Experiment assistance system, experiment assistance method and experiment assistance program
The experiment support system addresses the challenge of incomplete experimental data by updating computational models with confidence intervals and tentative values, enhancing efficiency and accuracy in composition search.
Patent Information
- Application Number
- JP2024040995
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2024-03-15
- Publication Date
- 2025-09-29
AI Technical Summary
Existing systems struggle to efficiently search for compositions when characteristic values cannot be obtained during experiments, especially when updating calculation models.
An experiment support system that updates a computational model using regression analysis with confidence intervals, determines new parameter sets, and sets tentative values when actual values are not obtained, increasing sample data to efficiently search for compositions.
Enables efficient composition search by reducing the number of experiments and costs, while improving the accuracy of the computational model through tentative value introduction and increased sample data.
Smart Images
Figure 2025141184000001_ABST
Abstract
Description
[Technical Field]
[0001] One aspect of the present disclosure relates to an experiment support system, an experiment support method, and an experiment support program. [Background technology]
[0002] A computer system for supporting experiments on compositions is known. For example, Patent Document 1 describes a system including a computing device configured to perform a method including determining a plurality of candidate predictive models based on experimental data and computationally derived data related to one or more monoclonal antibodies (mAbs) and determining an optimal predictive model from the plurality of candidate predictive models, and a user device configured to display an output of the predictive model. [Prior art documents] [Patent documents]
[0003] [Patent Document 1] Special Publication No. 2023-548364 Summary of the Invention [Problem to be solved by the invention]
[0004] When experiments are repeated while updating a calculation model that shows the relationship between parameters that indicate the conditions for generating a composition and the characteristic values of the composition, it is sometimes the case that the characteristic values cannot actually be obtained for some parameters. A mechanism for efficiently searching for compositions even when such a situation occurs is desired. [Means for solving the problem]
[0005] An experiment support system according to one aspect of the present disclosure includes at least one processor, and the at least one processor repeatedly updates a computational model that indicates the relationship between a parameter set, which is a set of one or more parameters that indicate the conditions for producing a target composition, and a characteristic value of the target composition obtained by an experiment based on the parameter set, together with a confidence interval, by regression analysis using sample data containing one or more pairs of parameter sets and characteristic values, while increasing the number of such pairs in the sample data. Each time the computational model is updated, a new parameter set is determined based on the updated computational model. If a new characteristic value of the target composition is not obtained by an experiment based on the new parameter set, a tentative characteristic value is set based on the confidence interval corresponding to the new parameter set, and the computational model is updated using sample data that further includes a new pair of the new parameter set and the tentative characteristic value.
[0006] The computational model obtained by regression analysis mathematically shows the relationship between a parameter set and a property value. However, in reality, even if an experiment is performed based on the parameter set, there may be cases where the property value cannot be obtained. In the above aspect, a new parameter set for the experiment can be determined by setting a tentative property value using the confidence interval of the computational model, increasing the amount of sample data, and then updating the computational model. As a result, compositions can be efficiently searched for. [Effects of the Invention]
[0007] According to one aspect of the present disclosure, compositions can be efficiently searched for. [Brief explanation of the drawings]
[0008] [Figure 1] FIG. 1 is a diagram illustrating an example of a functional configuration of an experiment support system. [Figure 2] 10 is a flowchart illustrating an example of processing by the experiment support system. [Figure 3] 10 is a flowchart showing a first example of updating a computational model. [Figure 4]10 is a flowchart showing a second example of updating a computational model. [Figure 5] FIG. 10 is a diagram showing a scene of updating a calculation model in the first example. [Figure 6] FIG. 10 is a diagram showing a scene of updating a calculation model in the second example. DETAILED DESCRIPTION OF THE INVENTION
[0009] Various examples of the present disclosure will be described in detail below with reference to the accompanying drawings. In the description of the drawings, the same or equivalent elements are designated by the same reference numerals, and redundant description will be omitted.
