Pharmaceutical support device, method for operating pharmaceutical support device, and program for operating pharmaceutical support device

Abstract

A pharmaceutical support device is provided with a processor that performs: a process for acquiring prescription information relating to a prescription for a candidate preservation solution, which is a candidate for a preservation solution for a biomedical product containing a protein; acquiring a plurality of types of measurement data measured in a test for confirming the preservation stability of the candidate preservation solution; using a machine learning model that outputs prediction data of a future time point of first measurement data, which is at least one of the plurality of types of measurement data; and inputting the prescription information, the first measurement data, and second measurement data, which is at least one of the plurality of types of measurement data other than the first measurement data, into a machine learning model, and outputting prediction data from the machine learning model.

Description

Technical Field

[0001] The technology of the present invention relates to a pharmaceutical preparation support device, a working method of the pharmaceutical preparation support device, and a working procedure of the pharmaceutical preparation support device. Background Art

[0002] Recently, biopharmaceuticals have attracted attention due to their high efficacy and few side effects. Biopharmaceuticals, for example, use proteins such as interferon and antibodies as active ingredients. Biopharmaceuticals are stored in a storage solution. In order to stably maintain the quality of biopharmaceuticals, it is important to make the prescription of the storage solution (also called the preparation prescription) suitable for the biopharmaceuticals.

[0003] The storage solution contains a buffer, an additive, and a surfactant. The prescription of the storage solution includes, for example, the type and concentration of each of the buffer, additive, and surfactant, and the hydrogen ion index (pH (Potential Of Hydrogen) value) of the storage solution itself.

[0004] In the past, when determining the prescription of the storage solution, a variety of storage solutions were prepared by changing the combination of buffer, additives and surfactants, and the storage stability of the protein in each storage solution was confirmed by actual experiments. However, such operations are very laborious. Therefore, for example, the following technology is proposed: as in <Theresa K. Cloutier, etc., Machine Learning Models of Antibody-Excipient Preferential Interactions for Use in Computational Formulation Design, Mol. Pharmaceutics, 17, 9, 3589-3599, 2020.> (hereinafter, marked as non-patent literature), based on the information of the protein of the biopharmaceutical and the information of the additive of the storage solution, the molecular dynamics (MD: Molecular Dynamics) method or the machine learning model is used to predict the storage stability of the protein for the additive without conducting experiments. Summary of the invention

[0005] Technical issues to be solved by the invention

[0006] In the techniques described in the non-patent literature, the storage stability of the storage solution is not confirmed by actual tests. Therefore, there is a risk that the predicted results will be far different from the actual storage stability. Therefore, the present inventors considered preparing a plurality of storage solutions and conducting tests for, for example, two weeks, and using the measurement data obtained by actual measurements at that time point, predicting the storage stability of the protein in each storage solution after, for example, four weeks.

[0007] However, even if the storage stability of the storage solution is the same, it involves many aspects, such as the storage stability of the protein against additives described in non-patent literature, the storage stability of the protein against aggregation, the storage stability of the protein against temperature, etc. Moreover, there is a correlation between these various storage stabilities. Therefore, for example, in the case of predicting the measurement data for confirming the storage stability against protein aggregation, it may not be possible to improve the prediction accuracy by using only the measurement data for confirming the storage stability against protein aggregation obtained at that time.

[0008] One embodiment of the technology according to the present invention provides a pharmaceutical preparation support device, an operating method of the pharmaceutical preparation support device, and an operating program of the pharmaceutical preparation support device, which are capable of improving the prediction accuracy of the storage stability of a storage solution.

[0009] Means for solving technical problems

[0010] The pharmaceutical support device of the present invention includes a processor, which performs the following processing: obtaining prescription information related to a candidate for a storage solution of a biopharmaceutical containing a protein, i.e., a prescription of a candidate storage solution; obtaining multiple measurement data actually measured in an experiment to confirm the storage stability of the candidate storage solution; using a machine learning model that outputs prediction data of a future time point of at least one of the multiple measurement data, i.e., a first measurement data; inputting the prescription information, the first measurement data, and at least one of the multiple measurement data, i.e., a second measurement data, other than the first measurement data, into the machine learning model, and outputting prediction data from the machine learning model.

[0011] Preferably, the measurement data is time series data measured at at least two time points.

[0012] Preferably, the prescription information includes at least one of the types of each of the buffer, additive, and surfactant contained in the candidate storage solution, the concentrations of each of the buffer, additive, and surfactant, and the hydrogen ion index of the candidate storage solution itself.

[0013] Preferably, the measured data includes data for confirming the storage stability of the protein against aggregation and data for confirming the storage stability of the protein against temperature.

[0014] Preferably, the measurement data is any one of aggregation analysis data of subvisible particles in a candidate storage solution for proteins, analysis data based on dynamic light scattering of proteins in a candidate storage solution, analysis data based on size exclusion chromatography of proteins in a candidate storage solution, and analysis data based on differential scanning calorimetry of proteins in a candidate storage solution.

[0015] Preferably, the machine learning model outputs a type of prediction data of the first measurement data.

[0016] Preferably, the machine learning model outputs two or more types of prediction data of the first measurement data.

[0017] Preferably, the processor also inputs feature quantities derived from protein information related to the protein into the machine learning model.

[0018] Preferably the protein is an antibody.

