subject can reflect the execution results and evaluation results of the training-related process actually executed in the execution environment 100, and can search for the optimal variable parameter values ​​and execution plans that can obtain the greatest possible gain, and can find the optimal training-related process with the largest gain.

[0235] Next, use Figure 17A The flowchart of the above-mentioned cultivation-related process optimization method is described in detail. In addition, in the cultivation-related process optimization system 1, it is preferred that each execution subject is prompted with multiple execution plans and execution instruction information generated by each listed execution process, and multiple different execution processes are executed simultaneously in the execution environment 100. At this time, sometimes the execution results and evaluation results are received separately for each different execution process. For such a case, use Figure 17B In the following description, Figure 17A In the flowchart of , in order to simplify the description, the following description focuses on the processing of one execution flow.

[0236] like Figure 17A As shown in the flowchart, the training-related process optimization device 2 begins the training-related process optimization process from the "Start" step. In subroutine SR1, the training-related process optimization device 2 performs a template execution flow generation process to generate a template execution flow. In the next subroutine SR2, the training-related process optimization device 2 performs a variable parameter value setting process to select multiple variable parameter values ​​from the search range of the variable parameter item 26g specified in the template execution flow.

[0237] In the following step S3, the training-related process optimization device 2 generates a list of multiple execution flows with different (or identical) variable parameter values ​​for the variable parameter item 26g. In the following subroutine SR4, the training-related process optimization device 2 performs execution plan generation processing, generating an execution plan for each execution flow generated in step S3. In the following subroutine SR5, the training-related process optimization device 2 performs execution instruction information generation processing, generating execution instruction information for the execution plan generated in subroutine SR4.

[0238] In the next step S6, the training-related process optimization device 2 sends the execution plan and execution instruction information generated for each execution process in the list to the corresponding execution subject in the execution plan. Thus, in the training-related process optimization system 1, in the execution environment 100, each execution subject performs the execution process according to the execution plan and execution instruction information.

[0239] The cultivation-related process optimization device 2 will, in the next step S7, receive the execution results and evaluation results obtained in the execution environment 100 by the sending / receiving unit 11, and in the next step S8, based on the execution plan and execution instruction information, make the execution process executed by the execution subject in the execution environment 100, the variable parameter value at this time, the execution result obtained from the execution environment 100, and the evaluation result also obtained from the execution environment 100 be recorded correspondingly in the database 8.

[0240] In the next step S9, the training-related process optimization device 2 determines whether the execution process is being executed according to the execution plan presented to the execution subject based on the execution results and evaluation results. If the execution process is not being executed according to the execution plan presented to the execution subject, it is determined that the execution plan needs to be changed (Yes). In step S10, the training-related process optimization device 2 analyzes the unexecuted portion of the execution plan and generates an execution plan that can be executed in the execution environment 100 through execution plan generation processing.

[0241] In addition, the training-related process optimization device 2 generates execution instruction information of the newly generated execution plan in the next step S11, and sends the newly generated execution plan and the execution instruction information to the corresponding execution subjects in the execution plan in the next step S12.

[0242] On the other hand, in the above-mentioned step S9, when the execution process is executed according to the execution plan prompted to the execution subject, there is no need to change the execution plan (No), and the training-related process optimization device 2 reflects the execution results and evaluation results obtained from the execution subject of the execution environment 100 in the next step S13, and determines whether to search again for the optimal variable parameter value from the search range of the template execution process.

[0243] For example, an administrator who has confirmed the execution results and evaluation results obtained from the execution subject of the execution environment 100 searches for the best variable parameter value again from the search range of the template execution process (Yes) when a command to continue searching for the variable parameter value is provided via the operation unit 10, or when the system is set to search for the variable parameter value for the evaluation result obtained from the execution subject of the execution environment 100 until a predetermined evaluation result is obtained, or when the system is set to search for the variable parameter value only a specified number of times after obtaining the execution results and evaluation results from the execution subject of the execution environment 100.

[0244] At this point, the training-related process optimization device 2 returns to subroutine SR2 again and repeats the above process until a negative result (No) is obtained in step S13. On the other hand, in step S13, if the training-related process optimization device 2 determines that the optimal variable parameter value is not to be searched again from the search range of the template execution process, the above training-related process optimization process ends.

[0245] Here, in Figure 17A In the flowchart of , for the sake of simplicity, the focus is on one execution process, and it is explained that when the best execution result and evaluation result are obtained, it is determined whether to end the cultivation-related process optimization processing. Therefore, regarding the summary of the processing of the cultivation-related process optimization device 2 when multiple execution plans and execution instruction information are prompted to the execution subject and the cultivation-related process optimization device 2 receives the execution results and evaluation results of multiple different execution processes from each execution subject, the following is used: Figure 17B Provide explanation.

[0246] exist Figure 17B In the example shown, four execution flows 0, 1, 2, and 3 are presented as execution flows for the cultivation-related process. Thus, in the cultivation-related process optimization system 1, it is preferred that multiple different execution flows be presented to each execution subject via the execution plan and execution instruction information, and that the multiple different execution flows be executed in parallel in the execution environment 100.

[0247] Here, assume that, regarding execution process 0, the adjustment of culture medium A, cell culture using culture medium A, adjustment of culture medium B, and cell culture using culture medium B have completed, resulting in the optimal execution result. On the other hand, assume that, regarding execution process 1, the adjustment of culture medium A (or adjustment of culture medium B) failed or was delayed, resulting in an execution result indicating that some operations (culture) were not executed. Assume that execution process 2 was not executed, and "execution process 3" is currently being executed, and no execution results have been obtained for each of these operations.

[0248] In this case, if Figure 17BAs shown, after receiving the execution result of execution process 0, the culture-related process optimization device 2 receives the execution result of an unexecuted operation (culture) due to the failure or delay of the adjustment of culture medium A or culture medium B regarding other execution processes 2.

[0249] At this time, in the cultivation-related process optimization device 2, the execution result of the optimal execution process 0 is stored in the database 8 for use in the subsequent generation of the regression model. In addition, in the cultivation-related process optimization device 2, even after receiving the optimal execution result for the execution process 0, the cultivation-related process optimization processing is not immediately terminated. The subsequent non-optimal execution result of the execution process 1 is also stored in the database 8. Such non-optimal execution results can also be used as reference data in the future as needed and flexibly used when generating new execution processes.

[0250] In this way, in the cultivation-related process optimization system 1, only one execution process 0 is focused on without ending the cultivation-related process optimization processing. The execution results and evaluation results of other execution processes 1, 2, and 3 are obtained as needed and stored in the database 8. It is preferred to end the cultivation-related process optimization processing based on the situation and the judgment of the manager, etc.

[0251] (1-2) Template execution process generation processing

[0252] (1-2-1) Configuration of the template execution process generation unit

[0253] Next, the template execution flow generation process for generating the above-mentioned template execution flow will be described. Figure 18 1 is a block diagram showing the configuration of the template execution process generation unit 15. Figure 18 As shown, the template execution process generation unit 15 has: a starting point execution process acquisition unit 1501, an execution performance result and evaluation performance result acquisition unit 1502, an associated execution process analysis unit 1503, a variable parameter item determination unit 1504, a search range setting unit 1505, a template execution process output unit 1506 and a constraint condition setting unit 1507.

[0254] Here, the starting execution process acquisition unit 1501 obtains the following information from the database 8 based on the contents of the culture-related processes and evaluation processes such as "adjustment of differentiation induction medium A", "culture based on differentiation induction medium A", "adjustment of differentiation induction medium B", "culture based on differentiation induction medium B", and "marker gene expression evaluation" input through the operation unit 10. Figure 2 The starting point execution flow that is common to the displayed contents (for example, processing objects, products, operations, etc.) and serves as the starting point for optimization.

[0255] The execution performance result and evaluation performance result acquisition unit 1502 is based on the contents of the training related process and the evaluation process input via the operation unit 10 and the contents of the operation summary column of the starting execution process, such as Figure 3 and Figure 4 As shown, when there are no execution performance results and evaluation performance results of the execution process that are identical to the starting point execution process in database 8, or when there are no execution performance results and evaluation performance results of the execution process that are identical to the starting point execution process, the execution performance results and evaluation performance results of the execution process (associated execution process) associated with the starting point execution process are obtained from database 8.

[0256] In addition, here, a case is described in which execution performance results and evaluation performance results that are identical to the contents of the training-related process and evaluation process input via the operation unit 10 and the contents of the operation summary column in the starting execution process are obtained from the database 8. However, for example, execution performance results and evaluation performance results that are similar to the contents of the input training-related process and evaluation process, and execution performance results and evaluation performance results that are similar to the contents of the operation summary column in the starting execution process can also be obtained from the database 8.

[0257] Here, the so-called similarity is determined, for example, based on the operation contents and their order contained in the culture-related process and the evaluation process (hereinafter collectively referred to as the process (Prosess)), and the names and characteristic values ​​of the inputs and outputs of the process. If it is the operation contents and their order, it is preferred to compare the standardized intermediate representations. Specifically, for example, the distance between the abstract syntax trees of the processes is calculated. In addition, based on the abstract syntax tree of a process, a tree automaton is created that accepts the abstract syntax trees of processes with similar representations, and a judgment is made as to whether they are accepted. For the comparison of the inputs and outputs of the processes, if it is simply a culture medium or a cell, it is preferred to compare the distance between the names, the content of each component of the culture medium, the genotype of the cells and other characteristic values.

[0258] By pre-determining such a definition of similarity in the execution performance result and evaluation performance result acquisition unit 1502, it is possible to obtain from the database 8 execution performance results and evaluation performance results that are similar to the contents of the input training-related process and evaluation process, as well as execution performance results and evaluation performance results that are similar to the contents of the operation summary column in the starting execution process.

[0259] If the execution performance results and evaluation performance results of an execution process identical to the starting execution process do not exist in database 8, and the associated execution process parsing unit 1503 reads the execution performance results and evaluation performance results of an execution process associated with the starting execution process (associated execution process) from database 8, it parses the execution performance results and evaluation performance results of the associated execution process to determine the variable parameter items and their search ranges from the starting execution process. The parsing of the execution performance results and evaluation performance results of the associated execution process and the determination of the variable parameter items and their search ranges from the starting execution process will be described later.

[0260] Based on the execution performance results and the evaluation performance results, the variable parameter item determination unit 1504 determines, from the starting execution flow, execution parameter items 26d and other items for which variable parameter items 26g can be set. The search range for the variable parameter value is set in variable parameter item 26g. The search range setting unit 1505 determines the search range set in variable parameter item 26g based on the execution performance results and the evaluation performance results, and sets the search range.

[0261] Alternatively, the administrator may set the variable parameter item 26g and its search range via the operation unit 10. The search range set in the variable parameter item 26g may include, for example: (i) a search range that further expands a predetermined range based on the range of variable parameter values ​​determined by the operation processing unit 7 or the administrator, indicating the maximum variable estimation range; or (ii) a search range that further narrows the range of variable parameter values ​​determined by the operation processing unit 7 or the administrator to a predetermined range that is considered promising (e.g., a range determined based on the administrator's past experience and knowledge, constraints, etc.), indicating the minimum variable estimation range. Furthermore, an example of narrowing the search range of variable parameter values ​​to an optimal range that avoids the content of the patent gazette based on patent publications or published patent publications, etc., will be described in detail in the fourth embodiment described below.

[0262] The search range setting unit 1505 can also perform inverse operations based on the deviation of the allowed quality, etc. while optimizing the variable parameter values ​​in an actual production site, for example, to narrow the search range of the variable parameter items, or narrow the search range to optimize the cumulative value of a series of evaluation performance results.

[0263] The template execution flow output unit 1506 sets the search range determined by the search range setting unit 1505 as the starting execution flow, generates a template execution flow based on the starting execution flow, and outputs the template execution flow. The constraint condition setting unit 1507 adds new constraints to the constraint condition item 26e in the template execution flow or modifies the constraints as needed.

[0264] (1-2-2) When there are no execution results or evaluation results for the same execution process as the starting point execution process

[0265] Here, if the execution performance result and evaluation performance result of the execution process identical to the starting execution process do not exist in the database 8, the following is used. Figures 19 to 28 , explains the situation of parsing the execution performance results and evaluation performance results of the associated execution process, and determining the variable parameter items and their search ranges from the starting point execution process.

[0266] Figure 19 It is a schematic diagram showing an example of a starting point execution process, which stipulates: a culture-related process, adjusting culture medium A, culturing cells based on culture medium A, adjusting culture medium B, culturing cells cultured in culture medium A in culture medium B to produce cells under desired conditions; and an evaluation process, performing marker gene expression evaluation on the obtained cells.

[0267] In this case, the starting point execution process acquisition unit 1501, for example, inputs the terms "adjustment of differentiation induction medium A", "adjustment of differentiation induction medium B", "culture based on differentiation induction medium A", and "culture based on differentiation induction medium B" through the operation unit 10 to determine the culture-related process that the manager wants to optimize. Furthermore, when the term "marker gene expression evaluation" for determining its evaluation process is input through the operation unit 10, the execution process of the same culture-related process and evaluation process is searched and acquired from the database 8 based on these terms as the starting point execution process.

[0268] In addition, for simplicity of explanation, although the example described here is of searching for and retrieving execution flows containing the same term as the term input by the administrator from database 8 as the starting execution flow, the present invention is not limited to this. For example, in addition to the term itself, the starting execution flow may be retrieved from database 8 based on the type represented by the term (e.g., operation, input, output, constraint, etc.), the flow of the operation, the order in which the processing objects are used before the operation, the order in which the artifacts obtained by the operation appear, the structure of the intermediate representation (e.g., abstract syntax tree), etc., and a predetermined degree of similarity between these.

[0269] Figure 19 An example of the configuration of the starting point execution process obtained by searching from the database 8 is shown. Figure 19 As shown, the starting point execution process specifies the contents related to the operations performed in the training association process and the evaluation process in the operation summary columns C6, C7, C8, C9, and C10 for each operation. Figure 2 Similarly, the starting point execution process includes, for example, an operation item 26a that specifies the content of the operation, an input item 26b that specifies the processing object of the operation, an output item 26c that specifies the result obtained by the operation, an execution parameter item 26d that specifies the numerical value associated with the operation, and a constraint item 26e that specifies the constraint conditions related to the operation.

[0270] Next, the execution performance result and evaluation performance result acquisition unit 1502 searches whether the execution performance results and evaluation performance results of the execution process that are the same as the starting point execution process are recorded in the database 8. However, if the execution performance results and evaluation performance results of the execution process that are the same as the starting point execution process are not recorded in the database 8, the execution performance results and evaluation performance results of the execution process (associated execution process) associated with the starting point execution process are obtained from the database 8.

[0271] In this case, the execution performance result and evaluation performance result acquisition unit 1502, for example, based on the terms such as "adjustment of differentiation induction medium A", "adjustment of differentiation induction medium B", "cultivation based on differentiation induction medium A", "cultivation based on differentiation induction medium B" specified in the culture association process in the starting execution process, and the terms such as "marker gene expression evaluation" specified in the evaluation process, determines the execution process containing these terms from the execution processes recorded in the database 8, uses the determined execution process as the associated execution process, and obtains the execution performance results and evaluation performance results of the associated execution process from the database 8.

[0272] Here, Figure 20 and Figure 21 This shows an example of how to associate the execution performance results of the execution process with the evaluation process. Figure 22 Indicate these Figure 20 and Figure 21 The evaluation performance results of the related execution processes (execution processes A and Z) are shown.

[0273] exist Figure 20 In the associated execution flow (execution flow A) shown, although there is no Figure 19 The operation summary column C8 of the "adjustment of differentiation induction medium B" and the operation summary column C9 of the "cell culture using differentiation induction medium B" of the starting execution process are shown, but the terms "adjustment of differentiation induction medium A", "cell culture using differentiation induction medium A", and "marker gene expression evaluation" in the starting execution process are included in the operation summary columns C6, C12, and C13. Figure 20 The associated execution flow shown and Figure 22The evaluation performance results of the execution process A shown are the execution performance results and evaluation performance results of the associated execution process.

[0274] In addition, Figure 21 In the associated execution process (execution process B) shown, all the operation summary columns of the starting execution process are present. That is, in the operation summary columns C6, C12, C13, C14, and C15, the same terms as those in the starting execution process are included, such as "adjustment of differentiation induction medium A," "cell culture based on differentiation induction medium A," "adjustment of differentiation induction medium B," "cell culture based on differentiation induction medium B," and "marker gene expression evaluation." As a result, the execution performance result and evaluation performance result acquisition unit 1502 acquires Figure 21 The associated execution flow shown and Figure 22 The evaluation performance results of the execution process B shown are the execution performance results and evaluation performance results of the associated execution process.