[0010] [System Configuration] The experimental support system according to the present disclosure is a computer system that supports experiments to generate a target composition. In the present disclosure, the target composition refers to a composition that an experimenter attempts to generate in order to obtain a desired composition. Ultimately, a certain target composition may be adopted as the desired composition. The experimental support system determines a parameter set to support the experiment. A parameter set refers to a collection of one or more parameters that indicate the conditions for generating the target composition. Individual parameters may indicate attributes of raw materials used to generate the composition or may indicate the conditions of a process performed in the experiment. The experimental support system automatically generates and provides a parameter set that is expected to generate a target composition that the experimenter may desire. It is expected that using this parameter set will enable efficient composition searching.
[0011] Typically, a target composition is produced through an experiment based on a parameter set, and a property value that indicates a characteristic of the target composition is obtained. The property value is a value that indicates a chemical or physical property of the target composition. However, an experiment based on a certain parameter set may not be able to obtain a property value of the target composition for reasons such as the inability to produce the target composition or the inability to measure the property value.
[0012] The experiment support system generates a new parameter set based on sample data containing one or more pairs of parameter sets and characteristic values. The experiment support system performs regression analysis using the sample data to generate a computational model that indicates the relationship between the parameter set and the characteristic values, and generates a new parameter set based on the computational model. The new parameter set is a parameter set that is not indicated by any of the multiple pairs used to generate the computational model. Regression analysis is a process for determining the relationship between input and output. The computational model obtained by regression analysis serves to estimate the relationship between the parameter set and a specific value. In one example, the computational model includes a function that indicates the relationship between the parameter set, which is an input value, and the characteristic value, which is an output value, and a confidence interval that indicates how likely the relationship is. In other words, the computational model indicates the relationship between the parameter set and the characteristic value together with the confidence interval.
[0013] The experimenter conducts an experiment based on the new parameter set. If a new characteristic value is obtained from this experiment, a new pair of the new parameter set and the new characteristic value is added to the sample data. That is, the sample data is updated to include the new pair. The experiment support system updates the computational model based on the updated sample data, and generates the next new parameter set based on the updated computational model. The experiment support system repeats the process of obtaining a new pair, updating the computational model, and generating a new parameter set.
[0014] As described above, there is a possibility that characteristic values cannot be obtained in an experiment based on a certain parameter set while an experiment is being repeated. In this case, the experiment support system sets provisional characteristic values corresponding to the parameter set. The experiment support system then updates the computational model based on sample data to which a new pair of the parameter set and provisional characteristic value has been added. The experiment support system generates the next new parameter set based on the updated computational model. Even if actual characteristic values cannot be obtained, the experiment system updates the computational model and provides a new parameter set, allowing the experimenter to conduct an experiment based on the new parameter set.
[0015] In this way, the experimenter can use the experiment support system to repeat the experiment regardless of whether actual property values are obtained, thereby enabling efficient composition search.
[0016] An experiment support system is composed of one or more computers. When multiple computers are used, these computers are connected via a communication network such as the Internet or an intranet to logically construct a single experiment support system.
[0017] The computer that constitutes the experiment support system generally comprises a processor, memory, and a communication interface as hardware devices. The processor is, for example, a CPU, and the memory is composed of flash memory, a hard disk, etc. Each function of the experiment support system is realized by the processor executing a program stored in the memory. The computer may further comprise input devices such as a keyboard and a mouse, and output devices such as a monitor and speakers.
[0018] The experiment support program for causing a computer to function as an experiment support system includes program code for implementing each functional module of the experiment support system. This experiment support program may be provided by being non-temporarily recorded on a tangible recording medium, such as a CD-ROM, DVD-ROM, or semiconductor memory. Alternatively, the experiment support program may be provided via a communications network as a data signal superimposed on a carrier wave. The provided experiment support program is recorded in memory, for example.
[0019] 1 is a diagram showing the functional configuration of an example experiment support system 10. In this example, the experiment support system 10 is connected to a user terminal 20 via a communication network such as the Internet or an intranet.