[0019] The working method of the pharmaceutical preparation support device of the present invention comprises the following steps:

[0020] Obtaining prescription information related to a candidate for a storage solution of a biopharmaceutical containing a protein, i.e., a prescription of a candidate storage solution; obtaining a plurality of measurement data actually measured in an experiment to confirm the storage stability of the candidate storage solution; using a machine learning model that outputs prediction data of a future time point of at least one of the plurality of measurement data, i.e., a first measurement data; and inputting the prescription information, the first measurement data, and at least one of the plurality of measurement data, i.e., a second measurement data, other than the first measurement data, into the machine learning model, and outputting prediction data from the machine learning model.

[0021] The operating program of the pharmaceutical preparation support device of the present invention is used to cause a computer to execute a process including the following steps:

[0022] Obtaining prescription information related to a candidate for a storage solution of a biopharmaceutical containing a protein, i.e., a prescription of a candidate storage solution; obtaining a plurality of measurement data actually measured in an experiment to confirm the storage stability of the candidate storage solution; using a machine learning model that outputs prediction data of a future time point of at least one of the plurality of measurement data, i.e., a first measurement data; and inputting the prescription information, the first measurement data, and at least one of the plurality of measurement data, i.e., a second measurement data, other than the first measurement data, into the machine learning model, and outputting prediction data from the machine learning model.

[0023] Effects of the Invention

[0024] According to the technology of the present invention, it is possible to provide a pharmaceutical preparation support device, an operating method of the pharmaceutical preparation support device, and an operating program of the pharmaceutical preparation support device that can improve the prediction accuracy of the storage stability of the storage solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is a diagram showing a pharmaceutical support server and an operator terminal.

[0026] Figure 2 This is a diagram showing a series of processes and prescription information for preparation of a candidate storage solution, addition of an antibody, a stress test, and measurement of measurement data.

[0027] Figure 3This is a diagram showing measurement data of the 0th week of the pressure test.

[0028] Figure 4 This is a diagram showing the measurement data of the first week of the pressure test.

[0029] Figure 5 This is a diagram showing measurement data of the second week of the pressure test.

[0030] Figure 6 This is a block diagram showing computers constituting the pharmaceutical manufacturing support server.

[0031] Figure 7 This is a block diagram showing a processing unit of a CPU of a pharmaceutical pharmaceutical support server.

[0032] Figure 8 It is a diagram showing a storage stability prediction model.

[0033] Fig. 9 This is a diagram showing an overview of predictions using the SVP cluster analysis data prediction model.

[0034] Fig.10 It is a table representing the learning data of the SVP cluster analysis data prediction model.

[0035] Fig.11 This is a diagram showing an overview of the processing in the learning phase of the SVP cluster analysis data prediction model.

[0036] Fig.12 This is a diagram showing an overview of predictions using the DLS analysis data prediction model.

[0037] Fig.13 This is a diagram showing an overview of predictions using a SEC analysis data prediction model.

[0038] Fig.14 This is a flowchart showing the processing flow of the pharmaceutical support server.

[0039] Fig.15 This is a diagram showing another example of prescription information.

[0040] Fig.16 This is a diagram showing a second embodiment of outputting prediction data of SVP aggregation analysis data, DLS analysis data, and SEC analysis data from one storage stability prediction model.

[0041] Fig.17 Yes means Fig.16 FIG. 4 is a diagram showing an overview of the processing in the training phase of the storage stability prediction model.

[0042] Fig.18This is a diagram showing a feature quantity derivation unit that derives feature quantities from antibody information.

[0043] Fig.19 It is a diagram showing a third embodiment in which feature values ​​are further input into the storage stability prediction model. DETAILED DESCRIPTION

[0044] [First embodiment]

[0045] As an example, Figure 1 As shown, the pharmaceutical support server 10 is connected to the operator terminal 11 via the network 12. The pharmaceutical support server 10 is an example of a "pharmaceutical support device" involved in the technology of the present invention. The operator terminal 11 is, for example, installed in a pharmaceutical facility and operated by an operator such as a pharmacist who is engaged in the pharmaceutical preparation of biopharmaceuticals in the pharmaceutical facility. The operator terminal 11 has a display 13 and an input device 14 such as a keyboard and a mouse. The network 12 is, for example, a WAN (Wide Area Network) such as the Internet or a public communication network. In addition, in Figure 1 In the example, only one operator terminal 11 is connected to the pharmaceutical manufacturing support server 10 , but in reality, a plurality of operator terminals 11 in a plurality of pharmaceutical facilities are connected to the pharmaceutical manufacturing support server 10 .

[0046] The operator terminal 11 sends a prediction request 15 to the pharmaceutical support server 10. The prediction request 15 is for causing the pharmaceutical support server 10 to select a candidate storage solution 35 (see Figure 2 ) is a request for prediction of the storage stability of a candidate storage solution 35. The prediction request 15 includes a prescription information / measurement data set group 16. The prescription information / measurement data set group 16 includes a plurality of prescription information 17 and measurement data 18 groups. A group of prescription information 17 and measurement data 18 is registered for each of a plurality of, for example, several to dozens of, candidate storage solutions 35. The prescription information 17 is information related to the prescription of the candidate storage solution 35. The measurement data 18 is data indicating the storage stability of the candidate storage solution 35 measured through actual tests. These prescription information 17 and measurement data 18 are also input by the operator operating the input device 14. In addition, although not shown in the figure, the prediction request 15 also includes a terminal ID (Identification Data) of the operator terminal 11 that is the source of the prediction request 15, etc.