[0275] The associated execution process analysis unit 1503 Figure 20 and Figure 21 The content of the execution performance results shown in the operation summary column C6 of each training-related process of the related execution flow is the same as Figure 22 Compare the contents of each evaluation performance result shown; analyze the correlation and dependency between the contents of the execution performance results obtained in the training association process (such as changes in the numerical values ​​of the execution parameter values) and the contents of the evaluation performance results obtained in the evaluation process; and infer the execution parameter items of the training association process that affect the evaluation performance results.

[0276] According to the associated execution process analysis unit 1503 of this embodiment, for example, as Figure 22 The "marker gene expression positivity rate" of the evaluation performance result of the "execution process B" shown is "82%", which is higher than that of the "execution process A". Therefore, it can be assumed that similarly, it is sufficient to search and derive an execution parameter value that is close to the value ("82%") of the "marker gene expression positivity rate" of the evaluation performance result of the "execution process B".

[0277] The associated execution process analysis unit 1503, for example, analyzes the contents of the "execution parameter values" of the execution processes A and B. Regarding the "basal culture medium concentration", "BMP4 concentration" and "glucose concentration", the values ​​of "execution process A" and "execution process B" are equal. Regarding the "VEGF concentration", the analysis result is that "execution process A" is "10μM" and the value is larger than "5μM" of "execution process B".

[0278] As a result, the associated execution process analysis unit 1503 obtains the following analysis result: If the value of "VEGF concentration" in the "execution parameter" is increased, it may be possible to produce cells with a "positive rate of marker gene expression" close to that obtained through "execution process B". In addition, only the execution performance results of execution processes A and Z are illustrated here, but the variable parameter value can also be determined by comparing the execution parameter values of multiple evaluation performance results executed in the past, for example, hundreds of evaluation performance results. At this time, the associated execution process analysis unit 1503 can also compare the execution parameter values of multiple processes that obtain evaluation performance results with a "positive rate of marker gene expression" value close to "82%", for example, within a range of about "5%", i.e., "77% - 87%".

[0279] The associated execution process analysis unit 1503 sends such an analysis result to the variable parameter item determination unit 1504. Based on the analysis result received from the associated execution process analysis unit 1503, the execution performance result of the associated execution process, and the evaluation performance result, as Figure 23 shown, the "BMP4 concentration", "VEGF concentration", and "glucose concentration" of the execution parameter item specified in the operation summary column C6 of "adjustment of differentiation induction medium A" in the starting execution process are set as the variable parameter item 26j.

[0280] The variable parameter item determination unit 1504 estimates the search range of the variable parameter item 26j based on the analysis result from the associated execution process analysis unit 1503, the execution performance result and evaluation performance result of the associated execution process, and the starting execution process in which the "BMP4 concentration", "VEGF concentration", and "glucose concentration" in the operation summary column C6 are set as the variable parameter item 26j. For example, it is set as "0 μM < BMP4 concentration < 10 μM", "0 μM < VEGF concentration < 15 μM", and "3 μM < glucose concentration < 15 μM". In addition, regarding the "basic medium concentration" here, a single value is set as the "execution parameter" instead of the "variable parameter". As a result, the template execution process output unit 1506 outputs the template execution process as Figure 23 shown.

[0281] In addition, Figure 24 and Figure 25 represent an example of the configuration of the execution process generated based on the Figure 23 shown template execution process. In the arithmetic processing unit 7, when the variable parameter value setting unit 16 receives Figure 23When the template execution process shown is executed, the variable parameter value setting unit 16 performs variable parameter value setting processing, selects multiple variable parameter values ​​from the search range of the variable parameter item 26j (in this example, "BMP4 concentration" is "5μM" or "6μM", etc.), and generates a list of multiple execution processes with each variable parameter value set through the execution process generation unit 17.

[0282] (1-2-3) Template execution process generates a flowchart of the processing flow

[0283] Next, use Figure 26 The flowchart of the template execution process generation process is described. Figure 26 As shown, the training-related process optimization device 2 starts the template execution process generation processing flow from the "start" step. In the next step S201, the manager inputs the training-related process and evaluation process to be optimized.

[0284] In the next step S202 , the template execution flow generation unit 15 acquires a start point execution flow, which is a start point of the search, from the database 8 based on the cultivation-related process and the evaluation process.

[0285] For simplicity, the description here describes a case where a search is performed on database 8 to obtain a starting execution flow as the search starting point. However, the present invention is not limited to this. For example, even if a search is performed on database 8 and a starting execution flow as the search starting point does not exist and cannot be obtained from database 8, a new starting execution flow can be created in the template execution flow generation unit 15 to obtain the new starting execution flow. Alternatively, an administrator can create a new starting execution flow through the operation unit 10. Alternatively, the template execution flow generation unit 15 can automatically create a rough starting execution flow based on the operation name and device name input by the administrator, using the specified training-related process and evaluation process operations as a template.

[0286] In the next step S203, the template execution flow generation unit 15 determines whether there are past execution performance results and evaluation performance results for the same execution flow as the starting execution flow in the database 8. If the template execution flow generation unit 15 determines that there are past execution performance results and evaluation performance results for the same execution flow as the starting execution flow in the database 8 (Yes), in the next step S204, the template execution flow generation unit 15 obtains the past execution performance results and evaluation performance results for the same execution flow as the starting execution flow from the database 8.

[0287] In the next step S205 , the template execution flow generation unit 15 determines a variable parameter item for which a variable parameter value can be set in the start point execution flow based on the past execution performance results and evaluation performance results acquired in step S204 .

[0288] In addition, regarding the setting of such variable parameter items, the administrator may carefully check the items during the execution process from the start point, and set predetermined items as variable parameter items based on the input of a selection command by the administrator via the operation unit 10 .

[0289] In the next step S206, the template execution flow generation unit 15 estimates the search range that can be set in the variable parameter item based on the past execution performance results and evaluation performance results obtained in step S204, and sets the estimated predetermined search range in the variable parameter item. In the next step S210, the template execution flow generation unit 15 outputs the template execution flow with the search range set in the variable parameter item, thereby terminating the above-described template execution flow generation process.

[0290] Furthermore, regarding the setting of the search range in such a variable parameter item, the administrator may carefully check the range of the variable parameter value and set the search range based on an input command input by the administrator via the operation unit 10 .

[0291] On the other hand, in the above-mentioned step S203, if the template execution process generation unit 15 determines that there are no execution performance results and evaluation performance results of the execution process identical to the starting execution process in the database 8 (No), then in the next step S207, based on the input training association process and evaluation process, the execution performance results and evaluation performance results of the associated execution process associated with the starting execution process are obtained from the database 8.

[0292] In the next step S208 , the template execution process generation unit 15 compares the execution performance results and the evaluation performance results of the plurality of related execution processes, and estimates the execution parameter items that affect the evaluation performance results from the starting execution process.

[0293] In the following step S209, the template execution flow generation unit 15 sets the execution parameter items that affect the evaluation performance results as variable parameter items. Based on the execution performance results and evaluation performance results of the associated execution flow, it estimates the search range that can be set in the variable parameter items. The template execution flow generation unit 15 then sets the estimated, specified search range in the variable parameter items. In the following step S210, the template execution flow generation unit 15 outputs the template execution flow with the search range set in the variable parameter items, thus terminating the aforementioned template execution flow generation process.

[0294] (1-3) Variable parameter value setting process

[0295] Next, the above-mentioned variable parameter value setting process flow will be described. Figure 27 1 is a block diagram showing the structure of the variable parameter value setting unit 16. Figure 27 As shown, the variable parameter value setting unit 16 includes a variable parameter value analyzing unit 1601 and a variable parameter value selecting unit 1602. The variable parameter value analyzing unit 1601 obtains, for example, execution performance results and evaluation performance results that are identical to or related to the starting execution flow from the database 8, and analyzes variable parameter values ​​that affect the evaluation performance results from within the search range using a regression model or the like that uses these execution performance results and evaluation performance results.

[0296] As regression models, various regression models that apply linear or nonlinear transformations can be used, such as principal component analysis (PCA), which uses the execution parameter values ​​of the execution performance results at the locations corresponding to the variable parameter items whose search range is set in the template execution process as explanatory variables and the evaluation performance results as the target variable; partial least squares (PLS), polynomial regression, Gaussian process regression, random forest regression, etc., which use the execution parameter values ​​of the execution performance results at the locations corresponding to the variable parameter items whose search range is set in the template execution process and the evaluation performance results as explanatory variables and the evaluation performance results as the target variable. Furthermore, as a regression model, for example, a machine learning model generated by machine training the execution performance results and the evaluation performance results can also be used.

[0297] As a machine learning model, for example, a neural network defined in the form of a combination of at least one matrix operation and at least one linear or nonlinear transformation can be applied. When training a machine learning model, there are supervised training and unsupervised training. When supervising an untrained machine learning model, it includes: using execution performance results, their evaluation performance results, and correct answer labels indicating whether the evaluation performance results are the expected results (for example, the accuracy of whether the evaluation performance results are the expected results), and adding correct answer labels for training when making the model learn execution performance results and evaluation performance results.

[0298] When performing unsupervised training on an untrained machine learning model, the execution performance results and their evaluation performance results are used to enable the model to learn the regularities and characteristics of the execution performance results and the evaluation performance results.

[0299] The variable parameter value setting unit 16 inputs an arbitrarily selected variable parameter value into the trained machine learning model in the variable parameter value analysis unit 1601 to obtain a prediction result (for example, the accuracy rate if it is supervised training, and the prediction evaluation result if it is unsupervised training), and selects a variable parameter value based on the prediction result in the variable parameter value selection unit 1602.

[0300] Furthermore, it is preferable to add execution results and evaluation results received from execution environment 100 to such a regression model each time as explanatory variables, target variables, and training data. This allows for the generation of a regression model that reflects the latest data on the execution process executed in execution environment 100, further optimizing the process of cultivating associations.

[0301] The variable parameter value selection unit 1602 selects a predetermined number of variable parameter values ​​from the search range of the variable parameter items based on the analysis results of the variable parameter value analysis unit 1601, for example, using Bayesian optimization, orthogonal arrays, or Latin hypercube methods.

[0302] Specifically, for example, the variable parameter value selection unit 1602 selects variable parameter values ​​based on past execution performance results and evaluation performance results using the regression model generated by the variable parameter value analysis unit 1601. As a method for selecting variable parameter values ​​using the regression model, in addition to defining a function in the search space based on the regression model and selecting the algorithm by optimizing the function, it is also possible to replace some or all of the variable parameter values ​​generated in this way at the discretion of the administrator.

[0303] Based on the regression model, the function defined on the search space is, for example, a function that quantitatively defines how much improvement in the evaluation results and how much new information can be obtained by generating a variable parameter value at a certain point in the search space (for example, Upper Confidence Bound, Expected Improvement, parallel Knowledge Gradient, Mutual Information, etc.). By optimizing given steps of these functions (for example, maximization, minimization, weighted sampling, etc.), a list of a predetermined number of variable parameter values ​​can be obtained.

[0304] When there are multiple variable parameter values ​​and priorities need to be assigned to them, in addition to determining the priority of the algorithm based on a given, appropriately determined benchmark (for example, in addition to sorting based on the values ​​of the above-mentioned function, when variable parameter values ​​are generated successively through methods such as LocalPenalization, the priority is assigned in ascending order according to the order in which the variable parameter values ​​are generated, etc.), the priority can also be arbitrarily assigned based on the judgment of the administrator.

[0305] In addition, the variable parameter value setting unit 16 may be configured to have a function of transforming or limiting the search range based on the attribute value of the variable parameter value (for example, in the case of the concentration of a substance in a solution, the flow rate, etc., when the variable parameter value is logarithmically transformed, the change in the target variable relative to the change in the explanatory variable is consistent, making it easier to see the contribution, for example, when the variable parameter value is logarithmically transformed). The variable parameter value may be estimated from the transformed or limited search range. For example, when it is known that a certain variable parameter value is a variable parameter value of a type that changes logarithmically relative to the evaluation value, the variable parameter value may also be logarithmically transformed accordingly.

[0306] In addition, the variable parameter value setting unit 16 can also perform inverse operations based on the deviation of the allowable quality, etc., to narrow the range of variable parameter values ​​from the search range, or select variable parameter values ​​from the search range to optimize the cumulative value of a series of evaluation performance results, for example, when optimizing while ensuring a certain quality in an actual production site.

[0307] For example, using Figure 28 The flowchart of the variable parameter value setting process is used to illustrate the above-mentioned variable parameter value setting process. The variable parameter value setting unit 16 starts the variable parameter value setting process from the start step. In the next step S301, for example, based on past execution performance results and evaluation performance results, a regression model for the search range is generated.

[0308] In the next step S302 , the variable parameter value setting unit 16 selects a plurality of variable parameter values ​​within the search range for the variable parameter item for which the search range is set, based on the regression model, and ends the variable parameter value setting process flow.

[0309] (1-4) Execution plan generation process

[0310] (1-4-1) Overview of Execution Plan Generation Processing

[0311] Next, the above-mentioned execution plan generation process flow will be described. Figure 29 : is a block diagram showing the structure of the execution plan generation unit 19. Figure 30 This is a flowchart showing the execution plan generation process. Figure 6 The following describes an example of an execution plan for the execution flow shown.

[0312] like Figure 29As shown, the execution plan generation unit 19 includes: an individual abstract syntax tree generation unit 1901, an execution process abstract syntax tree generation unit 1902, an extended abstract syntax tree generation unit 1903, an execution environment information acquisition unit 1904, a partially ordered set generation unit 1905, a candidate plan generation unit 1906 and an execution plan selection unit 1907.

[0313] In this case, if Figure 30 As shown, the execution plan generation unit 19 starts the execution plan generation process from the start step and moves to the next subroutine SR41 and step S404. In the subroutine SR41, the execution plan generation unit 19 performs individual abstract syntax tree generation processing (described later) by the individual abstract syntax tree generation unit 1901, as shown in FIG. Figure 6 As shown, a separate abstract syntax tree is generated for each operation specified in the operation summary columns C1, C2, C3, C4, and C5 in the execution flow.

[0314] here, Figure 31 Indicates that individual abstract syntax trees are generated, starting from Figure 6 The structure of the individual abstract syntax tree generated in the operation summary column C1 of the "adjustment of differentiation induction medium A" in the execution flow shown. Figure 32 Indicates that individual abstract syntax trees are generated, starting from Figure 6 The structure of the individual abstract syntax tree generated in the operation summary column C2 of the "Cell culture using differentiation induction medium A" of the execution flow shown.

[0315] Figure 33 Indicates that individual abstract syntax trees are generated, starting from Figure 6 The structure of the individual abstract syntax tree generated in the operation summary column C3 of the "adjustment of differentiation induction medium B" in the execution flow shown. Figure 34 Indicates that individual abstract syntax trees are generated, starting from Figure 6 The structure of the individual abstract syntax tree generated in the operation summary column C4 of the "Cell culture using differentiation induction medium B" of the execution flow shown.

[0316] On the other hand, in step S404, the execution plan generation unit 19 obtains the execution environment information from the database 8 or the execution environment 100 through the execution environment information acquisition unit 1904. Figure 10A The execution environment information E shown.

[0317] The execution plan generation unit 19 performs execution flow abstract syntax tree generation processing (described later) in the subroutine SR42 through the execution flow abstract syntax tree generation unit 1902, integrates the multiple individual abstract syntax trees generated in the above subroutine SR41, and generates the following: Figure 8 、 Figure 9A and Figure 9BThe execution flow abstract syntax tree t is shown. When there is only one individual abstract syntax tree, the individual abstract syntax tree is treated as the execution flow abstract syntax tree.

[0318] In the next subroutine SR43, the execution plan generation unit 19 performs an extended abstract syntax tree generation process (described later) by the extended abstract syntax tree generation unit 1903, so that Figure 8 、 Figure 9A and Figure 9B The execution flow abstract syntax tree t shown reflects Figure 10A The content of the execution environment information E shown is generated as follows Figure 12 、 Figure 13 and Figure 14 The extended abstract syntax tree t′ is shown.

[0319] In the next step S405, the execution plan generation unit 19 generates the following from the extended abstract syntax tree t′ through the partially ordered set generation unit 1905: Figure 10B The execution plan generation unit 19 determines the start time of the execution process (for example, 8:30 on January 2, 2020) by the candidate plan generation unit 1906 in the next step S406, and generates a set A″ ( ) of multiple time allocation plans for each operation in the order in which the operations are executed, each of which is assigned the execution start time and execution end time that the execution subject can execute. Figure 15 ).