[0020] The experiment support system 10 includes a processor 101 and a memory 102. In one example, the processor 101 functions as an acquisition unit 11, an update unit 12, a determination unit 13, and a selection unit 14. The acquisition unit 11 is a functional module that acquires experimental data for one experiment. The acquisition unit 11 stores the experimental data in the memory 102. The update unit 12 is a functional module that updates a calculation model by regression analysis using sample data stored in the memory 102. The update unit 12 stores the calculation model in the memory 102. The determination unit 13 is a functional module that determines a new parameter set based on the updated calculation model. The determination unit 13 adds the new parameter set to the sample data in the memory 102 and presents the new parameter set to the user. The selection unit 14 is a functional module that presents a parameter set corresponding to the optimal characteristic value of the sample data to the user as an optimal solution.
[0021] The user terminal 20 is a computer used by a user (for example, an experimenter) of the experiment support system 10. The user terminal 20 may be any of a variety of computers, such as a personal computer, a workstation, a tablet terminal, a smartphone, or a wearable terminal.
[0022] [System Operation] An example of processing by the experiment support system 10 will be described with reference to Figure 2, along with an example of an experiment support method according to the present disclosure. Figure 2 is a flowchart showing this example as process flow S1. Process flow S1 is based on the premise that memory 102 stores experimental data including one or more actual characteristic values obtained through one or more experiments, sample data based on the experimental data, and a current computational model generated by regression analysis using the sample data.
[0023] In step S11, the acquisition unit 11 acquires experimental data for one experiment. The user operates the user terminal 20 to input the results of an experiment based on a new parameter set as experimental data. The experimental data includes at least the new parameter set and further includes either a new characteristic value of the target composition obtained by an experiment based on the parameter set or flag information indicating that no new characteristic value was obtained by the experiment. The user terminal 20 may transmit the experimental data to the experiment support system 10, and the acquisition unit 11 may receive the experimental data. Alternatively, the user terminal 20 may store the experimental data in a predetermined storage device and notify the experiment support system 10, and the acquisition unit 11 may read the experimental data from the storage device in response to the notification. In either case, the acquisition unit 11 stores the experimental data for one experiment in the memory 102.
[0024] In step S12, the update unit 12 updates the computational model using the experimental data and existing sample data. Several examples of the details of this update are shown with reference to Figures 3 and 4. Figure 3 is a flowchart showing a first example of updating the computational model as step S12A. Figure 4 is a flowchart showing a second example of updating as step S12B. Steps S12A and S12B are both examples of step S12.
[0025] A first example will be described. In step S1201, the update unit 12 determines whether the experimental data includes a new characteristic value. That is, the update unit 12 determines whether a new characteristic value of the target composition has actually been obtained through the experiment. If the experimental data includes a new characteristic value (YES in step S1201), the process proceeds to step S1202.
[0026] In step S1202, the update unit 12 adds a new pair of a new parameter set and a new characteristic value based on the experimental data to the sample data in the memory 102. Through this process, the sample data is updated.
[0027] On the other hand, if the experimental data does not include a new characteristic value, that is, if the experimental data includes flag information (NO in step S1201), the process proceeds to step S1203.
[0028] In step S1203, the update unit 12 sets a provisional characteristic value based on the current calculation model. The update unit 12 identifies a confidence interval corresponding to the new parameter set from the current calculation model, and sets a provisional characteristic value based on the confidence interval.
[0029] In step S1204, the update unit 12 adds a new pair of a new parameter set and a temporary characteristic value to the sample data in the memory 102. Through this process, the sample data is updated.
[0030] After step S1202 or step S1204, the process proceeds to step S1205. In step S1205, the update unit 12 updates the computational model by regression analysis using the updated sample data. If the sample data is updated in step S1204, the update unit 12 updates the computational model using sample data that further includes new pairs of a new parameter set and a tentative characteristic value. The update unit 12 may use Gaussian process regression, kernel density estimation, or a deep neural network as a regression analysis algorithm to generate a computational model including a function that indicates the relationship between the parameter set and the characteristic value. A trained model generated by a deep neural network is an example of a function. When using any of Gaussian process regression, kernel density estimation, and a deep neural network, the update unit 12 may generate a computational model including a confidence interval. For example, the update unit 12 calculates or sets the confidence interval based on the variance of the generated function.