[0047] When receiving the prediction request 15, the pharmaceutical preparation support server 10 derives the prediction data 19 at a future point in time of the measurement data 18. The pharmaceutical preparation support server 10 distributes the prediction data 19 to the operator terminal 11 that is the source of the prediction request 15. When receiving the prediction data 19, the operator terminal 11 displays the prediction data 19 on the display 13 so that the operator can view the prediction data 19.

[0048] As an example, Figure 2 As shown, the prescription information 17 includes the type 25 of the additive contained in the candidate storage solution 35 and the hydrogen ion index 26 of the candidate storage solution 35. The prescription information 17 is added with a candidate storage solution ID for uniquely identifying the candidate storage solution 35. The prescription information 17 is created by the operator based on his / her own experience, etc. The type 25 of the additive is numerically expressed in a manner such that, for example, the additive A is "1" and the additive B is "2".

[0049] The operator actually prepares the candidate storage solution 35 in the test tube 36 according to the prescription information 17. Then, the antibody 37 is added to the prepared candidate storage solution 35. Then, a 2-week stress test is performed: including a freeze / thaw test of the candidate storage solution 35 to which the antibody 37 is added is repeated multiple times (for example, 3 times); a stirring test of stirring the candidate storage solution 35 to which the antibody 37 is added with a stirrer; and a time-dependent deterioration test of leaving the candidate storage solution 35 to which the antibody 37 is added. During the 2-week stress test, the measurement data 18 is actually measured using the measuring device 38. In addition, the antibody 37 is an example of the "protein" involved in the technology of the present invention.

[0050] exist Figure 2 In the figure, examples of measurement data of the 0th week of the pressure test (marked as measurement data (0th week)) 18_0, measurement data of the 1st week of the pressure test (marked as measurement data (1st week)) 18_1, and measurement data of the 2nd week of the pressure test (marked as measurement data (2nd week)) 18_2 are shown. Here, the 0th week of the pressure test refers to the start of the pressure test. Also, the 1st week of the pressure test is a time point one week after the start of the pressure test, and the 2nd week of the pressure test is a time point two weeks after the start of the pressure test. The measuring device 38 sends these measurement data 18_0 to 18_2 to the operator terminal 11.

[0051] The freeze / thaw test and the stirring test were performed in week 0 of the pressure test. On the other hand, the time-dependent degradation test was performed from week 0 to week 2 of the pressure test. The operator prepared candidate storage solutions 35 for the freeze / thaw test, the stirring test, and the time-dependent degradation test, respectively, and added the antibody 37 to the prepared candidate storage solutions 35.

[0052] As an example, Figure 3 to Figure 5As shown, the measurement data 18 includes aggregation analysis data 40 of sub-visible particles (SVP: Sub-Visible Particle) of the antibody 37 in the candidate storage solution 35 (hereinafter referred to as SVP aggregation analysis data) and analysis data 41 based on dynamic light scattering (DLS: Dynamic Light Scattering) of the antibody 37 in the candidate storage solution 35. The SVP aggregation analysis data 40 indicates the amount of SVP of the antibody 37 in the candidate storage solution 35 that causes the decrease in the efficacy of the biopharmaceutical. The DLS analysis data 41 also indicates the amount of nanoparticles of the antibody 37 in the candidate storage solution 35 that causes the decrease in the efficacy of the biopharmaceutical.

[0053] Furthermore, the measurement data 18 includes analysis data 42 based on size exclusion chromatography (SEC: Size Exclusion Chromatography) of the antibody 37 in the candidate storage solution 35 (hereinafter referred to as SEC analysis data). The SEC analysis data 42 represents the molecular weight distribution of the antibody 37 in the candidate storage solution 35. The SVP aggregation analysis data 40, the DLS analysis data 41, and the SEC analysis data 42 are data for confirming the storage stability against aggregation of the antibody 37. That is, the SVP aggregation analysis data 40, the DLS analysis data 41, and the SEC analysis data 42 are examples of "data for confirming the storage stability against aggregation of proteins" involved in the technology of the present invention. In addition, similarly to the prescription information 17, the measurement data 18 is assigned a candidate storage solution ID for uniquely identifying the candidate storage solution 35.

[0054] exist Figure 3In the figure, the measurement data 18_0 includes SVP aggregation analysis data (labeled as SVP aggregation analysis data (freeze / thaw test) in the figure) 40_0A, DLS analysis data (labeled as DLS analysis data (freeze / thaw test) in the figure) 41_0A and SEC analysis data (labeled as SEC analysis data (freeze / thaw test) in the figure) 42_0A measured after the freeze / thaw test. In addition, the measurement data 18_0 includes SVP aggregation analysis data (labeled as SVP aggregation analysis data (stirring test) in the figure) 40_0B, DLS analysis data (labeled as DLS analysis data (stirring test) in the figure) 41_0B and SEC analysis data (labeled as SEC analysis data (stirring test) in the figure) 42_0B measured after the stirring test. Moreover, the measurement data 18_0 includes SVP aggregation analysis data of the time-dependent degradation test (marked as SVP aggregation analysis data (time-dependent degradation test) 40_0C in the figure), DLS analysis data (marked as DLS analysis data (time-dependent degradation test) in the figure) 41_0C and SEC analysis data (marked as SEC analysis data (time-dependent degradation test) in the figure) 42_0C.