[0320] Furthermore, the execution plan generation unit 19 selects a candidate plan that satisfies the constraints specified by the execution flow, etc. from the plurality of time allocation plans through the candidate plan generation unit 1906, and generates a set A″′ of candidate plans ( Figure 16 ).

[0321] In the next step S408, the execution plan generation unit 19 selects an execution plan from multiple candidate plans through the execution plan selection unit 1907 based on pre-set selection conditions such as "selecting the candidate plan with the earliest execution end time of "marker gene expression evaluation" as the final operation of the execution process", and ends the execution plan generation processing flow.

[0322] (1-4-2) Individual Abstract Syntax Tree Generation Process

[0323] Next, the above-mentioned individual abstract syntax tree generation process will be described. Figure 35 This is a flowchart showing an example of a process flow for generating a separate abstract syntax tree. Figure 6 The operation summary column C2 of "Cell culture using differentiation induction medium A" of the execution process 1 is generated. Figure 39Examples of individual abstract syntax trees are shown.

[0324] like Figure 35 As shown, the individual abstract syntax tree generation unit 1901 starts executing the individual abstract syntax tree generation process flow from the "start" step, and in the next step S4101, selects the operation item 26a in the operation summary column C2 for generating the individual abstract syntax tree from the execution flow.

[0325] In the next step S4102, the individual abstract syntax tree generation unit 1901 generates an output item 26c in the operation summary column C2, as shown in FIG. Figure 36 As shown in "step 1" of the output item 26c, the output object node n1 representing the content of the output item 26c (here, "human iPS cells (in the process of differentiation)") is generated. In the next step S4103, the individual abstract syntax tree generation unit 1901 generates the output object node n1 as shown in "step 1" of the output item 26c. Figure 36 As shown in "step 2" of , an output label node n2 indicating "output" is added to the child node of the output target node n1 via an edge.

[0326] In the next step S4104, the individual abstract syntax tree generation unit 1901 generates an operation item 26a in the operation summary column C2, such as Figure 36 As shown in "step 3" of the operation item 26a, the operation content (here, "cell culture based on differentiation induction medium A" indicating the operation content) is displayed. Figure 36 The operation name node n3 (which is only represented as "cell culture") is added to the child node of the output label node n2 via an edge. In the next step S4105, the individual abstract syntax tree generation unit 1901 generates the operation name node n3 as follows: Figure 36 As shown in "step 4" of , the input label node n4 representing "input" is added to the child node of the operation name node n3 via an edge.

[0327] In the next step S4106, the individual abstract syntax tree generation unit 1901 generates an operation summary column C2 based on the input item 26b. Figure 37 As shown in "step 5" of , an input object node n5 representing the content of the input item 26b (here, "differentiation induction medium A" and "human iPS cells") is added via an edge.

[0328] In the next step S4107, the individual abstract syntax tree generation unit 1901 determines whether there is an execution parameter value or a variable parameter value in the operation summary column C2. Figure 6In the operation summary column C2 of the execution flow shown, the execution parameter item 26d contains an execution parameter value (there is also a variable parameter item 26g), and a positive result (Yes) is obtained in step S4107, so the individual abstract syntax tree generation unit 1901 transfers to the next step S4108.

[0329] In step S4108, the individual abstract syntax tree generation unit 1901 generates Figure 37 As shown in "step 6" of FIG, a parameter node group n6 indicating the content of the execution parameter item 26d (or the variable parameter item 26g if there is a variable parameter item 26g) is added to the child node of the operation name node n3 via an edge.

[0330] In addition, parameter node group n6 has: parameter label node n 60 , indicating the node is a parameter; and static label node n 62 , appended to the parameter label node n via the edge 60 In addition, if there is a variable parameter value, it is connected to the parameter label node n. 60 The same method is used to append dynamic label nodes.

[0331] In addition, the static label node n represents the execution parameter value 62 In the example, the execution parameter value (here "training period") is added as a child node through the edge, and the parameter name node n 63 In the child nodes of , a parameter value node n representing the numerical value of the execution parameter is added via an edge 64 (Here, it is "6 days"). In addition, when there is a variable parameter value, as described above, a parameter name node representing the parameter name specified by the variable parameter value is added as a child node via an edge to the dynamic label node representing the variable parameter value, and a parameter value node representing the numerical value of the variable parameter value is added as a child node of the parameter name node via an edge.

[0332] When it is determined in step S4107 that there are no execution parameter values ​​or variable parameter values, the individual abstract syntax tree generation unit 1901 proceeds to the next step S4109 .

[0333] In step S4109, the individual abstract syntax tree generation unit 1901 determines whether there are any constraints related to the operation in the constraint item 26e of the operation summary column C3. If so, the process proceeds to the next step S4110. Whether the constraints specified in the constraint item 26e are directly related to the operation, related to the input item 26b, related to the output item 26c, or related to the execution parameter item 26d or the variable parameter item 26g is determined when the execution flow is generated. Furthermore, while the constraint item 26e of the operation summary column C2 described here contains a constraint related to the input item 26b (here, "within 60 minutes after the completion of the adjustment of differentiation induction medium A"), there are no constraints related to the operation (e.g., "cell culture (the operation in the operation summary column C2) must be within 10 days").

[0334] As described above, there are no constraints related to the operation in the operation summary column C2, and therefore, they are not shown in the diagram. Assuming that there are constraints related to the operation, the individual abstract syntax tree generation unit 1901, in step S4110, adds a constraint node group n7 (not shown) representing the contents of the constraints related to the operation to the child nodes of the operation name node n3 based on the constraint item 26e.

[0335] In this case, in the operation-related constraint condition node group n7, the constraint label node n representing the content of the operation-related constraint condition is 70 The node n3 is appended to the child node of the operation name via an edge, and when the constraint condition is a static condition (a fixed condition that does not change), the static label node n 71 (not shown) Append to constraint label node n via edge 70 (not shown). In addition, as the static label node n 71 The child node of the constraint name node n represents the name of the constrained object 72 (not shown) is added via an edge as the constraint name node n 72 Child node, representing the parameter value node n of the constrained parameter value 73 (Not shown) Added via edges. In addition, when the constraint condition is a dynamic condition, a node with the dynamic label is added in the same process as the above process.

[0336] On the other hand, if it is determined in step S4109 that there are no constraints related to the operation, the individual abstract syntax tree generation unit 1901 proceeds to the next step S4111. Figure 38 As shown in "step 7" of FIG, the actuator label node n8 indicating the execution subject of the operation is added to the child node of the operation name node n3 via an edge.

[0337] In the next step S4112 , the individual abstract syntax tree generation unit 1901 determines whether or not there is a constraint related to the input item 26 b in the constraint item 26 e of the operation summary column C3 , and if so, the process moves to the next step S4113 .

[0338] In step S4113, the individual abstract syntax tree generation unit 1901 generates Figure 39 As shown in "step 8" of FIG, based on the constraint item 26e, a constraint node group n9 indicating the content of the constraint related to the input item 26b is added to the child node of the corresponding input object node n5.

[0339] Here, in the constraint node group n9, the output label node n 90 Append the edge to the child node of the corresponding input object node n5, and change the intermediate label node n 91 Append to output label node n 90 In addition, in the constraint node group n9, the input label node n is connected via the edge 92 Append to intermediate label node n 91 The input object node n representing the input object that serves as the judgment basis for the start and end of the operation specified in the constraint condition is connected to the child node of 93 Append to the input label node n 92 on the child node of (in this case, since the constraint condition of starting the operation "within 60 minutes after the adjustment of differentiation induction medium A is completed" is stipulated, here, "differentiation induction medium A" becomes the judgment basis for starting the operation stipulated by the constraint condition).

[0340] Furthermore, in the constraint condition node group n9, the constraint label node n representing the content of the constraint condition related to the input item 26b is 94 Append to intermediate label node n via edge 91 The child node of the static label node n will indicate that the constraint condition is a static condition (fixed condition that does not change). 95 , Constraint name node n representing the name of the constrained object (here "time") 96 , parameter value node n representing the constrained parameter value (here "0 minutes to 60 minutes") 97, appended serially to the constraint label node n via the edge 94 On the child node of .

[0341] If it is determined in step S4112 that there is no constraint condition associated with the input item 26 b , the individual abstract syntax tree generation unit 1901 proceeds to the next step S4114 .

[0342] In this way, the individual abstract syntax tree generation unit 1901 can generate an individual abstract syntax tree for the operation summary column C2 of the execution flow, for example.

[0343] In step S4114, the individual AST generation unit 1901 determines whether there is an operation item 26a within the execution flow for which an individual AST tree has not been generated. If the individual AST generation unit 1901 determines that there is an operation item 26a within the execution flow for which an individual AST tree has not been generated (Yes), then in the next step S4115, the individual AST generation unit 1901 selects the operation items 26a in the other operation summary columns C1, C3, C4, and C5 within the execution flow for which individual AST trees have not been generated, and then returns to step S4102, repeating the above process until a negative result (No) is obtained in step S4114.

[0344] On the other hand, in step S4114, if the individual abstract syntax tree generation unit 1901 determines that there is no operation item 26a (No) in the execution process for which an individual abstract syntax tree has not been generated, this means that individual abstract syntax trees have been generated for the operation items 26a in all operation summary columns C1, C2, C3, C4, and C5 in the execution process, and the above-mentioned individual abstract syntax tree generation processing flow is terminated.

[0345] (1-4-3) Execution process abstract syntax tree generation processing

[0346] Next, the execution flow abstract syntax tree generation process of integrating the above-mentioned individual abstract syntax trees will be described. Figure 40 This is a flowchart showing an example of the steps for generating an abstract syntax tree for the execution process. Figure 6 In the execution flow shown, the operation summary column C4 of "Cell culture using differentiation induction medium B" generates Figure 41 The individual abstract syntax trees of Figure 42 An example of unifying individual abstract syntax trees.

[0347] While this description describes the case where multiple individual ASTs are generated and then combined, depending on the template execution flow, there may be a single operation summary column, resulting in only one individual AST. In this case, as described above, the generated individual AST is treated as the execution flow AST.

[0348] like Figure 40 As shown, the execution flow abstract syntax tree generation unit 1902 starts the "execution flow abstract syntax tree generation process flow" from the "start" step, and in the next step S4201, from Figure 31 、 Figure 32 、 Figure 33 as well as Figure 34 ( Figure 41 and Figure 34 A predetermined individual syntax tree (here, for example, an individual syntax tree ( Figure 42 )).

[0349] In the next step S4202, the execution flow abstract syntax tree generation unit 1902 performs the following operations: Figure 42 As shown, the set N of input tag nodes without child nodes included in the individual abstract syntax tree of the "evaluation process" selected in the previous step S4201 is determined. l .

[0350] In the next step S4203, as Figure 41 As shown, the execution flow abstract syntax tree generation unit 1902 determines the individual abstract syntax tree of "cell culture based on differentiation induction medium B". The individual abstract syntax tree of "cell culture based on differentiation induction medium B" has the same Figure 42 The input object nodes (here, the input object nodes with the label "(differentiation-induced) human iPS cells") included in the set N1 are n 100 Output object node n with the same label (e.g., a node name that identifies the node) 200 , and in the output object node n 200 Has no parent node.

[0351] In the next step S4204, the execution flow abstract syntax tree generation unit 1902 generates the Figure 41 Output object node n of the individual abstract syntax tree of "cell culture based on differentiation induction medium B" shown 200 Replace with Figure 42 The set N of individual abstract syntax trees for “marker gene expression evaluation” shownl The input object node n contained in 100 ,connect Figure 42 Individual abstract syntax trees and Figure 43 Individual abstract syntax trees are generated such as Figure 43 The intermediate abstract syntax tree shown.

[0352] In the next step S4205, the abstract syntax tree generation unit 1902 determines whether there is a set N of input label nodes that do not have child nodes and that can be matched with the set N of input label nodes that are included in the intermediate abstract syntax tree. l Here, the set N of input label nodes that do not have child nodes included in the intermediate abstract syntax tree is l When there are more than one connectable individual abstract syntax trees and intermediate abstract syntax trees, the execution flow abstract syntax tree generation unit 1902 transfers to the above step S4203 again to determine the new set N that can be connected to the intermediate abstract syntax tree. l Thus, in step S4205, the above process is repeated until there is no set N of input label nodes that do not have child nodes and that can be connected to the intermediate abstract syntax tree. l The individual abstract syntax trees and intermediate abstract syntax trees connected by the elements.

[0353] On the other hand, in step S4205, if there is no set N of input tag nodes that do not have child nodes and that can be included in the intermediate abstract syntax tree, l In the case of an individual abstract syntax tree and an intermediate abstract syntax tree connected by the elements, the execution flow abstract syntax tree generation unit 1902 uses the intermediate abstract syntax tree as the final execution flow abstract syntax tree t′( Figure 8 9) and output, ending the above-mentioned execution process abstract syntax tree generation processing flow.

[0354] (1-4-4) Extended Abstract Syntax Tree Generation Process

[0355] Next, the above-mentioned extended abstract syntax tree generation process flow will be described. Figure 44 This is a flowchart showing an example of an extended abstract syntax tree generation process. Figure 10A The execution environment information E and Figure 8 、 Figure 9A and Figure 9B The execution flow abstract syntax tree t shown in the figure generates an extended abstract syntax tree t′( Figure 12 、 Figure 13 and Figure 14 ) examples.

[0356] like Figure 44 As shown, the extended abstract syntax tree generation unit 1903 starts the "extended abstract syntax tree generation process flow" from the "start" step, and in the next step S4301, Figure 10A The execution environment information E specifies the execution subject. Here, for simplicity, the example of selecting "culture medium mixing device P" as the execution subject is described. In the next step S4302, the extended abstract syntax tree generation unit 1903 determines, based on the execution environment information E, the operations that can be performed by the execution subject "culture medium mixing device P" selected in the previous step S4301.

[0357] For example, in Figure 10A In the execution environment information E shown, it is determined that: if the culture medium mixing device P can become the execution subject, then only the operations of adjusting the culture medium A and adjusting the culture medium B can be performed; if the cell culture devices X and Y can become the execution subject, then only the operation of culturing cells can be performed; if the flow cytometer Q can become the execution subject, then only the operation of evaluating the expression of marker genes can be performed.

[0358] In the next step S4303, the extended abstract syntax tree generation unit 1903 generates an execution subject name node n of "culture medium mixing device P" with the execution subject selected in the previous step S4301 as a label (for example, an execution subject name that can identify the execution subject). x More specifically, for example, if it is a culture medium mixing device P, the execution subject name node n is labeled with "culture medium mixing device P" which can identify the culture medium mixing device P. x If it is a cell culture device X, then the execution subject name node n is labeled with "cell culture device X" which can identify the cell culture device X. x .

[0359] In the next step S4304, the extended abstract syntax tree generation unit 1903 selects the operation name node n3 corresponding to the operation (i.e., "adjustment of the culture medium") that can be performed by the "culture medium mixing device P" of the execution subject selected in the previous step S4301 from the execution process abstract syntax tree t. In step S4301, when the "culture medium mixing device P" that can perform the operation of "adjustment of the culture medium" is selected, the extended abstract syntax tree t' is used. Figure 12 To express, select "adjustment of culture medium" from the execution flow abstract syntax tree t (ie, Figure 9A Operation name node n3 of "adjustment of culture medium").

[0360] In the next step S4305, the extended abstract syntax tree generation unit 1903 adds the execution subject name node n of the "medium mixing device P" via the edge x , as a child node of the actuator label node n8 connected to the operation name node n3 of "adjustment of culture medium" selected in the previous step S4304.

[0361] In the next step S4306, the extended abstract syntax tree generation unit 1903 determines the execution time related to the operation of "adjustment of culture medium" that can be executed by the "culture medium mixing device P" selected in the previous step S4301 according to the execution environment information E, such as Figure 12 As shown, the execution time node n of the execution time (here "30 minutes") is represented x1 Node n as the execution subject name of "culture medium mixing device P" x The child nodes of are appended via edges.

[0362] In the next step S4307, the extended abstract syntax tree generation unit 1903 determines whether other operation name nodes n3 corresponding to other operations executable by the "culture medium mixing device P" selected in the previous step S4301 exist in the execution process abstract syntax tree t.

[0363] For example, when "culture medium mixing device P" is selected, other operations that can be performed by "culture medium mixing device P" are as follows: Figure 9B The operation "adjustment of culture medium" also exists in the execution flow. Therefore, in the next step S4308, the extended abstract syntax tree generation unit 1903 selects the operation name node n3 of "adjustment of culture medium" corresponding to the operation executable by the "culture medium mixing device P" selected in the previous step S4301 from the execution flow abstract syntax tree t, and returns to step S4305 to repeat the above process.