[0031] Next, a second example will be described. In step S1211, the update unit 12 determines whether or not the experimental data includes a new characteristic value. This process is the same as step S1201. If the experimental data includes a new characteristic value (YES in step S1211), the process proceeds to step S1212. If the experimental data does not include a new characteristic value, that is, if the experimental data includes flag information (NO in step S1211), the process skips step S1212 and proceeds to step S1213.
[0032] In step S1212, the update unit 12 adds a new pair of a new parameter set and a new characteristic value based on the experimental data to the sample data in the memory 102. This process is the same as step S1202.
[0033] In step S1213, the update unit 12 determines whether there is a parameter set for which actual characteristic values have not been obtained through experiments. The parameter set for which actual characteristic values have not been obtained may be one or more parameter sets stored in the memory 102 and corresponding to tentative characteristic values, or may be a parameter set acquired together with flag information in step S11.
[0034] If there is a parameter set for which actual characteristic values have not been obtained (YES in step S1213), the process proceeds to step S1214. If there is no such parameter set (NO in step S1213), the process skips steps S1214 to S1216 and proceeds to step S1217.
[0035] In step S1214, the update unit 12 identifies one or more pairs containing actual property values as an experimental sample group from the sample data in the memory 102. That is, the experimental sample group corresponds to one or more experiments in which a target composition was produced and the property values of the target composition were actually obtained.
[0036] In step S1215, the update unit 12 provisionally updates the computational model by regression analysis using the experimental sample group without using one or more pairs including the provisional characteristic values. As in step S1205, various algorithms can be used for the regression analysis in this process. Because the provisional characteristic values are not used in the regression analysis here, in the provisionally updated computational model, confidence intervals are again set for each of one or more parameter sets corresponding to one or more provisional characteristic values.
[0037] In step S1216, the update unit 12 sets provisional property values based on the provisionally updated computational model to set a non-experimental sample group. The update unit 12 sets provisional property values for each of one or more parameter sets for which no actual property value was obtained, as follows: That is, the update unit 12 sets provisional property values based on the confidence interval corresponding to the parameter set in the provisionally updated computational model. For parameter sets associated with provisional property values as part of the sample data, the provisional property values may be updated by the setting. For parameter sets acquired with flag information, provisional property values are set for the first time. The update unit 12 generates pairs of parameter sets and the set provisional property values. The update unit 12 sets the generated one or more pairs as a non-experimental sample group. The non-experimental sample group corresponds to one or more experiments for which a property value of the target composition was not obtained.
[0038] The update unit 12 identifies the set of the identified experimental sample group and the set non-experimental sample group as the latest sample data, and the sample data is updated by this process.
[0039] In step S1217, the update unit 12 updates the computational model by regression analysis using the updated sample data (i.e., the experimental sample group and the non-experimental sample group). This process is also an example of updating the computational model using sample data that further includes new pairs of new parameter sets and tentative characteristic values. The update unit 12 performs regression analysis using the algorithm used in step S1215.
[0040] Returning to FIG. 2 , in step S13, the determination unit 13 determines a new parameter set based on the updated computational model. In one example, the determination unit 13 determines the new parameter set by an optimization method based on the computational model. The determination unit 13 may generate the new parameter set so that a characteristic value predicted based on the computational model is closer to a target value of the optimization method than the characteristic value used to generate the computational model, i.e., the multiple characteristic values indicated by the updated sample data. The target value refers to an optimal value in an optimization problem addressed by the optimization method. Examples of target values include a minimum value in a minimization problem and a maximum value in a maximization problem. Alternatively, the determination unit 13 may generate the new parameter set so that the confidence interval in at least a part of the relationship between the parameter set and the characteristic value is reduced.
[0041] In one example, the experiment support system updates the computational model and generates a new parameter set (i.e., steps S12 and S13) using Bayesian optimization. In this case, the update unit 12 uses Gaussian process regression to estimate a function that indicates the relationship between the parameter set and the characteristic value, and calculates the confidence interval of the function. The determination unit 13 calculates a given acquisition function based on the result of the Gaussian process regression, and generates a parameter set that maximizes the acquisition function as a new parameter set.
[0042] In step S14, the decision unit 13 presents the new parameter set to the user. In one example, the decision unit 13 transmits the parameter set to the user terminal 20. The user terminal 20 receives and displays the parameter set.