[0055] The measurement data 18_0 includes analysis data 43 based on differential scanning calorimetry (DSC: Differential Scanning Calorimetry) of the antibody 37 in the candidate storage solution 35 in addition to the SVP aggregation analysis data 40, the DLS analysis data 41, and the SEC analysis data 42 (hereinafter referred to as DSC analysis data). The DSC analysis data 43 indicates values ​​related to thermophysical properties such as the glass transition point and the crystallization temperature of the antibody 37 in the candidate storage solution 35. The DSC analysis data 43 is data for confirming the storage stability of the antibody 37 against temperature. That is, the DSC analysis data 43 is an example of "data for confirming the storage stability of the protein against temperature" involved in the technology of the present invention.

[0056] exist Figure 4 In the embodiment, the measurement data 18_1 includes SVP aggregation analysis data 40_1C, DLS analysis data 41_1C, and SEC analysis data 42_1C of the time-dependent degradation test. Figure 5 , the measurement data 18_2 includes SVP aggregation analysis data 40_2C, DLS analysis data 41_2C, and SEC analysis data 42_2C of the time-dependent degradation test. Unlike the measurement data 18_0, the measurement data 18_1 and 18_2 do not include the DSC analysis data 43.

[0057] Thus, the measurement data 18 includes a variety of measurement data such as SVP aggregation analysis data 40, DLS analysis data 41, SEC analysis data 42, and DSC analysis data 43. Furthermore, the measurement data 18 includes measurement data 18_0 of the 0th week of the pressure test, measurement data 18_1 of the 1st week, and measurement data 18_2 of the 2nd week. Therefore, the measurement data 18 is time series data measured at at least two time points.

[0058] Moreover, the measurement data 18 includes data measured after a freeze / thaw test such as the SVP aggregation analysis data 40_0A, data measured after a stirring test such as the DLS analysis data 41_0B, and data measured for a time-dependent degradation test such as the SVP aggregation analysis data 40_1C. In this way, it is preferred that the measurement data 18 includes data obtained from different tests.

[0059] As an example, Figure 6 As shown, the computer constituting the pharmaceutical pharmaceutical support server 10 includes a storage 50 , a memory 51 , a CPU (Central Processing Unit) 52 , a communication unit 53 , a display 54 , and an input device 55 . These are connected to each other via a bus 56 .

[0060] The memory 50 is a hard disk drive built into a computer constituting the pharmaceutical support server 10 or connected via a cable or a network. Alternatively, the memory 50 is a disk array formed by connecting a plurality of hard disk drives. The memory 50 stores control programs such as an operating system, various application programs, and various data attached to these programs. In addition, a solid-state drive may be used instead of a hard disk drive.

[0061] The memory 51 is a working memory for the CPU 52 to perform processing. The CPU 52 loads the program stored in the storage 50 into the memory 51 and performs processing according to the program. Thus, the CPU 52 centrally controls each part of the computer. The CPU 52 is an example of a "processor" involved in the technology of the present invention. In addition, the memory 51 can be built into the CPU 52.

[0062] The communication unit 53 controls the transmission of various information with external devices such as the operator terminal 11. The display 54 displays various screens. Various screens have operation functions based on GUI (Graphical User Interface). The computer constituting the pharmaceutical support server 10 receives input of operation instructions from the input device 55 through various screens. The input device 55 is a keyboard, a mouse, a touch panel, a microphone for voice input, etc.

[0063] As an example, Figure 7As shown, a working program 60 is stored in the memory 50 of the pharmaceutical support server 10. The working program 60 is an application program for causing the computer to function as the pharmaceutical support server 10. That is, the working program 60 is an example of a "working program of a pharmaceutical support device" involved in the technology of the present invention. The memory 50 also stores a storage stability prediction model 61.

[0064] When the operation program 60 is activated, the CPU 52 of the computer constituting the pharmaceutical preparation support server 10 cooperates with the memory 51 and the like to function as a receiving unit 65 , a read / write (hereinafter referred to as RW (Read Write)) control unit 66 , a predicting unit 67 , and a distribution control unit 68 .

[0065] The receiving unit 65 receives the prediction request 15 from the operator terminal 11. As described above, the prediction request 15 includes the prescription information / measurement data set group 16, and the prescription information / measurement data set group 16 includes the prescription information 17 and the measurement data 18. Therefore, the receiving unit 65 obtains the prescription information 17 and the measurement data 18 by receiving the prediction request 15. The receiving unit 65 outputs the prescription information / measurement data set group 16 to the RW control unit 66. In addition, the receiving unit 65 outputs the terminal ID of the operator terminal 11 (not shown) to the distribution control unit 68.

[0066] The RW control unit 66 controls the storage of various data in the memory 50 and the reading of various data in the memory 50. For example, the RW control unit 66 stores the prescription information / measurement data set group 16 from the receiving unit 65 in the memory 50. In addition, the RW control unit 66 reads the prescription information / measurement data set group 16 from the memory 50 and outputs the prescription information / measurement data set group 16 to the prediction unit 67. In addition, the RW control unit 66 reads the storage stability prediction model 61 from the memory 50 and outputs the storage stability prediction model 61 to the prediction unit 67.

[0067] The prediction unit 67 derives the prediction data 19 from the prescription information 17 and the measurement data 18 using the storage stability prediction model 61. The prediction unit 67 outputs the derived prediction data 19 to the distribution control unit 68.

[0068] The distribution control unit 68 controls distribution of the prediction data 19 to the operator terminal 11 that is the transmission source of the prediction request 15 . At this time, the distribution control unit 68 specifies the operator terminal 11 that is the transmission source of the prediction request 15 based on the terminal ID from the receiving unit 65 .