[0364] On the other hand, when a negative result is obtained in step S4307, this means that in the execution process abstract syntax tree t, all the operation name nodes n3 (such as Figure 13 and Figure 14 As shown in FIG. 3 , in the operation name node n3 of “adjustment of culture medium”, the execution subject name node n4 of “culture medium mixing device P” is added. x and execution time node n x1 At this time, the extended abstract syntax tree generation unit 1903 transfers to the next step S4309.

[0365] In step S4309, the extended abstract syntax tree generation unit 1903 determines whether there are other unselected execution subjects (for example, cell culture devices X, Y or flow cytometer Q) in the execution environment information E. If not, it means that all execution subjects in the execution environment information E are used as nodes in the execution process abstract syntax tree t, and the specified extended abstract syntax tree t′ is generated, thereby ending the above-mentioned extended abstract syntax tree generation processing flow.

[0366] On the other hand, in step S4309, when there are other unselected execution subjects (for example, cell culture devices X, Y or flow cytometer Q) in the execution environment information E, the extended abstract syntax tree generation unit 1903 selects other execution subjects from the execution environment information E in the next step S4310, returns to step S4302 again, and repeats the above processing until a negative result is obtained in step S4309.

[0367] (1-4-5) Generation and processing of partially ordered sets

[0368] Next, the generation process of the above-mentioned poset is described. In this case, the poset generation unit 1905 generates the following from the extended abstract syntax tree: Figure 10B The poset generation unit 1905 is as shown in FIG. Figure 45 As shown in "step 1", the operation name nodes n3 of the extended abstract syntax tree t' are extracted, such as "adjustment of differentiation induction medium A" and "adjustment of differentiation induction medium B" that do not have an operation name node n3 in their descendant nodes.

[0369] The partially ordered set generation unit 1905 extracts the operation name node n3 of "Cultivation based on differentiation induction medium A" (abbreviated as "cell culture" in the extended abstract syntax tree t') from the operation name node n3 of the extended abstract syntax tree t', which has only the operation name node n3 of "Adjustment of differentiation induction medium A" that does not have the operation name node n3 in its descendant node as a child node, as shown in FIG. Figure 45As shown in "step 2" of the code, the operation name node n3 "Adjustment of differentiation inducing medium A" is added as a child node of the operation name node n3 "Cultivation using differentiation inducing medium A" via an edge. Furthermore, in the extended abstract syntax tree t', the operation name node n3 "Cultivation using differentiation inducing medium B" (referred to as "cell culture" in the extended abstract syntax tree t') has not only the operation name node n3 "Adjustment of differentiation inducing medium B" as a child node but also the operation name node n3 "Cultivation using differentiation inducing medium A" as a child node. Therefore, there is no operation name node that has only the operation name node n3 "Adjustment of differentiation inducing medium B" as a child node. Therefore, no other operation name node n3 is added to the operation name node n3 "Adjustment of differentiation inducing medium B."

[0370] Next, as shown in "step 3", the operation name node n3 of "Cultivation based on differentiation induction medium B" which does not have an operation name node n3 in its descendant nodes and has the operation name nodes n3 of "Cultivation based on differentiation induction medium A" and "Adjustment of differentiation induction medium B" as child nodes is extracted as the child node of the operation name node n3 of "Cultivation based on differentiation induction medium B", and the operation name nodes n3 of "Cultivation based on differentiation induction medium A" and "Adjustment of differentiation induction medium B" are added through edges.

[0371] The poset generation unit 1905 is similarly Figure 45 As shown in "step 4", based on the extended abstract syntax tree t′, the operation name node n3 of "marker gene expression evaluation" with the operation name node n3 of "culture based on differentiation induction medium B" as a child node is extracted as the child node of the operation name node n3 of "marker gene expression evaluation", and the operation name node n3 of "culture based on differentiation induction medium B" is added by the edge to obtain the partial order set as the final processing result.

[0372] (1-5) Execution instruction information generation process

[0373] Next, the above-mentioned execution instruction information generation process flow will be described. Figure 46 : is a block diagram showing the configuration of the execution instruction information generating unit 20. Figure 47 This is a flowchart showing the execution instruction information generation process. Figure 46 As shown, the execution instruction information generating unit 20 includes an operation summary column selecting unit 2001 , an operation manual generating unit 2002 , and a setting information generating unit 2003 .

[0374] In addition, in this embodiment, the operation manual and setting information described later are used as execution instruction information, but the present invention is not limited to this. For example, only the operation manual can be used as execution instruction information, or only the setting information can be used as execution instruction information.

[0375] In this case, if Figure 47 As shown, the execution instruction information generating unit 20 starts the "execution instruction information generating process flow" from the "start" step. The operation summary bar selecting unit 2001 selects the execution instruction information generating unit 2001 in step S501, for example, Figure 6 In the illustrated execution flow, a predetermined operation summary column C1 is selected, and the process proceeds to the next steps S502 and S503.

[0376] In the next step S502 , the operation manual generating unit 2002 reads the operation instruction format preliminarily associated with the operation summary column C1 selected in step S501 from the database 8 .

[0377] For example, the operation instruction format may be a command in a predetermined format using a language readable by the culture medium mixing device P, etc., so that the content of the execution process and the content of the execution plan generated based on the execution process can be indicated in a manner that the culture medium mixing device P, etc., as the execution subject, can be identified using the predetermined language. Alternatively, in the case where a human is the execution subject operating the culture medium mixing device P, etc., the operation instruction format may be an article in a predetermined format using natural language, etc., and any format may be used as long as the content of the execution process and the content of the execution plan generated based on the execution process can be indicated in a manner that is understandable to humans.

[0378] In the next step S504, the operation manual generation unit 2002 reads the time information of the execution plan in addition to the constraint information, execution parameter values, variable parameter values, etc. of the operation summary column C1 selected in step S501, and inputs the read content into the area predetermined as the operation instruction format to generate an operation manual.

[0379] The operation manual can be in any format as long as the content of the execution process and the content of the execution plan generated based on the execution process can be created in a way that humans can understand using natural language. In addition, in the case where there is no human intervention in the device such as the culture medium mixing device P as the execution subject, the operation manual can be output in natural language.

[0380] On the other hand, in step S503, the setting information generation unit 2003 generates setting information such as a program for operating the execution subject, i.e., the culture medium mixing device P, in the operation summary column C1 selected in step S501, based on the execution start date and time when the execution subject starts to execute the operation, the execution end date and time when the execution subject ends to execute the operation, the constraints of the execution process, the execution parameter values, and the variable parameter values ​​determined in the execution plan.

[0381] In step S505 , the execution instruction information generating unit 20 outputs the operation manual and setting information generated based on the operation summary column C1 selected in step S501 and the execution plan as execution instruction information.

[0382] In the next step S506, the execution instruction information generation unit 20 determines whether there is an unselected operation summary column in the execution process. If so (Yes), the operation summary column selection unit 2001 selects the operation summary column C2 that was not selected in step S501 in step S507, and returns to the above step S502 again.

[0383] In this way, the execution instruction information generating unit 20 repeatedly performs the above-mentioned process until a negative result (No) is obtained in step S506 , thereby generating execution instruction information for each of all operation summary columns C1 , C2 , C3 , C4 , and C5 in the execution flow.

[0384] (1-6) Modification

[0385] In addition, in the above (1-1) to (1-5), for example, Figure 6 As shown in the operation summary columns C1 to C4 in the execution process 1 of , the case where a culture-related process consisting of a culture medium adjustment process and a cell culture process is applied is described. However, the culture-related process may also be a culture-related process that includes at least one of the culture medium adjustment process and the cell culture process, wherein the culture medium adjustment process includes operations related to the adjustment of the culture medium (operation summary columns C1 and C3); and the cell culture process includes operations related to cell culture based on the culture medium (operation summary columns C2 and C4). That is, in the above-mentioned embodiment, for example, only one of the operation summary columns C1 to C4 may be applied as a culture-related process.

[0386] As an example, in the execution process there is only Figure 6In the case of the operation summary column C1, the culture-related process optimization device 2 generates a template execution process and an execution process, etc., which cause the culture medium mixing device P to only execute the "adjustment of the differentiation inducing culture medium A" (the operation in the operation summary column C1). In more detail, for example, when the user inputs a command for executing only the operation in the operation summary column C1 via the operation unit 10, the culture-related process optimization device 2 generates a plurality of execution processes with different variable parameter values ​​from the operation summary column C1, sends these plurality of execution processes to the culture medium mixing device P, and causes the culture medium mixing device P to repeatedly execute the "adjustment of the differentiation inducing culture medium A" according to each execution process.

[0387] Here, the variable parameter values ​​in the operation summary column C1 can be determined based on past execution results, as described above, or can be set in advance by the user via the operation unit 10. Thus, a plurality of culture media A are generated, each adjusted based on a plurality of combinations of variable parameter values. A predetermined evaluation process is then executed for each of the generated culture media A, obtaining evaluation results for each culture media A. The execution results and the evaluation results are stored in the culture-related process optimization device 2. The culture-related process optimization device 2 records the obtained execution process, variable parameter values, execution results, and evaluation results in association with each other in the database 8.

[0388] (1-7) Actions and Effects

[0389] In the above configuration, the culture-related process optimization method according to this embodiment obtains a starting point execution flow that serves as the starting point of the search, wherein the starting point execution flow specifies multiple operations to be performed sequentially in the culture-related process related to cell culture as operation items, and specifies information related to the operations (acquisition step). In the culture-related process optimization method, one or more variable parameter items capable of setting variable parameter values ​​are determined in the starting point execution flow (variable parameter item determination step). Based on past execution performance results and evaluation performance results, the variable parameter values ​​are set in the variable parameter items determined in the variable parameter item determination step, and an execution flow is generated (execution flow generation step).

[0390] Furthermore, in the cultivation-related process optimization method, an execution result of an execution subject executing according to the execution process is obtained in the execution environment (an execution result obtaining step), and an evaluation result of the execution result is obtained (an evaluation result obtaining step). In the cultivation-related process optimization method, the execution process, the variable parameter value, the execution result, and the evaluation result are recorded in association with each other (a storage step).

[0391] Thus, the cultivation-related process optimization method can be used as a clue to the execution process for searching for optimal production conditions that achieve the greatest possible gain when the execution subject actually executes the cultivation-related process in the execution environment 100 based on these execution processes, variable parameter values, the execution results, and the evaluation results. This allows the optimal production conditions (such as culture medium adjustment conditions and cell culture conditions) with the greatest possible gain to be found with the fewest possible experiments. Furthermore, because the cultivation-related process optimization method can find the optimal production conditions with the greatest possible gain with the fewest possible experiments, it can reduce the total cost and effort required to search for production conditions and find the optimal cultivation-related process with the greatest possible gain.

[0392] (2) Cultivation-related process optimization method according to the second embodiment

[0393] Next, a cultivation-related process optimization method according to a second embodiment will be described. In the cultivation-related process optimization method according to the second embodiment, when selecting a variable parameter item set in a template execution process, an item selection simulation is performed. Based on the results of the item selection simulation, a variable parameter item is selected from the starting execution process.

[0394] In addition, in the cultivation-related process optimization method according to the second embodiment, when selecting a variable parameter value from the search range set in the variable parameter item of the template execution process, a variable parameter value selection simulation is performed, and based on the result of the variable parameter value selection simulation, a variable parameter value is selected from the search range in each variable parameter item.

[0395] In the cultivation-related process optimization method according to the second embodiment, when optimizing the execution process for execution in the execution environment 100 through successive optimization, before actually executing the "execution process" in the execution environment 100 and performing successive optimization, the range of the types of variable parameter items is narrowed down in advance through project selection simulation, or the range of variable parameter values ​​is limited in advance using the simulation analysis results obtained based on the variable parameter value selection simulation, thereby reducing the number of condition studies in the execution environment 100.

[0396] When optimizing an execution process executed in a certain execution environment 100, if the number of variables that can be optimized is large, or if the variables include categorical variables (for example, the type of added reagent), a large number of experiments are required if a simple sequential optimization search is performed. Since most of the culture-related processes executed in the execution environment 100 generally require a large amount of time and money to execute a single execution process, it is unrealistic to have the execution subject execute the execution process a large number of times for optimization based on the execution process generated by the culture-related process optimization system 1.

[0397] The second embodiment of the cultivation-related process optimization method utilizes simulation analysis results to actually execute an "execution process" in execution environment 100 by the execution subject, thereby limiting the range of optimized variable parameter items and variable parameter values. By using the cultivation-related process optimization method to limit the range, the number of execution processes actually executed in execution environment 100 can be significantly reduced.

[0398] In particular, in the case of a complex execution environment 100 and a system with many hidden variables, such as a biofactory, the prediction accuracy of the system's actions based on simulation is poor (it is also difficult to align the + dimensions). Therefore, the predicted values ​​based on the optimal conditions of the simulation are often not directly consistent with the optimal production conditions in the real world. On the other hand, the relationship between the responsiveness of the objective function to the type of variable of the changing conditions is common to some extent in the simulation and the real world. If this is the case, even in the case of a complex execution environment 100, by simulation, by changing the experimental conditions globally (variable parameter items and variable parameter values), for example, it is possible to roughly classify the variables that change the value of the objective function and the variables that do not change, or it is possible to roughly grasp the response characteristics of the target variable to the explanatory variable. By referring to this information, the search range (variable parameter items and variable parameter values) can be set from the variables predicted to contribute to the value of the objective function, or by determining the shape of the regression model used, it is possible to efficiently search for variable parameter items and variable parameter values.

[0399] Here, the cultivation-related process optimization method according to the second embodiment will be described, particularly using a biofactory with a large number of hidden variables as an example. More specifically, consider the following cultivation-related process: Escherichia coli X, introduced with a plasmid expressing an enzyme P for producing a certain compound A, is cultured in a culture medium M. By inducing expression of enzyme P, compound A (hereinafter also referred to as the target substance) is produced. The following describes the project selection simulation and the variable parameter value selection simulation, respectively.

[0400] (2-1) Template execution flow generation using project selection simulation

[0401] First, the situation of generating a template execution process in the culture-related process optimization method is explained, wherein, in the template execution process, a project selection simulation is used to determine the most suitable raw materials constituting the culture medium M that maximize the yield [g / L] of the target substance A as variable parameter items.

[0402] Here, for example, when there are hundreds of candidate raw materials that constitute the culture medium M, in order to obtain a culture medium M that can maximize the yield of the target substance A during the culture-related process, it is relatively burdensome to actually make the execution subject execute the "execution process" in the execution environment 100 to determine which raw material is best to select from these hundreds of candidate raw materials, or which raw material combination is best, or how to mix the raw materials in the best way, and make a determination based on the obtained execution results and evaluation results.

[0403] Therefore, in the second embodiment, when generating a template execution process, by executing a project selection simulation, the range of raw materials constituting the culture medium M that can maximize the yield of the target substance A is narrowed down in advance from hundreds of candidate raw materials constituting the culture medium M, thereby significantly reducing the number of execution processes actually performed in the execution environment 100.

[0404] here, Figure 48 : is a block diagram showing the configuration of the template execution flow generation unit 51 according to the second embodiment. Figure 49 1 is a flow chart showing the template execution process generation process according to the second embodiment. Figure 48 As shown, the template execution process generation unit 51 includes: a starting point execution process acquisition unit 1501, an execution performance result and evaluation performance result acquisition unit 1502, a candidate project selection unit 53, a project selection simulation analysis unit 54, a variable parameter project analysis unit 55, a search range setting unit 56, a template execution process output unit 1506, and a constraint condition setting unit 1507.

[0405] like Figure 49 As shown, the training-related process optimization device 2 starts the "template execution process generation processing flow" from the "start" step. In the next step S201, the manager inputs the training-related process and evaluation process to be optimized.

[0406] In the next step S202, the starting point execution flow acquisition unit 1501 acquires the starting point execution flow that serves as the starting point for the search from the database 8 based on the culture association process and the evaluation process. In step S51, the candidate item selection unit 53 determines a plurality of candidate raw materials (also referred to as candidate variable parameter items) that constitute the culture medium M for obtaining an evaluation result that maximizes the yield of the target substance A in the evaluation process, and arbitrarily selects a specified dissolution amount as the parameter value for item selection of each candidate raw material. In addition, the number of candidate variable parameter items (candidate raw materials) is not particularly limited, and is set here to be two or more (multiple), such as 9 or 10, but may also be one. In addition, the determination of the candidate variable parameter items (candidate raw materials) can be determined by the administrator or based on the results of the item selection simulation evaluation conducted in the past, which will be described later. There is no particular limitation on the method of determination.