[0043] As shown in step S15, a series of processes from acquiring experimental data to presenting a new parameter set can be repeatedly executed. If the process is repeated (NO in step S15), the experimenter conducts an experiment based on a new parameter set. As a result of the experiment, the experimenter identifies a new characteristic value or confirms that no characteristic value was obtained. In repeated step S11, the acquisition unit 11 acquires new experimental data related to the experiment. In repeated step S12, the update unit 12 updates the computational model again using the experimental data and existing sample data. In repeated step S13, the determination unit 13 determines a new parameter set based on the updated computational model. In step S14, the determination unit 13 presents the parameter set to the user. In this way, the experiment support system 10 repeatedly updates the computational model by regression analysis using sample data containing one or more pairs of parameter sets and characteristic values while increasing the number of such pairs in the sample data. Each time the computational model is updated, the experiment support system 10 determines a new parameter set based on the updated computational model.
[0044] If the presentation of a new parameter set is terminated due to the completion of a series of experiments (YES in step S15), the process proceeds to step S16. In step S16, the selection unit 14 presents the optimal solution to the user. For example, the selection unit 14 refers to the sample data in the memory 102 to identify actual characteristic values that match or are closest to the target value of the optimization problem, and selects a parameter set corresponding to this characteristic value from the sample data. The selection unit 14 presents the parameter set to the user as the optimal solution. The selection unit 14 may present a pair of the selected parameter set and the corresponding characteristic value to the user as the optimal solution. In one example, the determination unit 13 transmits the optimal solution to the user terminal 20. The user terminal 20 receives and displays the optimal solution. The experimenter or user can refer to the optimal solution to perform subsequent tasks such as additional experiments, verification, product design or development, etc.
[0045] The updating of the computational model will be further described with reference to Figures 5 and 6. Figure 5 is a diagram showing one scene of updating the computational model in the first example. Figure 6 is a diagram showing one scene of updating the computational model in the second example. In both Figures 5 and 6, it is assumed that the regression analysis is Gaussian process regression and the optimization method is Bayesian optimization.
[0046] 5 shows the update of the computational model 200 using three states ST11, ST12, and ST13. The computational model 200 includes a function f that indicates the relationship between the parameter set x and the characteristic value y, a confidence interval, and an acquisition function. In FIG. 5, the function f, the confidence interval, and the acquisition function are represented by a curve 210, a region 220, and a curve 230, respectively.
[0047] The update unit 12 updates the computational model 200 by Gaussian process regression using sample data consisting of four pairs 201 to 204 of parameter sets and characteristic values (step S1205). A state ST11 indicates the computational model 200. The determination unit 13 determines a new parameter set x that maximizes the acquisition function. new is generated (step S13).
[0048] Parameter set x new Assume that no new characteristic value is obtained by the experiment based on the parameter set x (NO in step S1201). In this case, the update unit 12 new The update unit 12 sets a provisional characteristic value based on the confidence interval 221 corresponding to the parameter set x (step S1203). The state ST12 indicates the setting. The update unit 12 may set the provisional characteristic value based on the confidence interval 221 and a target value of the Bayesian optimization. In this example, the target value is assumed to be the minimum value in the minimization problem. The update unit 12 sets the value in the confidence interval 221 that is farthest from the minimum value 240 as the provisional characteristic value, and updates the parameter set x new and the temporary characteristic value are added to the sample data (step S1204). The minimum value 240 is the limit value (lower limit value) corresponding to the target value in the minimization problem. The setting of such temporary characteristic values is performed by the parameter set x newThis can be said to be a process that was carried out in response to the fact that the above did not contribute to solving the minimization problem, i.e., to searching for the target value.
[0049] The update unit 12 updates the computational model 200 by Gaussian process regression using sample data consisting of five pairs 201 to 205 (step S1205). A state ST13 indicates the computational model 200. The determination unit 13 determines the next new parameter set x that maximizes the acquisition function. new2 is generated (step S13).