[0069] As an example, Figure 8As shown, the storage stability prediction model 61 is composed of an SVP cluster analysis data prediction model 61A, a DLS analysis data prediction model 61B, and a SEC analysis data prediction model 61C. The SVP cluster analysis data prediction model 61A, the DLS analysis data prediction model 61B, and the SEC analysis data prediction model 61C are constructed, for example, by a neural network or a support vector regression (SVR: Support Vector Regression) algorithm. The SVP cluster analysis data prediction model 61A, the DLS analysis data prediction model 61B, and the SEC analysis data prediction model 61C are examples of "machine learning models" involved in the technology of the present invention.

[0070] As an example, Fig. 9 As shown, the prediction unit 67 inputs the prescription information 17 of a certain candidate preservation solution 35 and the group of measurement data 18_0 to 18_2 into the SVP cluster analysis data prediction model 61A. Then, the prediction data 19A is output from the SVP cluster analysis data prediction model 61A. The prediction data 19A is data that predicts the SVP cluster analysis data 40 of the 4th week of the stress test. In this case, the SVP cluster analysis data 40 becomes an example of the "first measurement data" involved in the technology of the present invention. And, as described above, the measurement data 18 includes the SVP cluster analysis data 40, the DLS analysis data 41, the SEC analysis data 42 and the DSC analysis data 43. Therefore, in this case, the DLS analysis data 41, the SEC analysis data 42 and the DSC analysis data 43 other than the SVP cluster analysis data 40 become an example of the "second measurement data" involved in the technology of the present invention.

[0071] As an example, Fig.10 As shown in Table 75, a plurality of learning prescription information 17L, learning measurement data 18_0L, learning measurement data 18_1L, learning measurement data 18_2L and correct answer data 19ACA are prepared as learning data 78 of the SVP cluster analysis data prediction model 61A (reference Fig.11 ). The learning prescription information 17L includes the types of additives contained in the storage solution prepared for learning (hereinafter, marked as the learning storage solution) and the hydrogen ion index of the learning storage solution. As in the case of the prescription information 17, the types of additives are digitized. The learning measurement data 18_0L to 18_2L are data obtained by measuring the storage stability of the learning storage solution prepared according to the learning prescription information 17L. In addition, the learning storage solution is a storage solution prepared in the past for various antibodies 37.

[0072] The measurement data 18_0L for learning includes SVP aggregation analysis data 40, DLS analysis data 41, and SEC analysis data 42 measured after the freeze / thaw test. Furthermore, the measurement data 18_0L for learning includes SVP aggregation analysis data 40, DLS analysis data 41, and SEC analysis data 42 measured after the stirring test. Furthermore, the measurement data 18_0L for learning includes SVP aggregation analysis data 40, DLS analysis data 41, and SEC analysis data 42 of the time-dependent degradation test. Furthermore, the measurement data 18_0L for learning includes DSC analysis data 43.

[0073] The learning measurement data 18_1L and 18_2L include SVP aggregation analysis data 40, DLS analysis data 41, and SEC analysis data 42 of the time-dependent degradation test.

[0074] The correct answer data 19ACA is data obtained by measuring the storage stability of the learning storage solution prepared according to the learning prescription information 17L, similarly to the learning measurement data 18_0L to 18_2L. Specifically, the correct answer data 19ACA is SVP aggregation analysis data 40 measured in the fourth week of the stress test.

[0075] As an example, Fig.11 As shown, in the learning phase of the SVP aggregation analysis data prediction model 61A, learning data 78 is assigned. The learning data 78 is Fig.10 The learning prescription information 17L, the learning measurement data 18_0L to 18_2L, and the correct answer data 19ACA shown in Table 75 of the embodiment of the present invention are one of the multiple groups. The learning prescription information 17L and the learning measurement data 18_0L to 18_2L in the learning data 78 are input into the SVP cluster analysis data prediction model 61A. Then, the learning prediction data 19AL is output from the SVP cluster analysis data prediction model 61A.

[0076] Based on the prediction data 19AL for learning and the correct answer data 19ACA, the loss calculation of the SVP cluster analysis data prediction model 61A using the loss function is performed. Then, based on the result of the loss calculation, the various coefficients of the SVP cluster analysis data prediction model 61A are updated and set, and the SVP cluster analysis data prediction model 61A is updated according to the updated settings.

[0077] In the learning stage of the SVP cluster analysis data prediction model 61A, the learning data 78 is replaced while the above series of processes are repeatedly performed, including inputting the learning prescription information 17L and the learning measurement data 18_0L~18_2L into the SVP cluster analysis data prediction model 61A, outputting the learning prediction data 19AL from the SVP cluster analysis data prediction model 61A, loss calculation, update setting and updating of the SVP cluster analysis data prediction model 61A. When the prediction accuracy of the learning prediction data 19AL relative to the correct answer data 19ACA reaches a predetermined set level, the repetition of the above series of processes is terminated. In this way, the SVP cluster analysis data prediction model 61A whose prediction accuracy reaches the set level is stored in the memory 50 as a storage stability prediction model 61 and is used by the prediction unit 67. In addition, regardless of the prediction accuracy of the learning prediction data 19AL relative to the correct answer data 19ACA, the learning can be terminated when the above series of processes are repeated a set number of times.