[0407] For example, when m candidate raw materials constituting the culture medium M are determined from the metabolic pathway network, past execution performance results and evaluation performance results related to the culture medium M are obtained from the database 8 through the execution performance result and evaluation performance result acquisition unit 1502. For example, m candidate raw materials can be determined from the raw materials of the culture medium M used in the execution performance results, etc., or m candidate raw materials can be determined from a specified area of ​​the metabolic pathway network centered on the raw materials of the culture medium M used in the execution performance results, etc.

[0408] The metabolic pathway network is data representing the pathway of chain chemical reactions occurring in cells in biochemistry and is previously stored in the database 8. The candidate item selection unit 53 obtains the metabolic pathway network from the database 8 and selects m candidate raw materials constituting the culture medium M based on the metabolic pathway network.

[0409] In step S52 , the project selection simulation analysis unit 54 performs a project selection simulation using the dissolved amounts of the m candidate raw materials determined in step S51 as input, and obtains an output result of the estimated yield of the target substance A as a project selection simulation evaluation result.

[0410] In addition, here, as an example, since the raw materials constituting the culture medium M are described, the parameter value for item selection is the dissolution amount, but in other culture-related processes, the parameter value for item selection may be, of course, concentration and mixing amount, temperature, time, etc.

[0411] Here, the project selection simulation uses, for example, a substance that has been pre-modeled, such as a biochemical reaction system operating under non-ideal conditions such as molecular mixing or localization, such as a cell simulation such as an E-cell. For example, if Escherichia coli X is cultured in a medium M composed of m selected raw materials and the expression of enzyme P is induced, it is possible to hypothetically simulate and predict the production level of target substance A.

[0412] The project selection simulation analysis unit 54, as a project selection simulation evaluation result, for example, takes the time T after the specified time T from the start of the project selection simulation as the starting time T, and based on the project selection simulation result, calculates the integral value of the production of the target substance A between the starting time T and the time (T+Δt) as the project selection simulation evaluation result.

[0413] In the next step S53, the candidate item selection unit 53 determines whether to select a new item selection parameter value (i.e., the dissolved amount of the candidate raw material) for the candidate variable parameter item (candidate raw material) and continue the item selection simulation. Whether to continue the item selection simulation can be determined, for example, by a manager. Alternatively, the candidate item selection unit 53 can determine whether a predetermined number of item selection simulations have been performed, or whether the desired item selection simulation evaluation results (here, a raw material with a significant yield variation of target substance A) have been obtained.

[0414] When the candidate item selection unit 53 determines in step S53 to continue the item selection simulation (Yes), that is, the manager determines to continue, or the candidate item selection unit 53 determines that the item selection simulation has not been performed the specified number of times, or the candidate item selection unit 53 determines that the expected item selection simulation evaluation result group has not been obtained, it returns to step S51 again, selects a new dissolution amount (parameter value for item selection) for the candidate raw material, and in the next step S52, uses the newly selected dissolution amount as input to perform the item selection simulation.

[0415] In this way, a plurality of candidate raw materials with different dissolution amounts are generated, and project selection simulation is performed for each combination of candidate raw materials with different dissolution amounts. The integrated value of the yield of the target substance A is calculated as the project selection simulation evaluation result.

[0416] On the other hand, in step S53, when the candidate item selection unit 53 determines not to continue the item selection simulation (No), that is, the manager determines not to continue, or when the candidate item selection unit 53 determines that the item selection simulation has been executed a specified number of times, or when the candidate item selection unit 53 determines that the desired item selection simulation evaluation result group has been obtained, it transfers to the next step S54.

[0417] In step S54, the variable parameter project analysis unit 55 can determine whether the candidate raw material is a raw material that has a greater impact on the production change or production increase based on the obtained project selection simulation evaluation results, according to the production change (or production increase) of the target substance A produced by changing the dissolution amount of the candidate raw material, and based on this, limit the raw materials that become variable parameter projects.

[0418] In the next step S55, the search range setting unit 56 estimates the search range of the dissolution amount that can be set for each raw material of the variable parameter item screened out in step S54, based on the past actual execution performance results and the trend of the evaluation performance results, as an influence on the production change (or production increase) of the target substance A, and sets the estimated specified search range for each raw material of the variable parameter item.

[0419] In step S211 , the template execution flow output unit 1506 sets the search range obtained by the search range setting unit 56 to the starting execution flow, generates a template execution flow based on the starting execution flow, outputs the template execution flow, and ends the above-mentioned template execution flow generation processing flow.

[0420] In addition, in the above embodiment, the type and quantity (m) of candidate raw materials are fixed, the dissolution amounts of each fixed combination of candidate raw materials are changed respectively, and the project selection simulation is repeated. However, it is of course also possible to appropriately change the type and quantity of the combined candidate raw materials and repeat the project selection simulation.

[0421] (2-2) Template execution flow generation processing using variable parameter value selection simulation

[0422] The following describes a variable parameter value selection simulation. In this case, in the culture-related process optimization method, when selecting variable parameter values ​​from the search range of the template execution process, the variable parameter value selection simulation is used to select variable parameter values ​​based on the narrowed range of the solubility (mol / L) of each of the multiple (e.g., m1) raw materials constituting the culture medium M, so as to maximize the yield (g / L) of the target substance A.

[0423] here, Figure 50 : is a block diagram showing the configuration of the variable parameter value setting unit 61 according to the second embodiment. Figure 51 Flowchart showing the flow of variable parameter value setting processing according to the second embodiment. Figure 50 As shown, the variable parameter value setting unit 61 includes a variable parameter value selection simulation analyzing unit 62 , a variable parameter value analyzing unit 1601 , and a variable parameter value selecting unit 1602 .

[0424] like Figure 51 As shown, the variable parameter value selection simulation analysis unit 62 starts the variable parameter value setting processing flow from the "Start" step. In the next step S70, the specified dissolution amount (for example, the dissolution amount L1 of raw material m1 in 1 (L), the dissolution amount L2 of raw material m2, the dissolution amount L3 of raw material m3, etc.) is randomly selected from the search range of each raw material set in the variable parameter item as a candidate variable parameter value (hereinafter also referred to as a candidate dissolution amount).

[0425] In the next step S71, the variable parameter value selection simulation analysis unit 62 takes the candidate dissolution amount of each raw material randomly selected in step S70 as input, performs a variable parameter value selection simulation, and obtains information about how the yield of target substance A is distributed when different concentrations of culture medium are input as the variable parameter value selection simulation evaluation result.

[0426] Here, in the variable parameter value selection simulation, for example, using a cell simulation such as an E-cell, a biochemical reaction system operating under non-ideal conditions such as molecular mixing or localized presence is preliminarily modeled. For example, assuming that Escherichia coli X is cultured in a culture medium M composed of m raw materials in predetermined dissolved amounts (for example, the dissolved amount of raw material m1 is L1, the dissolved amount of raw material m2 is L2, the dissolved amount of raw material m3 is L3, etc.), and the expression of enzyme P is induced, the expected yield of target substance A can be hypothetically simulated.

[0427] In the variable parameter value selection simulation analysis unit 62, as the variable parameter value selection simulation evaluation result, for example, the time when the specified time T has passed since the start of the variable parameter value selection simulation is taken as the starting time T, and based on the project selection simulation result, the integral value of the yield of the target substance A between the starting time T and the time (T+Δt) is calculated as the variable parameter value selection simulation evaluation result.

[0428] In the next step S72, the variable parameter value selection simulation analysis unit 62 determines whether to continue to reselect candidate dissolution amounts from the search range and perform the variable parameter value selection simulation. If the variable parameter value selection simulation analysis unit 62 determines in step S72 that the variable parameter value selection simulation is to be continued (Yes), that is, when the administrator determines to continue the variable parameter value selection simulation, or when the variable parameter value selection simulation analysis unit 62 determines that the variable parameter value selection simulation has not been executed a predetermined number of times, or when the variable parameter value selection simulation analysis unit 62 determines that the desired variable parameter value selection simulation evaluation result group has not been obtained, the process returns to the above-mentioned step S70, randomly selects a candidate dissolution amount from the search range, and performs the variable parameter value selection simulation again.

[0429] In this way, a plurality of combination candidates with different dissolution amounts of m raw materials are generated, and a variable parameter value selection simulation is performed for each combination candidate. For each combination candidate, an integrated value of the yield of the target substance A is calculated as a variable parameter value selection simulation evaluation result.

[0430] On the other hand, in step S72, when the variable parameter value selection simulation analysis unit 62 determines not to continue the variable parameter value selection simulation (No), that is, when the manager determines not to continue the variable parameter value selection simulation, or the variable parameter value selection simulation analysis unit 62 determines that the variable parameter value selection simulation has been executed a specified number of times, or the variable parameter value selection simulation analysis unit 62 determines that the expected variable parameter value selection simulation evaluation result group has been obtained, transfer to the next step S73.

[0431] In step S73, the variable parameter value analyzing unit 1601 obtains a distribution trend of candidate variable parameter values ​​and the variable parameter value selection simulation evaluation results, for example, based on the variable parameter value selection simulation evaluation results. Based on this distribution trend, the range of dissolution amounts is restricted from the search range of each raw material. In the following step S74, the variable parameter value selecting unit 1602 selects a variable parameter value (dissolution amount) for each raw material from the range of variable parameter values ​​restricted by the distribution trend of the variable parameter values ​​and the variable parameter value selection simulation evaluation results, generates multiple execution flows with different dissolution amounts for each raw material in the variable parameter item, and terminates the aforementioned variable parameter value setting process.

[0432] In addition, there is no particular limitation on the method of limiting the range of variable parameter values ​​based on the results of the variable parameter value selection simulation and selecting variable parameter values ​​from the limited range, but it is preferred to analyze the evaluation results of the variable parameter value selection simulation and select variable parameter values ​​that can be expected to obtain the best evaluation results.

[0433] here, Figure 52 63A indicates that, for example, in order to simplify the explanation, two variable parameter items (raw materials) are used, and the candidate variable parameter values ​​(candidate dissolution amounts) are changed for each variable parameter item, and a variable parameter value selection simulation is performed. The variable parameter value selection simulation evaluation results are used as the variable parameter value selection simulation evaluation results, and a schematic diagram of the distribution trend of the candidate variable parameter values ​​and the variable parameter value selection simulation evaluation results is obtained when the output result of the expected yield of the target substance A is estimated.

[0434] Here, Figure 52 Figure 63A shows an example in which the candidate dissolution amounts of the first raw material a1 during the variable parameter value selection simulation are plotted on the horizontal axis, and the candidate dissolution amounts of the second raw material b1 are plotted on the vertical axis, with the respective variable parameter value selection simulation evaluation results being displayed by color coded. Based on the distribution trend of these variable parameter value selection simulation evaluation results, the range of estimated optimal variable parameter values ​​is narrowed, and from this narrowed range of variable parameter values, the variable parameter values ​​(dissolution amounts) for each raw material specified in the execution process are selected.

[0435] In addition, here, the variable parameter value analysis unit 1601 can be configured as follows: for example, the candidate dissolution amount of each raw material used in the variable parameter value selection simulation is used as an explanatory variable, and the variable parameter value selection simulation evaluation result is used as a target variable to generate a regression model, and based on the analysis result of the regression model, the range of the optimal dissolution amount of the raw material is limited, and the variable parameter value is selected from the limited range.

[0436] In addition, at this time, the variable parameter value analysis unit 1601 can also be configured as: for example, in addition to using the candidate dissolution amounts of each raw material used in the variable parameter value selection simulation as explanatory variables, the past execution performance results and execution results (dissolution amounts) stored in the database 8 are also used as explanatory variables, and the evaluation performance results and evaluation results (yield) of the execution performance results and the variable parameter value selection simulation evaluation results are used as target variables to generate a regression model, and based on the analysis results of the regression model, the range of the optimal dissolution amount of the raw material is restricted, and the variable parameter value is selected from the restricted range.

[0437] In addition, although the configuration here is to execute a variable parameter value selection simulation and restrict the range of the variable parameter value (dissolution amount) of the raw material where the yield variation of the target substance A becomes large, and select the variable parameter value of the raw material based on this, the present invention is not limited to this.

[0438] As another embodiment, for example, it can also be configured as follows: using the input and output of the simulation of variable parameter value selection as training data, training a machine learning model (regression model) such as a neural network, and obtaining a trained regression model that can differentiate the output with the input. Even for a trained regression model, an analytical result that is approximate to the simulation of variable parameter value selection can be obtained in advance, and the range of the optimal variable parameter value is limited through the regression model a.

[0439] In this case, it is desirable to pre-store such a trained regression model a in the database 8. Thus, when limiting the dissolution amount of each raw material that has a large yield variation of the target substance A, the variable parameter value analyzing unit 1601 can use the trained regression model a to limit the range of the optimal dissolution amount of the raw material without performing a variable parameter value selection simulation.

[0440] When using the trained regression model a, the burden of computational processing can sometimes be reduced compared to the variable parameter value selection simulation, and analytical results that are similar to the evaluation results of the variable parameter value selection simulation can be obtained in a shorter time than the variable parameter value selection simulation, and the simulation can be used to efficiently search for the dissolution amount as a variable parameter value.

[0441] In addition, when training regression model a, not only the input and output of the simulation are selected by variable parameter values, but also the execution performance results and evaluation performance results recorded in database 8 can be used as training data to train the machine learning model (regression model a).

[0442] Alternatively, as another embodiment, the following regression model b can be used to limit the range of variable parameter values. For example, the input and output of a variable parameter value selection simulation are used as training data to generate a trained regression model a, which extracts characteristic hyperparameter values ​​as features by varying the variable parameter value (the amount of raw material dissolved).

[0443] Next, the final regression model b can be generated by using the feature quantities such as the hyperparameter values ​​extracted from the trained regression model a as a reference, using the past execution performance results (the raw materials of the culture medium M used in the past) or the evaluation performance results (the output of the target substance A at this time) as explanatory variables, and the evaluation performance results as the target variable.

[0444] In addition, the feature quantities such as the hyperparameter values ​​extracted from the trained regression model a are used as a reference to generate the final regression model b, for example, as described below.

[0445] 1. The regression model f(x / w) takes a d-dimensional vector x as input and sets k (an integer 1≤k) weight variables w.

[0446] 2. Regressing the data using the regression model in 1 above means updating i of the k weight variables w (an integer where 1 ≤ i ≤ k) so that the output of the regression model is appropriate for the data. (For example, this involves determining the weights of the regression model that approximate the input and output of a simulation using variable parameter values.)

[0447] 3. As a method for transferring the feature quantity of regression model a obtained in 2 above to another regression model b (such as transplanting the knowledge of regression model a trained to approximate the input and output of the variable parameter value selection simulation to the final regression model b used to actually select the variable parameter value), there are generally the following patterns (1) and (2). For example, when regression model a and the final regression model b are multiplied in the same function form,

[0448] (1) Substitute some or all of the k weight variables of the trained regression model a into the corresponding weight variables of the final regression model b.

[0449] (2) Extract one or more relationships or inequalities c(w) = 0 or c(w) > 0, etc. (or created by the administrator) established between the k weight variables of regression model a, so that the weight variable w of the final regression model b (during training) also satisfies the relationship.

[0450] In addition, when the regression model a and the final regression model b do not have the same functional form, the feature value of the regression model a can be transferred to another regression model b by training the final regression model b in a manner that approximates the input and output of the regression model a.

[0451] The variable parameter value analyzing unit 1601 uses the final regression model obtained in this way to limit the range of the dissolved amount of the raw material in which the variation in the yield of the target substance A becomes large.

[0452] Here, as a simpler example of generating the final regression model b with reference to the feature quantity extracted from the above-mentioned trained regression model a, the following is used Figure 52 63B, 63C, 63D, and 63E are explained.

[0453] Figure 52 Figures 63B, 63C, and 63D are schematic diagrams schematically illustrating, for example, two raw materials as variable parameter items, a search range ER1 for each variable parameter value specified in the template execution process. Furthermore, the feature quantities extracted from the trained regression model a using the variable parameter value selection simulation evaluation results are schematically illustrated as regions ER2, ER3, and ER4. In this case, by explicitly describing constraints (without unknowns or degrees of freedom), the search range ER1 can be restricted to regions ER2, ER3, and ER4, allowing these regions to be extracted as features.

[0454] For example, Figure 52 63E is a schematic diagram showing that the search range is not limited to a range that can be clearly expressed as a region as in 63D, but can also have one or more unknowns or degrees of freedom. While 63D uniquely defines the narrowed search range, 63E represents the search range itself as an unknown c. During the actual search, for example, by describing in regression model b the variable parameter values ​​used to generate the search range represented by this unknown c, the search range is effectively restricted.