[0050] 6 shows the update of the computational model 300 using four states ST21, ST22, ST23, and ST24. The computational model 300 includes a function f that indicates the relationship between the parameter set x and the characteristic value y, a confidence interval, and an acquisition function. In FIG. 6, the function f, the confidence interval, and the acquisition function are represented by a curve 310, a region 320, and a curve 330, respectively.
[0051] The update unit 12 updates the computational model 300 by Gaussian process regression using sample data consisting of five pairs 301 to 305 of parameter sets and characteristic values (S1217). The pairs 301 to 304 contain actual characteristic values, and the pair 305 contains the parameter set x tmp and the tentative characteristic values. A state ST21 indicates the calculation model 300. The determination unit 13 determines a new parameter set x new is generated (step S13).
[0052] Parameter set x new Assume that new characteristic values are obtained by an experiment based on (YES in step S1211). In this case, the update unit 12 updates the parameter set x new A new pair 306 of the parameter set X and the new characteristic value is added to the sample data (step S1212). tmpSince there is a parameter set x (YES in step S1213), the update unit 12 specifies the pairs 301 to 304, 306 as an experimental sample group (step S1214). The update unit 12 provisionally updates the computational model by regression analysis using the experimental sample group without using the pair 305 (step S1215). A state ST22 shows the provisionally updated computational model 300. The update unit 12 specifies the parameter set x tmp A provisional characteristic value is set based on the confidence interval 321 corresponding to the parameter set x (step S1216). In this example, the target value is assumed to be the minimum value in the minimization problem. The update unit 12 sets the value farthest from the minimum value 340 in the confidence interval 321 as the provisional characteristic value. The minimum value 340 is the limit value (lower limit value) corresponding to the target value in the minimization problem. State ST23 indicates the setting. Such provisional characteristic value setting is performed by the parameter set x tmp This can be said to be a process following the fact that the parameter set x tmp The pair 305' with the set tentative characteristic value is set as a non-experimental sample group (step S1216). The pair 305' can be said to be the updated pair 305. The update unit 12 identifies the set of the identified experimental sample group and the set test sample group as the latest sample data.
[0053] The update unit 12 updates the computational model 300 by Gaussian process regression using sample data consisting of six pairs 301 to 304, 305′, and 306 (step S1217). A state ST24 indicates the computational model 300. The determination unit 13 determines the next new parameter set x that maximizes the acquisition function. new2 is generated (step S13).
[0054] [Variations] The technology according to the present disclosure has been described in detail above based on various examples. However, the present disclosure is not limited to the above examples. The technology according to the present disclosure can be modified in various ways without departing from the spirit of the present disclosure.
[0055] The experiment support system according to the present disclosure does not need to include a selection unit, that is, a computer system different from the experiment support system may present the optimal solution to the user.
[0056] In the above example, the experiment support system is configured as a server in a client-server system. As another example, the experiment support system may be implemented in a stand-alone computer. Alternatively, the experiment support system may be implemented in a user terminal that can access a predetermined database that stores experimental data, sample data, or computational models via a communication network.
[0057] The processing steps of the method executed by at least one processor are not limited to the above examples. For example, some of the above steps may be omitted, or the steps may be executed in a different order. Furthermore, any two or more of the above steps may be combined, or some of the steps may be modified or deleted. Alternatively, other steps may be executed in addition to the above steps.
[0058] In the present disclosure, when comparing the magnitude of two numerical values, either of the two criteria "greater than or equal to" and "greater than" may be used, or either of the two criteria "less than or equal to" and "less than" may be used.
[0059] In this disclosure, the expression "at least one processor executes a first process, executes a second process, ... executes an nth process" or a corresponding expression indicates a concept including a case where the entity executing the n processes from the first process to the nth process, i.e., the processor, changes midway through. In other words, this expression indicates a concept including both a case where all n processes are executed by the same processor and a case where the processor changes among the n processes according to an arbitrary policy.