[0078] As an example, Fig.12 As shown, the prediction unit 67 inputs the prescription information 17 of a certain candidate preservation solution 35 and the group of measurement data 18_0 to 18_2 into the DLS analysis data prediction model 61B. Then, the prediction data 19B is output from the DLS analysis data prediction model 61B. The prediction data 19B is data that predicts the DLS analysis data 41 of the 4th week of the pressure test. In this case, the DLS analysis data 41 becomes an example of the "first measurement data" involved in the technology of the present invention. In addition, the SVP aggregation analysis data 40, the SEC analysis data 42 and the DSC analysis data 43 other than the DLS analysis data 41 become an example of the "second measurement data" involved in the technology of the present invention.

[0079] And, as an example, Fig.13 As shown, the prediction unit 67 inputs the prescription information 17 of a certain candidate preservation solution 35 and the group of measurement data 18_0 to 18_2 into the SEC analysis data prediction model 61C. Then, the prediction data 19C is output from the SEC analysis data prediction model 61C. The prediction data 19C is data for predicting the SEC analysis data 42 of the 4th week of the pressure test. In this case, the SEC analysis data 42 becomes an example of the "first measurement data" involved in the technology of the present invention. In addition, the SVP aggregation analysis data 40, the DLS analysis data 41 and the DSC analysis data 43 other than the SEC analysis data 42 become an example of the "second measurement data" involved in the technology of the present invention.

[0080] In this way, the SVP cluster analysis data prediction model 61A, the DLS analysis data prediction model 61B, and the SEC analysis data prediction model 61C constituting the storage stability prediction model 61 respectively output prediction data 19A, 19B, and 19C of a first measurement data.

[0081] The learning data of the DLS analysis data prediction model 61B is the same as the learning data 78 of the SVP aggregate analysis data prediction model 61A, except that the correct answer data is the DLS analysis data 41 of the fourth week of the stress test. The learning stage of the DLS analysis data prediction model 61B is also the same as the learning stage of the SVP aggregate analysis data prediction model 61A, except that the prediction data for learning is the DLS analysis data 41 of the fourth week of the stress test. Therefore, the illustration and detailed description are omitted.

[0082] Similarly, the learning data of the SEC analysis data prediction model 61C is the same as the learning data 78 of the SVP aggregation analysis data prediction model 61A, except that the correct answer data is the SEC analysis data 42 of the fourth week of the stress test. The learning stage of the SEC analysis data prediction model 61C is the same as the learning stage of the SVP aggregation analysis data prediction model 61A, except that the prediction data for learning is the SEC analysis data 42 of the fourth week of the stress test. Therefore, the illustration and detailed description are omitted.

[0083] Next, regarding the effect based on the above structure, as an example, refer to Fig.14 First, if the operation program 60 is started in the pharmaceutical support server 10, Figure 7 As shown, the CPU 52 of the pharmaceutical manufacturing support server 10 functions as a receiving unit 65 , an RW control unit 66 , a predicting unit 67 , and a distribution control unit 68 .

[0084] The operator inputs a plurality of desired prescription information 17 and measurement data 18 via the input device 14 . Then, a prediction request 15 including the prescription information / measurement data set group 16 obtained in this way is transmitted from the operator terminal 11 to the pharmaceutical manufacturing support server 10 .

[0085] In the pharmaceutical preparation support server 10, the prediction request 15 is received by the receiving unit 65. Thus, the prescription information / measurement data set group 16, that is, the prescription information 17 and the measurement data 18 are acquired (step ST100). The prescription information / measurement data set group 16 is output from the receiving unit 65 to the RW control unit 66, and is stored in the memory 50 under the control of the RW control unit 66 (step ST110).

[0086] The prescription information / measurement data set group 16 is read out from the memory 50 by the RW control unit 66 (step ST120 ). The prescription information / measurement data set group 16 is output from the RW control unit 66 to the prediction unit 67 .

[0087] In the prediction unit 67, the prescription information 17 and the measurement data 18 of a candidate storage solution 35 are input into the storage stability prediction model 61, thereby outputting the prediction data 19 from the storage stability prediction model 61 (step ST130). Fig. 9 As shown, the prescription information 17 and the measurement data 18_0 to 18_2 are input into the SVP cluster analysis data prediction model 61A, thereby outputting the prediction data 19A from the SVP cluster analysis data prediction model 61A. Fig.12 As shown, the prescription information 17 and the measurement data 18_0 to 18_2 are input into the DLS analysis data prediction model 61B, thereby outputting the prediction data 19B from the DLS analysis data prediction model 61B. Fig.13 As shown, the prescription information 17 and the measurement data 18_0 to 18_2 are input to the SEC analysis data prediction model 61C, thereby outputting the prediction data 19C from the SEC analysis data prediction model 61C. The prediction data 19A to 19C are output from the prediction unit 67 to the distribution control unit 68. The process of step ST130 is continuously repeated while all the prediction data 19 of the plurality of candidate storage solutions 35 are not output ("No" in step ST140).

[0088] When all the prediction data 19 of the plurality of candidate storage solutions 35 are output (YES in step ST140 ), the prediction data 19 are distributed to the operator terminal 11 that is the transmission source of the prediction request 15 under the control of the distribution control unit 68 (step ST150 ).

[0089] The prediction data 19 is displayed on the display 13 of the operator terminal 11. Based on the prediction data 19, the operator selects a candidate storage solution 35 to be adopted as a storage solution for the biopharmaceutical from among a plurality of candidate storage solutions 35.

[0090] As described above, the CPU 52 of the pharmaceutical support server 10 includes a receiving unit 65 and a prediction unit 67. The receiving unit 65 acquires the prescription information 17 and the measurement data 18 by receiving the prediction request 15. The prescription information 17 is information related to the prescription of the candidate storage solution 35, which is a candidate of the storage solution of the biopharmaceutical containing the antibody 37. The measurement data 18 is data actually measured in a test for confirming the storage stability of the candidate storage solution 35.