[0455] (2-3) Functions and Effects

[0456] With the above configuration, in the cultivation-related process optimization method according to the second embodiment, similarly to the first embodiment, the cultivation-related process optimization process generates: an execution flow for the cultivation-related process and its evaluation process; an execution plan, which is data indicating the timing at which each execution subject should coordinately perform each operation within the execution flow; and execution instruction information instructing the execution subject in execution environment 100 to execute the corresponding operation according to the execution plan. Therefore, the cultivation-related process optimization method according to the second embodiment can also achieve the same effects as the first embodiment.

[0457] Furthermore, in the culture-related process optimization method according to the second embodiment, a parameter value (dissolution amount) for item selection is selected for a candidate variable parameter item (candidate raw material) that is estimated to be a variable parameter item capable of obtaining a predetermined evaluation result. An item selection simulation is performed through computational processing, using the parameter value for item selection of the selected candidate raw material as input and the evaluation result (change in target substance yield) as output. Based on the results of the item selection simulation, the variable parameter items are preliminarily restricted (variable parameter item determination step). Thus, in the second embodiment, since the raw material constituting the culture medium M that maximizes the yield of the target substance A can be pre-screened from among hundreds of candidate raw materials constituting the culture medium M, the number of execution processes actually performed in the execution environment 100 can be significantly reduced.

[0458] Furthermore, in the culture-related process optimization method according to the second embodiment, candidate variable parameter values ​​(candidate raw materials) estimated to yield a predetermined evaluation result are selected, and a variable parameter value selection simulation is executed through computational processing, with the selected candidate variable parameter values ​​as input and the evaluation result (change in target substance yield) as output. Based on the results of the variable parameter value selection simulation, the range of variable parameter values ​​is pre-restricted (variable parameter value determination step). Thus, in the second embodiment, the range of raw material solubility that maximizes the yield of target substance A can be pre-restricted among the raw materials constituting the culture medium M, thereby significantly reducing the number of execution processes actually performed in the execution environment 100.

[0459] Alternatively, as an example of the variable parameter value determination step, the input and output of the variable parameter value selection simulation can be used as training data to generate a trained regression model capable of differentiating the output by the input. Based on this trained regression model, the range of the variable parameter value can be constrained. This allows for obtaining analytical results that approximate the evaluation results of the variable parameter value selection simulation in a shorter time than the variable parameter value selection simulation, enabling efficient variable parameter value searches.

[0460] Furthermore, as an example of a variable parameter value determination step, the input and output of a variable parameter value selection simulation can be used as training data to generate a trained regression model for extracting feature quantities corresponding to changes in candidate variable parameter values, and a final regression model using the feature quantities extracted from the trained regression model can be generated. Based on the final regression model, the range of the variable parameter value can be limited.

[0461] As described above, in the cultivation-related process optimization method, through the project selection simulation and the variable parameter value selection simulation, the variable parameter projects of the template execution process and the variable parameter values ​​of the execution process can be limited to a certain extent, thereby significantly reducing the number of execution processes actually performed in the execution environment 100.

[0462] In particular, in the cultivation-related process of a complex system such as a biofactory where there are many hidden variables and the execution environment 100 is a complex system, the use of these item selection simulations and variable parameter value selection simulations is effective in reducing the number of searches by the execution subject in the execution environment 100 for optimal production conditions through the execution process.

[0463] In addition, in the culture-related process optimization method according to the second embodiment, the culture-related process of producing compound A by inducing the expression of enzyme P by culturing Escherichia coli X in culture medium M is described as an example, but it is of course also applicable to the cell culture-related processes according to the above embodiments.

[0464] (3) Retrieval of related execution processes

[0465] Next, the cultivation-related process optimization method according to the third embodiment is described. In the above embodiment, it is set to retrieve and obtain the execution performance results and evaluation performance results of the related execution process from the database 8 based on the terms specified in the operation item 26a in the starting execution process. In this embodiment, the retrieval is performed in a wider range. For example, it is also possible to retrieve and obtain the execution performance results and evaluation performance results of the related execution process from the database 8 based not only on the operation item 26a in the starting execution process, but also based on the terms specified in any one of the other input items 26b, output items 26c, execution parameter items 26d, and constraint items 26e.

[0466] In addition to this, it is also possible to search not only for simple term identity, but also for substances related to types or properties, specifics, components, composition, varieties, genes, homology, production conditions, etc., specified in at least the operation item 26a, input item 26b, output item 26c, execution parameter item 26d and constraint item 26e in the starting execution process, such as (i) technical treatment and processing methods such as culture methods, heating methods, cooling methods, forming methods, compression methods or screening methods; (ii) processing objects of operations such as basic culture media and raw materials; (iii) differentiation induction culture media (adjusted), target substances or molded products, and results obtained by operations; (iv) evaluation methods such as qualitative and quantitative evaluation, appearance evaluation, formability evaluation or quality evaluation, and pre-determine substances related to types or properties, specifics, components, composition, varieties, genes, homology, production conditions, etc., and based on the predetermined matters, retrieve and obtain the execution performance results and evaluation performance results of the related execution processes from the database 8.

[0467] For example, Figure 2Taking the starting execution process of "cell culture based on differentiation induction medium B" shown as an example, as the execution performance results and evaluation performance results of the associated execution processes, the following can also be retrieved and obtained from the database 8: (i) execution performance results and evaluation performance results of a culture medium whose composition or type is close to the differentiation induction medium B specified in the input item 26b of the starting execution process, a culture medium whose composition is close to the differentiation induction medium B, etc.; (ii) execution performance results and evaluation performance results whose results are close to the marker gene expression evaluation specified in the output item 26c during the evaluation process of the starting execution process; (iii) execution performance results and evaluation performance results of a culture method that is the same as the culture method specified in the execution parameter item 26d of the starting execution process, a similar culture temperature, a similar culture time, a similar evaluation index, etc.; (iv) execution performance results and evaluation performance results that are within the specified range of the time constraint or execution condition constraint specified in the constraint condition item 26e of the starting execution process.

[0468] In addition, in addition to retrieving and obtaining from the database 8 the execution performance results and evaluation performance results of the associated execution process the execution performance results and evaluation performance results that contain the same terms as those specified in the operation item 26a, input item 26b, output item 26c, execution parameter item 26d or constraint item 26e of the starting execution process, it is also possible to not only perform a search based on the identity of simple terms, but also predetermine matters related to these matters in terms of composition, variety, gene, homology, production conditions, etc., and retrieve the execution performance results and evaluation performance results of the associated execution process from the database 8 based on the predetermined matters.

[0469] (4) Cultivation-related process optimization method according to the fourth embodiment

[0470] (4-1) Overview of the Cultivation-Related Process Optimization Method According to the Fourth Embodiment

[0471] Next, a cultivation-related process optimization method according to a fourth embodiment will be described. Figure 53 1 is a block diagram showing the overall configuration of a cultivation-related process optimization system 301 that executes the cultivation-related process optimization method according to the fourth embodiment. Figure 53As shown, this cultivation-related process optimization system 301 differs from the first embodiment described above in that an optimal range search unit 303 is provided in the cultivation-related process optimization device 302, and the cultivation-related process optimization device 302 is connected to the patent information management system 101a and the information management system 101b via the network 4. Below, to avoid duplication, the description of the same configuration as the first embodiment is omitted, and the description focuses on the differences from the first embodiment.

[0472] In the cultivation-related process optimization device 302, a patent information management system 101a and an information management system 101b, such as a patent information platform (J-PlatPat (registered trademark)), a foreign patent information service (FOPISER), Espacenet (registered trademark), or PATENTSCOPE (registered trademark), which stores patent gazettes (a gazette indicating the contents of patent applications granted after examination by the JPO) and published patent gazettes (a gazette indicating the contents of patent applications before they are granted), are connected via network 4. In this case, when a prescribed patent information acquisition command is given by an administrator (user) via the operation unit 10 to acquire a prescribed patent gazette or published patent gazette, the cultivation-related process optimization device 302 transmits the patent acquisition command from the transmission / reception unit 11 to the patent information management system 101a via network 4.

[0473] Thus, the patent information management system 101a reads the prescribed patent gazette or published patent gazette corresponding to the patent information acquisition command from the database, and transmits the obtained patent gazette data as patent gazette data to the cultivation-related process optimization system 301 via the network 4. The cultivation-related process optimization system 301 receives the patent gazette data transmitted from the patent information management system 101a in response to the patent information acquisition command via the transmission / reception unit 11, and stores the data in the database 8. The cultivation-related process optimization device 302 displays the obtained patent gazette data on the display unit 9, allowing the administrator to visually recognize the contents of the patent gazette or published patent gazette based on the patent gazette data.

[0474] Furthermore, in response to an operation command from the operation unit 10, the culture-related process optimization device 302 can obtain, as needed, various technical information related to culture, such as a culture medium adjustment process indicating a method for adjusting a culture medium, cells cultured in the culture medium, and a method for adjusting the culture medium, and a cell culture process indicating a method for culturing cells, from the information management system 101b connected via the network 4. In this way, the culture-related process optimization device 302 can obtain patent publication data and other various technical information as existing conditions via the network 4.

[0475] Here, technical information other than patent publication data that can be obtained as a precondition by the training association process optimization device 302 includes not only publicly known technical information but also non-publicly known technical information. Specifically, various technical information can be used, such as academic papers, textbooks, laboratory notes, product manuals, technical information recorded on homepages, and technical information collected and created by specific individuals. Furthermore, this embodiment primarily describes the case where patent publications, i.e., patent publication data, that have been examined and granted rights by the JPO, are obtained as a precondition.

[0476] In this case, when the training-related process optimization device 302 obtains patent publication data as the existing conditions, the optimal range search unit 303 searches for an optimal range of variable parameter values ​​that can generate an execution process that avoids the existing conditions. Specifically, the optimal range search unit 303 searches for an optimal range of variable parameter values ​​that can generate an execution process outside the range of the existing conditions, which does not satisfy at least one of the multiple components included in the existing conditions, within a search range of variable parameter values ​​that is specified based on the starting execution process, past execution performance results, and evaluation performance results.

[0477] For example, in a patent gazette, the scope of rights is defined based on the content of the claims described in the patent gazette. In principle, technical matters that fully meet multiple constituent elements described in the claims constitute patent infringement.

[0478] In the cultivation-related process optimization device 302 according to the fourth embodiment, an optimal range search process is performed to determine an optimal range of variable parameter values, wherein the optimal range of variable parameter values ​​does not satisfy at least one or more of the multiple constituent elements specified in the claims of the patent gazette and can generate a template execution process outside the scope of the patent right. In the cultivation-related process, an optimal evaluation result with the largest possible gain is obtained, and an execution process that avoids existing conditions is obtained through repeated experiments. In particular, as described above, when the patent gazette is selected as the existing condition, an optimal evaluation result with the largest possible gain is obtained in the cultivation-related process, and an execution process that does not infringe the patent right can be obtained. On the other hand, when a known or unknown cultivation-related process that the manager wishes to avoid is selected as the existing condition, an optimal evaluation result with the largest possible gain is obtained in the cultivation-related process, and a new execution process that avoids known or unknown cultivation-related processes can be obtained.

[0479] (4-2) Optimal range search process

[0480] Next, the above-mentioned optimal range search process will be described. Figure 543 is a block diagram showing the configuration of the optimal range search unit 303. Figure 55 3 is a flowchart showing the optimal range search process flow executed by the cultivation association process optimization device 302. Figure 54 As shown, the optimal range search unit 303 includes a patent information analysis unit 304, an optimal range logical expression generation unit 305, and a logical expression analysis unit 306. Furthermore, the patent information analysis unit 304 includes a claim analysis unit 3041 and a constituent element analysis unit 3042, and the optimal range logical expression generation unit 305 includes an existing condition logical expression generation unit 3051, a search range logical expression generation unit 3052, and an optimal range logical expression analysis unit 3053.

[0481] In this case, if the cultivation-related process optimization device 302 obtains the existing conditions from the external system via the network 4, Figure 55 As shown, the "optimal range search process" begins at the "Start" step and proceeds to the next step S87. In step S87, the optimal range search unit 303 determines whether the acquired existing conditions are patent publication data. Whether the acquired existing conditions are patent publication data can be determined, for example, based on whether the source of the acquired existing conditions is the patent information management system 101a, or based on whether the existing conditions contain the words "patent publication" or "published patent publication." Alternatively, the administrator can input the determination result that the existing conditions are patent publication data via the operation unit 10.

[0482] If a positive result is obtained in step S87, this indicates that the acquired existing conditions are patent information data. In this case, in the next step S88, the patent information analysis unit 403 of the optimal range search unit 303 analyzes the dependency relationships of the claims to be analyzed (hereinafter referred to as existing claims) included in the patent publication data through the claim analysis unit 3041. Furthermore, as an example of patent disclosure data, a case where the content of the existing claims is applied to the patent publication after being granted rights as described below will be described.

[0483] “Existing claim 1:

[0484] A mesenchymal stem cell culture medium comprising: bone morphogenetic protein-4 (BMP4) at a concentration of less than 3 μM, vascular endothelial growth factor (VEGF) at a concentration of less than 5 μM, and / or stem cell growth factor (SCF) at a concentration of 5 μM or more and less than 10 μM.

[0485] “Current Claim 2:

[0486] The mesenchymal stem cell culture medium according to claim 1 further comprises less than 5 μM of sugars.

[0487] “Current Claim 3:

[0488] The mesenchymal stem cell culture medium according to claim 1 or 2 further comprises substance K.

[0489] “Current Claim 4:

[0490] The mesenchymal stem cell culture medium according to claim 3, wherein the concentration of the substance K is less than 5 μM.

[0491] In this case, the claim analysis unit 3041 pre-stores information such as information on the format prescribed by the JPO for patent application documents consisting of claims and descriptions, and dependency analysis information on general formats and terms used when expressing dependency relationships between existing claims in the claims. The claim analysis unit 3041 extracts claim information from patent gazette data and analyzes the dependency analysis information based on known natural language processing techniques (e.g., Reference 1: Sheremetyeva, S., Nirenburg, S., & Nirenburg, I. (1996). Generating patent claims from interactive input. In Eighth International Natural Language Generation Workshop (https: / / www.researchgate.net / publication / 2480091_Generating_Patent_Claims_From_Interactive_Input), Reference 2: Sheremetyeva, S. (2003, July). Natural language analysis of patent claims. In Proceedings of the ACL-2003 workshop on Patent corpus processing). g(pp.66-73).(https: / / aclanthology.org / W03-2008.pdf)), parses the contents of existing claims 1-4, and for each of existing claims 1-4, determines whether it is an independent claim without a dependent record, and for a dependent claim, determines to which claim it is dependent, etc. The claim parsing unit 3041 then structures the dependency relationship of existing claims 1-4 by generating a claim syntax tree, such as a tree structure and a directed acyclic graph (DAG), with the existing claim on the dependent side as the parent and the dependent existing claim as the child.

[0492] Figure 56The following shows an example of a claim syntax tree generated by the claim analysis unit 3041 based on existing claims 1 to 4. In this example, based on the description of the claims, existing claim 1 is defined as the parent, existing claim 2 is defined as the child of existing claim 1, existing claim 3 is defined as the child of existing claims 1 and 2, and existing claim 4 is defined as the child of existing claim 3.

[0493] In this way, if the dependency relationship of existing claims 1 to 4 is analyzed, then in the next step S89, the constituent element analysis unit 3042 divides the recorded content into multiple constituent elements according to existing claims 1 to 4 respectively, and, for example, based on terms such as ", (punctuation mark)", "and", "or" in the text, analyzes the correlation between the constituent elements. For example, the constituent element parsing unit 3042 predefines terms used to express the relationship between constituent elements, and uses this definition information and well-known natural language processing technology (for example, comparative document 3: Shinmori, A., Okumura, M., Marukawa, Y., & Iwayama, M. (2003, July). Patent claim processing for readability-structure analysis and term explanation. In Proceedings of the ACL-2003 workshop on Patent corpusprocessing (pp. 56-65). (https: / / aclanthology.org / W03-2007.pdf)) to parse the recorded content according to existing claims 1 to 4, and generate a constituent element syntax tree for each existing claim 1 to 4, which uses a tree structure to express the relationship between constituent elements.

[0494] here, Figure 57 This is a schematic diagram showing the structure of a component syntax tree generated by the component analysis unit 3042 based on conventional claim 1. Here, conventional claim 1 can be divided into the component "containing less than 3 μM of BMP4," the component "containing less than 5 μM of VEGF," the component "containing 5 μM or more and less than 10 μM of SCF," and the component "medium for mesenchymal stem cells," based on the use of terms such as ", (punctuation mark)" or "and / or" in the text.

[0495] In addition, according to the terms such as “, (punctuation mark)” or “and / or” in the existing claim 1, the scope of the right of the existing claim 1 is the following four forms of “medium for mesenchymal stem cells”.