[0060] [Note] As can be seen from the various examples above, the present disclosure includes the following aspects. (Appendix 1) at least one processor; the at least one processor: a parameter set, which is a set of one or more parameters indicating the conditions for producing a target composition, and a computational model showing the relationship between the parameter set and a characteristic value of the target composition obtained by an experiment based on the parameter set, together with a confidence interval, is repeatedly updated by regression analysis using sample data containing one or more pairs of the parameter set and the characteristic value, while increasing the number of such pairs in the sample data; determining a new parameter set based on the updated computational model each time the computational model is updated; when a new property value of the target composition is not obtained by the experiment based on the new parameter set, a tentative property value is set based on the confidence interval corresponding to the new parameter set, and the computational model is updated using the sample data further including a new pair of the new parameter set and the tentative property value. Experiment support system. (Appendix 2) wherein the at least one processor, in updating the computational model, Identifying one or more pairs including the characteristic values obtained by the experiment from the sample data as an experimental sample group; tentatively updating the computational model through the regression analysis using the experimental sample group; For each of one or more parameter sets for which the characteristic value was not obtained by the experiment, setting the provisional characteristic value based on the confidence interval corresponding to the parameter set in the provisionally updated computational model, and generating a pair of the parameter set and the provisional characteristic value; updating the computational model by the regression analysis using the sample data composed of the experimental sample group and a non-experimental sample group composed of the one or more generated pairs; 1. An experiment support system as described in Appendix 1. (Appendix 3) the at least one processor: determining the new parameter set by an optimization method based on the updated computational model; If the new characteristic value cannot be obtained based on the new parameter set, the tentative characteristic value is set based on the confidence interval corresponding to the new parameter set and a target value of the optimization method. 10. The experiment support system according to claim 1 or 2. (Appendix 4) the at least one processor sets, as the provisional characteristic value, a value in the confidence interval corresponding to the new parameter set that is farthest from a limit value corresponding to the target value; The experiment support system described in Appendix 3. (Appendix 5) 1. An experiment support method executed by an experiment support system having at least one processor, comprising: a step of repeatedly updating a computational model that indicates the relationship between a parameter set, which is a set of one or more parameters that indicate the conditions for producing a target composition, and a characteristic value of the target composition obtained by an experiment based on the parameter set, together with a confidence interval, by regression analysis using sample data containing one or more pairs of the parameter set and the characteristic value, while increasing the number of such pairs in the sample data; determining a new parameter set based on the updated computational model each time the computational model is updated; When a new property value of the target composition is not obtained by the experiment based on the new parameter set, setting a tentative property value based on the confidence interval corresponding to the new parameter set, and updating the computational model using the sample data further including a new pair of the new parameter set and the tentative property value; Experimental support methods, including: (Appendix 6) a step of repeatedly updating a computational model that indicates the relationship between a parameter set, which is a set of one or more parameters that indicate the conditions for producing a target composition, and a characteristic value of the target composition obtained by an experiment based on the parameter set, together with a confidence interval, by regression analysis using sample data containing one or more pairs of the parameter set and the characteristic value, while increasing the number of such pairs in the sample data; determining a new parameter set based on the updated computational model each time the computational model is updated; When a new property value of the target composition is not obtained by the experiment based on the new parameter set, setting a tentative property value based on the confidence interval corresponding to the new parameter set, and updating the computational model using the sample data further including a new pair of the new parameter set and the tentative property value; An experimental support program that runs the following on a computer.
[0061] The computational model obtained by regression analysis mathematically shows the relationship between a parameter set and a property value. However, in reality, even if experiments are conducted based on a parameter set, property values may not be obtained. In Supplementary Notes 1, 5, and 6, a new parameter set for an experiment can be determined by setting tentative property values using the confidence interval of the computational model, increasing sample data, and then updating the computational model. This results in efficient composition search. For example, it can reduce the number of experiments, time, and cost required to obtain the target composition. In addition, introducing tentative property values and increasing sample data before performing regression analysis can also be expected to improve the accuracy of the computational model.
[0062] According to Appendix 2, the confidence interval of a computational model updated without using tentative characteristic values is an experimentally based calculation result, and therefore can be said to have a higher likelihood than when the computational model is updated using tentative characteristic values. Therefore, by using the confidence interval, the tentative characteristic values can be reset to a more plausible value. Then, by updating the computational model using sample data including the reset tentative characteristic values, a more accurate computational model can be obtained. As a result, the search for the target composition can be more efficiently carried out.