[0091] The prediction unit 67 uses the storage stability prediction model 61 that outputs prediction data 19 of a future time point of at least one of the plurality of measurement data 18 such as the SVP aggregation analysis data 40, the DLS analysis data 41, the SEC analysis data 42, and the DSC analysis data 43, that is, the first measurement data. The prediction unit 67 inputs the prescription information 17, the first measurement data, and at least one second measurement data other than the first measurement data among the plurality of measurement data 18 into the storage stability prediction model 61, and outputs the prediction data 19 from the storage stability prediction model 61.

[0092] For example, the prediction unit 67 inputs the prescription information 17, the SVP aggregation analysis data 40_0 to 40_2 from week 0 to week 2 of the pressure test, the DLS analysis data 41_0 to 41_2 from week 0 to week 2 of the pressure test, the SEC analysis data 42_0 to 42_2 from week 0 to week 2 of the pressure test, and the DSC analysis data 43 into the SVP aggregation analysis data prediction model 61A, and outputs the SVP aggregation analysis data 40 of the 4th week, i.e., the prediction data 19A, from the SVP aggregation analysis data prediction model 61A. Since not only the first measurement data but also the second measurement data related to the first measurement data contribute to the prediction of the prediction data 19, the prediction accuracy of the storage stability of the candidate storage solution 35 can be improved compared to the case where the second measurement data is not input into the storage stability prediction model 61.

[0093] The measurement data 18 is time series data measured at at least two time points. Therefore, a plurality of data that become prediction clues of the prediction data 19 can be input into the storage stability prediction model 61, thereby further improving the prediction accuracy of the prediction data 19.

[0094] like Figure 2 As shown, the prescription information 17 includes the type of additive 25 and the hydrogen ion index 26. Therefore, at least the type of additive 25 and the hydrogen ion index 26 suitable for the storage solution of the biopharmaceutical can be selected.

[0095] The measurement data 18 includes data for confirming the storage stability against aggregation of the antibody 37, such as the SVP aggregation analysis data 40, the DLS analysis data 41, and the SEC analysis data 42. In addition, the measurement data 18 includes data for confirming the storage stability against temperature of the antibody 37, such as the DSC analysis data 43. Therefore, the prediction data 19 reflecting the storage stability of the antibody 37 against both aggregation and temperature can be obtained.

[0096] like Figure 3As shown in FIG. 1 , the measurement data 18 includes SVP aggregation analysis data 40, DLS analysis data 41, SEC analysis data 42, and DSC analysis data 43 of the antibody 37 in the candidate storage solution 35. Therefore, the prediction data 19 that more accurately represents the storage stability of the candidate storage solution 35 can be obtained.

[0097] like Fig. 9 , Fig.12 and Fig.13 As shown, the storage stability prediction model 61 outputs one type of prediction data 19 of the first measurement data. Therefore, the prediction accuracy of the prediction data 19 can be improved compared to a case where one storage stability prediction model 61 outputs two or more types of prediction data 19 of the first measurement data.

[0098] Biopharmaceuticals containing antibody 37 as a protein are called antibody pharmaceuticals, and are widely used in the treatment of chronic diseases such as cancer, diabetes, and rheumatoid arthritis, as well as in the treatment of rare diseases such as hemophilia and Crohn's disease. Therefore, according to this example in which the protein is antibody 37, the development of antibody pharmaceuticals widely used in the treatment of various diseases can be further promoted.

[0099] Fig.11 The learning of the SVP aggregate analysis data prediction model 61A shown and the learning of the DLS analysis data prediction model 61B and the SEC analysis data prediction model 61C which are omitted from the illustration can be performed in the pharmaceutical support server 10 or in another device other than the pharmaceutical support server 10. Furthermore, the SVP aggregate analysis data prediction model 61A, the DLS analysis data prediction model 61B, and the SEC analysis data prediction model 61C can continue to be learned after being stored in the memory 50.

[0100] As the prescription information 17, the type 25 of the additive contained in the candidate storage solution 35 and the hydrogen ion index 26 of the candidate storage solution 35 itself are exemplified, but the invention is not limited thereto. Fig.15 As shown in the prescription information 80, the type 81 and concentration 82 of the buffer contained in the candidate storage solution 35, the concentration 83 of the additive, and the type 84 and concentration 85 of the surfactant can be included. As can be seen from the example of L-arginine hydrochloride and refined white sugar as additives, a plurality of buffers, additives, and surfactants may be used. In this case, the type and concentration are registered for each of the plurality of types.

[0101] Thus, the prescription information may include at least one of the types 81, 25, and 84 of the buffer solution, additive, and surfactant contained in the candidate storage solution 35, the concentrations 82, 83, and 85 of the buffer solution, additive, and surfactant contained in the candidate storage solution 35, and the hydrogen ion index 26 of the candidate storage solution 35 itself. In addition, the molecular formula of the buffer solution, additive, and surfactant may be added to the prescription information.