[0496] (i) "Mesenchymal stem cell culture medium containing less than 3 μM BMP4"

[0497] (ii) "Mesenchymal stem cell culture medium containing less than 5 μM VEGF"

[0498] (iii) "Mesenchymal stem cell culture medium containing 5 μM or more and less than 10 μM SCF"

[0499] (iv) "A culture medium for mesenchymal stem cells comprising less than 3 μM BMP4, less than 5 μM VEGF, and 5 μM or more and less than 10 μM SCF."

[0500] Furthermore, in this specification, the term "constituent elements of a prior art claim" refers not only to elements expressed in terms such as substances, synthesis methods, sites, positions, directions, time, quantity, concentration, temperature, and pressure, as defined in the prior art claims, but also to elements expressed by terms that combine these terms to form a single meaning. For example, in the aforementioned prior art claim 1, constituent elements composed of terms include "mesenchymal stem cell culture medium," "BMP4," "less than 3 μM," "VEGF," "less than 5 μM," "SCF," "5 μM or more," and "less than 10 μM." Furthermore, constituent elements composed of terms that form a single meaning include "containing less than 3 μM BMP4," "containing less than 5 μM VEGF," "containing 5 μM or more and less than 10 μM SCF," and "containing less than 3 μM BMP4, less than 5 μM VEGF, and 5 μM or more and less than 10 μM SCF."

[0501] The constituent element parsing unit 3042 analyzes the constituent elements of the existing claim 1 and the relationships between the constituent elements based on the terms such as ", (punctuation mark)" or "and / or" in the sentence of the existing claim 1, the definition information of the terms indicating the relationship between the constituent elements, and the well-known natural language processing technology, determines the constituent elements in units of the terms as described above and in units of the article, and generates a constituent element syntax tree ( Figure 57 ).

[0502] At this time, the component analysis unit 3042 sets the elaboration node n as the starting point. 200 , as the refined node n 200 Set the target node n as the child node of 201 and attribute node n 202 , these target nodes n 201 and attribute node n 202Through the edge and refined node n 200 The component analysis unit 3042 extracts the noun "medium for mesenchymal stem cells" which is the "title of the invention" at the end of the existing claim 1 and the target node n. 201 Establish a corresponding relationship, analyze the " / " of "and / or" that divides the existing claim 1 into the above 4 forms, and connect "or" that means satisfying any one of the constraints specified by the child node with the attribute node n 202 Establish a corresponding relationship.

[0503] The constituent element analysis unit 3042 is configured as an attribute node n of "or" which represents the relationship between "and" and "or" in "and / or" in the conventional claim 1. 203 The child node of the set is an "or" node n that represents the "and / or" relationship between the constituent elements of the article unit, and only needs to satisfy any one of the constraints specified by the child node. 205 and an "and" node n representing "and" that needs to satisfy all constraints specified by its child nodes 206 , connecting them to attribute node n via edges 203 .

[0504] The component analysis unit 3042 is configured as a node n representing "OR" that satisfies any one of the constraints. 205 A plurality of constraint nodes n are set as child nodes, each representing the contents of the constituent elements divided by "or". 2051 、n 2052 、n 2053 , connecting them to node n via edges 205 In addition, the component analysis unit 3042 2051 、n 2052 、n 2053 Set the variable node n 207 and parameter node n 208 As child nodes, connect them to each constraint node n through edges. 2051 、n 2052 、n 2053 connect.

[0505] At the constraint node n according to this embodiment 2051 In the example, the label “BMP4” is connected to the variable node n via an edge. 207 Connect the parameter "less than 3μM" to the parameter node n 208 Correspondingly, the constituent element of the conventional claim 1 shown in (i) above, "containing less than 3 μM of BMP4" is defined. 2052The label "VEGF" is connected to the variable node n via an edge. 207 Connect the parameter "less than 5μM" to the parameter node n 208 Accordingly, the constituent element of the conventional claim 1 shown in (ii) above, "containing less than 5 μM of VEGF" is defined.

[0506] Furthermore, at the constraint node n 2053 The label "SCF" is connected to the variable node n via an edge 207 Connection, at parameter node n 208 Set node n to represent the logical AND of "and" 2081 . And, at node n representing the logical AND 2081 In the example, set the lower limit node n as a child node 2082 and upper limit node n 2083 , they are connected by edges. At the lower limit node n 2082 On the other hand, regarding the node n constrained by the parent node 2053 The specified "SCF" corresponds to the label "5 μM or more" of the lower limit value of the parameter defined in the existing claim 1, and is marked at the upper limit value node n 2083 On the other hand, about the node n that is also constrained by the parent node 2053 The specified “SCF” corresponds to the labeling of the upper limit value of the parameter defined in the existing claim 1 as “less than 10 μM”.

[0507] And, with node n representing “or” 205 Same as attribute node n 203 Another node n representing "and" is set as a child node 206 On the top, set the constraint node n that needs to satisfy all constraints 2061 、n 2062 、n 2063 , which are nodes n representing “and” 206 The child nodes of are connected via edges. In addition, here, the constraint node n 2061 、n 2062 、n 2063 The content of the node n represents "or" 205 The constraint node n specified on 2051 、n 2052 、n 2053 The same, so the description is omitted.

[0508] Furthermore, the constituent element analysis unit 3042 generates constituent element syntax trees for the remaining existing claims 2 to 4, similarly to existing claim 1. As described above, the constituent element analysis unit 3042 analyzes the mutual relationships between the multiple constituent elements specified in each existing claim 1 to 4 for each existing claim 1 to 4, and generates a constituent element syntax tree that represents the mutual relationship between the constituent elements of existing claims 1 to 4.

[0509] In addition, in the process of generating a constituent element syntax tree for each of the conventional claims 1 to 4, for example, the well-known technology "Patent claim processing for readability: Structure analysis and term explanation (https: / / www.researchgate.net / publication / 228569678_Patent_claim_proc essing_for_readability_Structure_analysis_and_term_explanation)" can be applied.

[0510] In addition, in this embodiment, a constituent element syntax tree is generated for each of the conventional claims 1 to 4. As an example, Figure 57 While the constituent element syntax tree of prior claim 1 has been described above, the present invention is not limited thereto. For example, the constituent element syntax trees of prior claim 2, etc., which are dependent on prior claim 1, can be integrated into the constituent element syntax tree of prior claim 1, thereby representing prior claims 1 to 4 using a single constituent element syntax tree.

[0511] Next, in step S90, the existing conditional logic expression generation unit 3051 generates existing conditional logic expressions representing the contents of each of the existing claims 1 to 4 based on the constituent element syntax trees generated for each of the existing claims 1 to 4. That is, the existing conditional logic expression generation unit 3051 uses logical symbols (e.g., (( ￢、∧、∨、→、← and etc.) to generate the existing conditional logic formula representing the syntax tree of the constituent elements. In addition, in order to simplify the description, the following description focuses on the existing claims 1 and 2, and the description of the existing claims 3 and 4 is omitted. The existing conditional logic formula of the existing claim 1 is the following formula (1), and the existing conditional logic formula of the existing claim 2 is the following formula (2) (V is a logical or (or), ∧ is a logical and (and)), wherein,

[0512] (BMP4<3μM)∨(VEGF<5μM)∨(5μM≤SCF<10μM)∨((BMP4<3μM)∧(VEGF<5μM)∧(5μM≤SCF<10μM))…(1)

[0513] Sugars <5μM…(2)

[0514] Next, the component analysis unit 3042 proceeds to step S91, where it analyzes the existing conditional logical formula generated in step S90, generating an existing conditional logical formula that converts the terms included in the existing conditional logical formula into terms that have a subordinate concept relationship (also referred to herein as an ontology). For example, in formula (2), which is the existing conditional logical formula of claim 2, a new existing conditional logical formula is generated, as shown in the following formula (3), which converts "sugar" into "glucose" and "trehalose," which have a subordinate concept relationship with "sugar."

[0515] (glucose < 5 μM) ∨ (trehalose < 5 μM)… (3)

[0516] The conversion to terms with such a subordinate concept relationship can be accomplished by using definition information that predefines the relationship between superordinate and subordinate concepts to identify substances with a subordinate concept relationship of "sugar," or by searching and identifying substances specified as a subordinate concept of "sugar" from the descriptions contained in the patent publication data being parsed. While the description herein describes the conversion of terms specified in existing claims to terms with a subordinate concept relationship, the present invention is not limited thereto; terms specified in existing claims can also be converted to terms with a superordinate concept relationship.

[0517] In addition, in step S91, the existing conditional logic formula generation unit 3051 parses each existing conditional logic formula of existing claims 1 to 4, and establishes a subordinate relationship association for the existing conditional logic formula generated for each existing claim 1 to 4 based on the subordinate relationship of existing claims 1 to 4 parsed in step S88. Figure 58 An example of structured data in which the existing conditional logical expressions of the existing claims 1 to 4 are associated and structured in a dependent relationship is shown.

[0518] In this case, the existing conditional logic formula generation unit 3051 is configured such that, for example, since the existing claim 2 is subordinate to the existing claim 1 (existing claim 1∧existing claim 2 (in Figure 582)), so that the existing conditional logical expression of the existing claim 2, which is a dependent item of the existing claim 1, is arranged adjacent to the existing conditional logical expression of the dependent existing claim 1. In addition, the existing conditional logical expression generating unit 3051 is arranged so that, for example, the existing claim 3 is respectively dependent on the existing claim 2 and the existing claim 1, which are dependent items of the existing claim 1 ("existing claim 1 ∧ existing claim 2 ∧ existing claim 3" (in Figure 58 Marked as "1∧2∧3"), existing claim 1∧existing claim 3 ( Figure 58 Marked as "1∧3")), therefore, for the existing claim 3 which is a dependent item of the existing claim 1 or 2, the existing conditional logical formula of the existing claim 3 is also configured adjacent to the existing conditional logical formulas of the dependent existing claim 1 and the existing claim 2 respectively.

[0519] Thus, the existing conditional logical expression generation unit 3051 generates structured data representing the dependency pattern of the existing claims for each row of existing claims 1 to 4. Based on the structured data, the existing conditional logical expression generation unit 3051 generates an existing conditional logical expression (hereinafter referred to as logical expression P) that establishes a dependency relationship between the individual existing conditional logical expressions of existing claims 1 to 4, as shown in the following equation (4).

[0520] Logical formula P

[0521] = the existing conditional logical formula of the existing claim 1∨

[0522] (Existing conditional logical formula of existing claim 1 ∧ Existing conditional logical formula of existing claim 2)∨

[0523] (the existing conditional logical formula of existing claim 1 ∧ the existing conditional logical formula of existing claim 2 ∧ the existing conditional logical formula of existing claim 3)∨

[0524] (the existing conditional logical formula of existing claim 1 ∧ the existing conditional logical formula of existing claim 2 ∧ the existing conditional logical formula of existing claim 3 ∧ the existing conditional logical formula of existing claim 4)∨

[0525] (Existing conditional logical formula of existing claim 1 ∧ Existing conditional logical formula of existing claim 3)∨

[0526] (The existing conditional logical formula of existing claim 1 ∧ the existing conditional logical formula of existing claim 3 ∧ the existing conditional logical formula of existing claim 4)…(4)

[0527] Since the logical formula P of all prior art claims 1 to 4 is complex, for simplicity of explanation, the following example uses the logical formula P' obtained by applying prior art claims 1 and 2. In this case, the logical formula P' is represented by the following formula (5).

[0528] Logical formula P′

[0529] = existing claim 1 ∨ (existing claim 1 ∧ existing claim 2)

[0530] =(BMP4<3μM)∨(VEGF<5μM)∨(5μM≤SCF<10μM)∨

[0531] ((BMP4<3μM)∧(VEGF<5μM)∧(5μM≤SCF<10μM))∨

[0532] ((BMP4<3μM)∨(VEGF<5μM)∨(5μM≤SCF<10μM)∨

[0533] ((BMP4<3μM)∧(VEGF<5μM)∧(5μM≤SCF<10μM)))∧((Glucose

[0534] <5μM)∨(trehalose <5μM))…(5)

[0535] Next, the process proceeds to step S95, where the search range logic expression generation unit 3052 obtains the search range of the variable parameter value as the optimization object generated by the template execution process generation unit 15 of the operation processing unit 7, and generates a search range logic expression Q representing the search range using logical symbols used in logic to represent logical expressions.

[0536] If a negative result is obtained in step S87, this indicates that the acquired existing conditions are not patent information data, that is, not patent publications or published patent publications, but rather are acquired from technical information such as books or papers. In this case, the optimal range search unit 303 proceeds to the next step S93. In step S93, the existing condition logic formula generation unit 3051 analyzes the acquired technical information, extracts the existing conditions to be analyzed from the technical information, generates an existing condition logic formula representing the existing conditions, and then proceeds to the next step S95. The method for generating the existing condition logic formula representing the existing conditions is not particularly limited. For example, the existing condition logic formula may be generated by an administrator, or the existing condition logic formula may be generated by automatically extracting existing conditions from technical information using known natural language processing techniques.

[0537] The following description focuses on the case where a patent publication with the following existing claims is obtained (a positive result is obtained in step S87). Furthermore, the following description uses the example of setting the search range of variable parameter values ​​to be optimized to be 0 μM or greater and less than 10 μM for BMP4, 0 μM or greater and less than 15 μM for SCF, and 10 μM or greater and less than 15 μM for glucose.

[0538] In this case, the search range logical expression generation unit 3052 generates the following equation (6) as the search range logical expression Q based on the search range of the variable parameter value described above.

[0539] Search range: Logical Q

[0540] =(0μM≤BMP4≤10μM)∧(0μM≤SCF≤15μM)∧

[0541] (10μM≤glucose≤15μM)…(6)

[0542] Next, the process proceeds to step S96, where the optimal range logical expression analysis unit 3053 generates an optimal range logical expression R representing an optimal range outside the range of the existing conditions within the search range of the variable parameter value based on the logical expression P′ obtained in step S91 and the search range logical expression Q obtained in step S95. In this embodiment, based on the aforementioned logical expression P′ (to simplify the explanation, the logical expression P is changed and will be used here) and the search range logical expression Q, the following equation (7) is obtained using the logical symbol "￢" representing negation and the logical symbol "∧" representing logical AND.

[0543] Optimal range logic R

[0544] =￢P′∧Q

[0545] =￢((BMP4<3μM)∨(VEGF<5μM)∨(5μM≤SCF<10μM)∨

[0546] ((BMP4<3μM)∧(VEGF<5μM)∧(5μM≤SCF<10μM))∨

[0547] (((BMP4<3μM)∨(VEGF<5μM)∨(5μM≤SCF<10μM)∨

[0548] ((BMP4<3μM)∧(VEGF<5μM)∧(5μM≤SCF<10μM)))∧((Glucose

[0549] <5μM)∨(Trehalose<5μM))))∧((0μM≤BMP4≤10)∧

[0550] (0μM≤SCF≤15μM)∧(10μM≤glucose≤15μM))…(7)

[0551] Next, in step S97, the optimal range logical expression parser 3053 converts the logical expression R obtained in step S96 into an optimal range logical expression in, for example, disjunctive normal form. The optimal range logical expression in disjunctive normal form is represented by R1∨R2∨…∨Rn (Ri=R′1∧…∧R′m), where Ri (1≤i≤n) represents a combination of real parameters not included in the right range and is a logical expression used to determine whether a parameter falls within the search range for a variable parameter value.

[0552] Here, ￢P′ can be expressed as the following formula (8).

[0553] ￢P′=(￢Ps∧￢Pp∧￢Pglt)∨(￢Ps∧￢Pp∧￢Pglt∧￢Pglc∧￢Pt)

[0554] Wherein, Ps: BMP4 < 3 μM

[0555] Pp: VEGF < 5 μM

[0556] Pglt: 5μM≤SCF<10μM

[0557] Pglc: glucose < 5 μM

[0558] Pt: Trehalose <5 μM…(8)

[0559] The search range logical formula Q can be expressed by the following formula (9).

[0560] Q=Qs∧Qglt∧Qglc

[0561] Where, Qs: 0 μM ≤ BMP4 ≤ 10 μM

[0562] Qglt: 0μM≤SCF≤15μM

[0563] Qglc:3μM≤glucose≤15μM…(9)

[0564] Formula (10) can be obtained from the above formulas (8) and (9).