[0063] According to Appendix 3, by setting provisional characteristic values taking into consideration not only the confidence interval but also the target value of the optimization method, it becomes possible to make the process of the optimization method converge more quickly, thereby enabling the search for the target composition to proceed more efficiently.
[0064] For parameter sets for which actual characteristic values cannot be obtained, it can be said that the calculation model will better reflect the actual situation if the characteristic values in the calculation model are set as far as possible from the target values of the optimization method. According to Appendix 4, setting temporary characteristic values in this way increases the accuracy of the calculation model and allows the processing by the optimization method to converge more quickly. [Explanation of symbols]
[0065] 10...experiment support system, 11...acquisition unit, 12...update unit, 13...determination unit, 14...selection unit, 20...user terminal, 200, 300...computational model.
Claims
1. at least one processor; the at least one processor: a parameter set, which is a set of one or more parameters indicating conditions for producing a target composition, and a computational model showing the relationship between the parameter set and a characteristic value of the target composition obtained by an experiment based on the parameter set, together with a confidence interval, is repeatedly updated by regression analysis using sample data containing one or more pairs of the parameter set and the characteristic value, while increasing the number of such pairs in the sample data; determining a new parameter set based on the updated computational model each time the computational model is updated; when a new property value of the target composition is not obtained by the experiment based on the new parameter set, a tentative property value is set based on the confidence interval corresponding to the new parameter set, and the computational model is updated using the sample data further including a new pair of the new parameter set and the tentative property value. Experiment support system.
2. wherein the at least one processor, in updating the computational model, Identifying one or more pairs including the characteristic values obtained by the experiment from the sample data as an experimental sample group; tentatively updating the computational model through the regression analysis using the experimental sample group; For each of one or more parameter sets for which the characteristic value was not obtained by the experiment, setting the provisional characteristic value based on the confidence interval corresponding to the parameter set in the provisionally updated computational model, and generating a pair of the parameter set and the provisional characteristic value; updating the computational model by the regression analysis using the sample data composed of the experimental sample group and a non-experimental sample group composed of the one or more generated pairs; The experiment support system according to claim 1 .
3. the at least one processor: determining the new parameter set by an optimization method based on the updated computational model; If the new characteristic value cannot be obtained based on the new parameter set, the tentative characteristic value is set based on the confidence interval corresponding to the new parameter set and a target value of the optimization method.
3. The experiment support system according to claim 1.
4. the at least one processor sets, as the provisional characteristic value, a value in the confidence interval corresponding to the new parameter set that is farthest from a limit value corresponding to the target value; The experiment support system according to claim 3 .
5. 1. An experiment support method executed by an experiment support system having at least one processor, comprising: a step of repeatedly updating a computational model that indicates the relationship between a parameter set, which is a set of one or more parameters that indicate the conditions for producing a target composition, and a characteristic value of the target composition obtained by an experiment based on the parameter set, together with a confidence interval, by regression analysis using sample data containing one or more pairs of the parameter set and the characteristic value, while increasing the number of such pairs in the sample data; determining a new parameter set based on the updated computational model each time the computational model is updated; When a new property value of the target composition is not obtained by the experiment based on the new parameter set, setting a tentative property value based on the confidence interval corresponding to the new parameter set, and updating the computational model using the sample data further including a new pair of the new parameter set and the tentative property value; Experimental support methods, including:
6. a step of repeatedly updating a computational model that indicates the relationship between a parameter set, which is a set of one or more parameters that indicate the conditions for producing a target composition, and a characteristic value of the target composition obtained by an experiment based on the parameter set, together with a confidence interval, by regression analysis using sample data containing one or more pairs of the parameter set and the characteristic value, while increasing the number of such pairs in the sample data; determining a new parameter set based on the updated computational model each time the computational model is updated; When a new property value of the target composition is not obtained by the experiment based on the new parameter set, setting a tentative property value based on the confidence interval corresponding to the new parameter set, and updating the computational model using the sample data further including a new pair of the new parameter set and the tentative property value; An experimental support program that runs the following on a computer.
Citation Information
Patent Citations
Methods and systems for biopharmaceutical development
JP2023548364A