[0565] ￢P′∧Q

[0566] =((￢Ps∧￢Pp∧￢Pglt)∨(￢Ps∧￢Pp∧￢Pglt∧￢Pglc∧￢Pt))

[0567] ∧(Qs∧Qglt∧Qglc)

[0568] =R1∨R2

[0569] Where,

[0570] R1=(￢Ps∧￢Pp∧￢Pglt∧Qs∧Qglt∧Qglc)

[0571] R2=(￢Ps∧￢Pp∧￢Pglt∧￢Pglc∧￢Pt∧Qs∧Qglt∧Qglc)

[0572] …(10)

[0573] In this way, in step S97, when the optimal range logical expression (R1∨…∨Rn, R1∨R2 in this embodiment) of the disjunctive normal form is transformed, it is transferred to the next step S98, and the logical expression analysis unit 306 extracts the range of variable parameter values ​​defined in the template execution process based on the optimal range logical expression of the disjunctive normal form obtained, and ends the optimal range search processing process.

[0574] At this time, in step S98, the logical expression analysis unit 306 replaces the truth value of the condition related to the parameter outside the search range of the variable parameter value (in this example, for example, Pp: VEGF < 5μM) in the optimal range logical expression of the disjunctive normal form. Furthermore, the logical expression analysis unit 306 determines the common intersection of the conditions related to the parameter range in each of R1 and R2 of the optimal range logical expression of the disjunctive normal form, and generates the optimal range logical expression by excluding the determined intersection. The following equation (11) represents the optimal range logical expression with the intersection removed from R1 of the optimal range logical expression of the disjunctive normal form, and the following equation (12) represents the optimal range logical expression with the intersection removed from R2 of the optimal range logical expression of the disjunctive normal form.

[0575] R1=(￢Ps∧￢Pp∧￢Pglt∧Qs∧Qglt∧Qglc)

[0576] →(3μM≤BMP4≤10μM)∧((0μM≤SCF≤5μM)∨(10μM≤SCF≤15μM))

[0577] ∧(3μM≤glucose≤15μM)…(11)

[0578] R2=(￢Ps∧￢Pp∧￢Pglt∧￢Pglc∧￢Pt∧Qs∧Qglt∧

[0579] Qglc)→(3μM≤BMP4≤10μM)∧((0μM≤SCF≤5μM)∨

[0580] (10μM≤SCF≤15μM))∧(5μM≤glucose≤15μM)…(12)

[0581] The logical expression analysis unit 306 uses R1 of the optimal range logical expression of the disjunctive normal form to actually replace the conditions related to the parameters outside the search range of the variable parameter value, and based on the above formula (11) with the intersection part removed, obtains the ranges of the three parameters shown in the following formula (13).

[0582] 3μM≤BMP4≤10μM,

[0583] (0μM≤SCF≤5μM)∨(10μM≤SCF≤15μM),

[0584] 3μM≤Glucose≤15μM…(13)

[0585] In addition, the logical expression analysis unit 306 uses R2 of the optimal range logical expression of the disjunctive normal form to actually replace the conditions related to the parameters outside the search range of the variable parameter value, and based on the above formula (12) with the intersection part removed, the ranges of the three parameters shown in the following formula (14) are obtained.

[0586] 3μM≤BMP4≤10μM

[0587] (0μM≤SCF≤5μM)∨(10μM≤SCF≤15μM)

[0588] 5μM≤Glucose≤15μM…(14)

[0589] The logic expression analysis unit 306 compares the conditions of the above formula (13) obtained based on the optimal range logic expression R1 of the disjunctive normal form and the conditions of the above formula (14) obtained based on the optimal range logic expression R2 of the disjunctive normal form, and extracts the final variable parameter value range defined in the template execution process. The final variable parameter value range is the parameter range included in the conditions of the above formula (13) and the conditions of the above formula (14).

[0590] In this embodiment, since the condition of the above formula (13) includes the condition of the above formula (14), the range of the parameter represented by the condition of the above formula (13) is extracted as the final variable parameter value range defined in the template execution process.

[0591] In the training association process optimization device 302, the template execution process generation unit 15 receives the range of variable parameter values ​​obtained by the optimal range search unit 303, and sets the range of variable parameter values ​​obtained by the optimal range search unit 303 as a new search range in the variable parameter items of the execution template through the template execution process generation unit 15. Specifically, Figure 5In the template execution process shown, for example, in the variable parameter item 26g in the operation summary column C1, "3μM≤BMP4 concentration≤10μM" is specified as the variable parameter value of "BMP4 concentration", and the range of the variable parameter value obtained by the optimal range search unit 303 is set as a new search range in the variable parameter item 26g.

[0592] Therefore, the variable parameter value setting unit 16, similar to the above-mentioned embodiment, can determine what variable parameter value to use to execute the execution process in the execution environment 100 within the set search range based on past execution performance results and evaluation performance results, and send multiple variable parameter values ​​selected from the search range to the execution process generation unit 17.

[0593] Furthermore, similar to the above-described embodiment, the execution flow generation unit 17 can write the variable parameter values ​​selected by the variable parameter value setting unit 16 into the variable parameter items 26g of the template execution flow, thereby generating multiple execution flows with different variable parameter values. In this way, the execution flow generation unit 17 can generate a list of multiple execution flows with different variable parameter values.

[0594] (4-3) Functions and Effects

[0595] In the above configuration, the cultivation-related process optimization device 302, as the existing condition acquisition unit, acquires the existing conditions related to the cultivation-related process via the sending / receiving unit 11 (existing condition acquisition step). Furthermore, the cultivation-related process optimization device 302 searches for an optimal range of variable parameter values ​​within a predetermined search range of variable parameter values ​​by the optimal range search unit 303, which does not satisfy at least one of the multiple components included in the existing conditions and can generate an execution flow outside the existing condition range (optimal range search step).

[0596] In this way, the cultivation-related process optimization device 302 sets the variable parameter value from the optimal range of the variable parameter value searched in the optimal range search step in the execution process generation step and generates the execution process.

[0597] As a result, in the cultivation-related process optimization device 302, in addition to achieving the same effects as the above-mentioned embodiment, it is possible to obtain a cultivation process that avoids existing conditions and has new conditions that have not existed before.

[0598] Furthermore, in the above embodiment, the case of applying an optimal range logical expression in a disjunctive normal form, which generates an optimal range logical expression in the form of a disjunction (or) of conjunctions (and), is described as the optimal range logical expression in standard form. However, the present invention is not limited to this. For example, an optimal range logical expression in a conjunctive normal form, which generates an optimal range logical expression in the form of a conjunction of disjunctions, or other various other normal forms may also be applied. Furthermore, as the result of analyzing the optimal range logical expression by the logical expression analysis unit, the case of obtaining an optimal range logical expression in standard form is described. However, the present invention is not limited to this. For example, as long as an optimal range logical expression that satisfies the condition of ￢P′∧Q can be derived, and an optimal range logical expression that represents an optimal range outside the range of the existing conditions can be obtained within the search range of the variable parameter value, various other solutions may be applied.

[0599] In addition, in the above embodiment, the sending / receiving unit 11 connected to the network 4 is described as being applied as the existing condition acquisition unit, but the present invention is not limited to this. An interface that can be connected to an external device can also be applied as the existing condition acquisition unit to obtain the existing conditions directly from the external device without going through the network 4.

[0600] Furthermore, in the above embodiment, as an example of existing conditions, a case where a patent publication after patent issuance or a published patent publication before patent issuance is used as patent publication data is described, but the present invention is not limited thereto. For example, patent publication data combining multiple patent publications after patent issuance, patent publication data combining multiple published patent publications before patent issuance, or patent publication data combining one or more patent publications after patent issuance and one or more published patent publications before patent issuance may also be used as existing conditions.

[0601] In addition, when multiple patent gazettes or published patent gazettes are used as patent gazette data as described above, logical expressions P1, P2, ..., PN are generated respectively to represent the scope of existing claims of the patent gazettes and / or published patent gazettes, and the following equation (15) is obtained which is the logical AND of the negation of these logical expressions and the search range logical expression Q. The same processing as in the above-mentioned embodiment can be performed, and the optimal range of the variable parameter value can be obtained.

[0602] (￢P1∧￢P2∧…∧￢PN)∧Q…(15)

[0603] In addition, as another embodiment, it may be a culture medium-related process optimization system that combines the configuration of the above-mentioned second embodiment and the configuration of the fourth embodiment, or a culture medium-related process optimization system that combines the configuration of the above-mentioned third embodiment and the configuration of the fourth embodiment.

[0604] Explanation of symbols

[0605] 1. 301: Cultivate a correlation process optimization system;

[0606] 2. 302: Cultivation of associated process optimization device;

[0607] 3a, 3b, 3c, 3d: communication devices;

[0608] 8: Database;

[0609] 11: Sending / receiving unit (execution result acquisition unit, evaluation result acquisition unit, current condition acquisition unit);

[0610] 15: Template execution process generation unit;

[0611] 16: variable parameter value setting unit;

[0612] 17: Execution process generation unit;

[0613] 19: Execution plan generation department;

[0614] 20: execution instruction information generating unit;

[0615] 303: Optimal range search unit;

[0616] 1501: Starting point execution process acquisition unit;

[0617] 1504: variable parameter item determination unit;

[0618] t: execution process abstract syntax tree (syntax tree);

[0619] t′: Extended Abstract Syntax Tree (syntax tree).

Claims

1. A method for optimizing a culture-related process, which is a method in a culture-related process optimization system, for optimizing a culture-related process associated with cell culture, comprising: an acquisition step, wherein a start point execution process acquisition unit acquires a start point execution process as a start point of the search, wherein the start point execution process specifies one or more operations performed in the association cultivation process and specifies content related to the operations; a variable parameter item determining step, wherein the variable parameter item determining unit determines one or more variable parameter items capable of setting variable parameter values ​​in the starting point execution process; a template execution flow generation step, wherein the template execution flow generation unit estimates a search range and generates a template execution flow in which the estimated search range is set, wherein the search range indicates a range of the variable parameter values ​​that can be set in the variable parameter item; an execution flow generating step of selecting the variable parameter value from the search range by an execution flow generating unit, setting the selected variable parameter value to the variable parameter item determined in the variable parameter item determining step, and generating an execution flow; an execution result obtaining step, wherein the execution result obtaining unit obtains the execution result when the execution subject actually executes the execution process in the execution environment; an evaluation result obtaining step, in which an evaluation result obtaining unit obtains an evaluation result of the execution result; and a storage step of recording the execution process, the variable parameter value, the execution result, and the evaluation result in a database correspondingly; In the variable parameter item determination step, the execution performance results and evaluation performance results of the execution subject in the execution environment when executing according to the same or related past execution process as the starting point execution process are compared to determine one or more variable parameter items for which the variable parameter value can be set in the starting point execution process. In the template execution process generation step, the execution performance result and the evaluation performance result when the execution subject in the execution environment executes according to the past execution process that is the same as or related to the starting point execution process are compared, the range of the variable parameter value that will affect the evaluation performance result is estimated, and the search range is set. In the execution flow generation step, the variable parameter value is determined based on the execution performance result when the execution subject in the execution environment executes according to the past execution flow that is the same as or related to the starting execution flow, The evaluation results include at least one of the cost, yield, quality, and execution time involved in the execution process of the cultivation association process; the deviation of cost, yield, quality, and execution time; and the deviation between cost, yield, quality, and execution time and their given target values. The culture-related process includes a culture-related process related to regenerative medicine for producing cells, or a culture-related process for producing biopharmaceuticals or compounds using cells.

2. The method for optimizing the cultivation association process according to claim 1, wherein: In the cultivation association process, the operation includes at least any one of the following: Operations related to culture medium adjustment; Operations related to cell culture based on the culture medium.

3. The method for optimizing the cultivation association process according to claim 1, wherein: The culture-related process is a medium adjustment process for adjusting a culture medium and / or a cell culture process for culturing cells.

4. The method for optimizing the cultivation association process according to claim 1, wherein: In the execution flow generation step, a regression model is generated based on the execution performance result, and the variable parameter value set in the variable parameter item is determined using the regression model.

5. The cultivation association process optimization method according to claim 1, wherein: Also includes: In the execution plan generating step, the execution plan generating unit generates an execution plan, wherein the execution plan indicates how the corresponding execution subjects coordinately and chronologically execute the plurality of operations specified in the execution flow in the execution environment.

6. The method for optimizing the cultivation association process according to claim 1, wherein: In the execution flow generation step, the constraint conditions when the execution subject executes the execution flow in the execution environment are set in the constraint condition items of the execution flow.

7. The method for optimizing the cultivation association process according to claim 1, wherein: In the variable parameter item determination step, an item selection simulation is performed through calculation processing and the variable parameter item is selected based on the result of the item selection simulation, wherein, in the item selection simulation, a parameter value for item selection is selected from candidate variable parameter items that can become the variable parameter items that can be inferred to be able to obtain a specified evaluation result, the selected parameter value for item selection is used as input, and the evaluation result is used as output.

8. The method for optimizing the cultivation association process according to claim 1, wherein: In the execution process generation step, a variable parameter value selection simulation is performed through calculation processing, and the range of the variable parameter value is limited based on the result of the variable parameter value selection simulation, wherein, in the variable parameter value selection simulation, a candidate variable parameter value that can become the variable parameter value that is inferred to be able to obtain a specified evaluation result is selected, the selected candidate variable parameter value is used as input, and the evaluation result is used as output.

9. The method for optimizing the cultivation association process according to claim 1, wherein: Before the execution process generation step, the process includes: an existing condition acquisition step of acquiring, by an existing condition acquisition unit, existing conditions related to the cultivation-related process; and an optimal range search step, wherein the optimal range search unit searches, within a predetermined search range of the variable parameter value, for an optimal range of the variable parameter value that does not satisfy at least one of the plurality of component elements included in the existing condition and that can generate the execution flow outside the range of the existing condition; In the execution flow generation step, the variable parameter value is set from the optimal range of the variable parameter value searched in the optimal range search step to generate the execution flow.

10. The method for optimizing the cultivation association process according to claim 9, wherein: The optimal range searching step comprises: an optimal range logical expression generating step, wherein an optimal range logical expression generating unit generates an optimal range logical expression that logically expresses an optimal range of the variable parameter value based on an existing condition logical expression that logically expresses the existing condition and a search range logical expression that logically expresses a preset search range of the variable parameter value; and In the analyzing step, a logic formula analyzing unit analyzes the optimal range logic formula to determine, for each variable parameter item, an optimal range of the variable parameter value that can generate a different execution flow outside the existing condition range.

11. The method for optimizing the cultivation-related process according to claim 9 or 10, wherein: The optimal range search step uses patent announcements or published patent gazettes as the existing conditions, In the existing condition acquisition step, The existing claims recorded in the patent publication or the published patent gazette are obtained as the existing conditions. The optimal range searching step comprises: The patent information analysis step analyzes the dependency relationship of multiple existing claims by the patent information analysis unit, and analyzes the mutual relationship between multiple constituent elements respectively specified in each existing claim.

12. A culture-related process optimization system for optimizing a culture-related process associated with cell culture, comprising: a starting point execution process acquisition unit configured to acquire a starting point execution process serving as a starting point for the search, wherein the starting point execution process specifies one or more operations to be performed in the association cultivation process and specifies contents related to the operations; a variable parameter item determining unit for determining one or more variable parameter items capable of setting variable parameter values ​​in the starting point execution process; a template execution flow generation unit configured to estimate a search range and generate a template execution flow in which the estimated search range is set, wherein the search range indicates a range of the variable parameter values ​​that can be set in the variable parameter item; an execution flow generating unit that selects the variable parameter value from the search range, sets the selected variable parameter value to the variable parameter item determined by the variable parameter item determining unit, and generates an execution flow; An execution result acquisition unit, which acquires an execution result when an execution subject executes according to the execution process in the execution environment; an evaluation result acquisition unit that acquires an evaluation result of the execution result; and The database records the execution process, the variable parameter value, the execution result and the evaluation result in correspondence, The variable parameter item determination unit compares the execution performance results and evaluation performance results of the execution subject in the execution environment when executing the same or related past execution process as the starting point execution process, and determines one or more variable parameter items for which the variable parameter value can be set in the starting point execution process. The template execution process generation unit compares the execution performance result and the evaluation performance result when the execution subject in the execution environment executes according to the past execution process that is the same as or related to the starting point execution process, estimates the range of the variable parameter value that will affect the evaluation performance result, and sets the search range. The execution flow generation unit determines the variable parameter value based on the execution performance result when the execution subject in the execution environment executes the same or related past execution flow as the starting point execution flow. The evaluation results include at least one of the following: cost, yield, quality, and execution time involved in the execution process of the cultivation association process; deviations from cost, yield, quality, and execution time; and deviations between cost, yield, quality, and execution time and their given target values. The culture-related process includes a culture-related process related to regenerative medicine for producing cells, or a culture-related process for producing biopharmaceuticals or compounds using cells.

Images