[0102] [Second embodiment]

[0103] As an example, Fig.16 As shown, the prediction unit 67 inputs the prescription information 17 of a certain candidate storage solution 35 and the group of measurement data 18_0 to 18_2 into the storage stability prediction model 90. Then, the prediction data 19A, 19B and 19C are output from the storage stability prediction model 90. That is, the storage stability prediction model 90 is a model that implements the SVP cluster analysis data prediction model 61A, the DLS analysis data prediction model 61B and the SEC analysis data prediction model 61C of the above-mentioned first embodiment through one function. The storage stability prediction model 90 is an example of a "machine learning model" involved in the technology of the present invention. And, in this case, the SVP cluster analysis data 40, the DLS analysis data 41 and the SEC analysis data 42 become an example of the "first measurement data" involved in the technology of the present invention. And, the DSC analysis data 43 becomes an example of the "second measurement data" involved in the technology of the present invention.

[0104] As an example, Fig.17 As shown, in the learning stage of the storage stability prediction model 90, learning data 92 is provided. The learning data 92 is a group of learning prescription information 17L, learning measurement data 18_0L~18_2L, and correct answer data 19ACA, 19BCA, and 19CCA. That is, the learning data 92 is data obtained by adding the correct answer data 19BCA and 19CCA to the learning data 78 of the first embodiment described above. Like the correct answer data 19ACA, the correct answer data 19BCA and 19CCA are data obtained by actually measuring the storage stability of the learning storage solution prepared according to the learning prescription information 17L. The correct answer data 19BCA is the DLS analysis data 41 of the 4th week of the pressure test. And, the correct answer data 19CCA is the SEC analysis data 42 of the 4th week of the pressure test.

[0105] The learning prescription information 17L and the learning measurement data 18_0L to 18_2L are input to the storage stability prediction model 90. Then, the storage stability prediction model 90 outputs the learning prediction data 19AL, 19BL, and 19CL.

[0106] Based on the learning prediction data 19AL and the correct answer data 19ACA, a loss calculation of the storage stability prediction model 90 using the loss function is performed. The result of the loss calculation is marked as loss LA. In addition, based on the learning prediction data 19BL and the correct answer data 19BCA, a loss calculation of the storage stability prediction model 90 using the loss function is performed. The result of the loss calculation is marked as loss LB. In addition, based on the learning prediction data 19CL and the correct answer data 19CCA, a loss calculation of the storage stability prediction model 90 using the loss function is performed. The result of the loss calculation is marked as loss LC. Then, based on the comprehensive loss L shown in the following formula (1) which is the weighted sum of these losses LA~LC, the various coefficients of the storage stability prediction model 90 are updated and set, and the storage stability prediction model 90 is updated according to the updated settings. In addition, α, β and γ are weights, α+β+γ=1.

[0107] L=LA×α+LB×β+LC×γ……(1)

[0108] In the learning phase of the storage stability prediction model 90, the input of the learning prescription information 17L and the learning measurement data 18_0L to 18_2L to the storage stability prediction model 90, the output of the learning prediction data 19AL, 19BL and 19CL from the storage stability prediction model 90, the loss calculation, the update setting and the update of the storage stability prediction model 90 are repeated while the learning data 92 is replaced. When the prediction accuracy of the learning prediction data 19AL for the correct answer data 19ACA, the prediction accuracy of the learning prediction data 19BL for the correct answer data 19BCA and the prediction accuracy of the learning prediction data 19CL for the correct answer data 19CCA respectively reach the preset setting level, the repetition of the above series of processes is terminated. In this way, the storage stability prediction model 90 whose prediction accuracy reaches the set level is stored in the memory 50 and used by the prediction unit 67. In addition, the learning may be terminated when the above series of processes are repeated for a set number of times regardless of the prediction accuracy of the learning prediction data 19AL etc. with respect to the correct answer data 19ACA etc.

[0109] Thus, in the second embodiment, the prediction data 19 of two or more first measurement data are output from one storage stability prediction model 90. Therefore, as in the first embodiment described above, there is no need to prepare a plurality of machine learning models for outputting the prediction data 19 of the first measurement data, such as the SVP aggregation analysis data prediction model 61A, the DLS analysis data prediction model 61B, and the SEC analysis data prediction model 61C, and the effort of learning the machine learning model can be saved. Furthermore, the effort of processing the output prediction data 19 can also be saved.

[0110] [Third embodiment]

[0111] As an example, Fig.18 As shown, the CPU of the pharmaceutical preparation support server of the third embodiment functions as a feature value deriving unit 95 in addition to the processing units 65 to 68 of the first embodiment. The feature value deriving unit 95 derives a feature value 97 based on the prescription information 17 and the antibody information 96 .

[0112] Antibody information 96 includes an amino acid sequence 98 of antibody 37 and a three-dimensional structure 99 of antibody 37. Antibody information 96 is obtained by analyzing antibody 37 using well-known techniques such as mass spectrometry, X-ray crystal structure analysis, and electron microscopy. The amino acid sequence 98 is a sequence that describes the order of peptide bonds of amino acids constituting antibody 37 such as asparagine (ASn for short), glutamine (Glu for short), and arginine (Arg for short) from the amino terminal to the carboxyl terminal. The amino acid sequence 98 is also called the primary structure. The three-dimensional structure 99 represents the secondary structure, tertiary structure, and quaternary structure of the amino acids constituting antibody 37. In addition to the exemplified α helix, the secondary structure also includes β folding, β turn, etc. The tertiary structure includes the exemplified dimer coiled coil structure, etc. The quaternary structure is a dimer, a trimer, and a tetramer. According to antibody 37, there are also antibodies (i.e., monomers) that do not have a quaternary structure. In this case, as exemplified, the quaternary structure is not registered. The three-dimensional structure 99 can also be obtained by inputting the amino acid sequence 98 into "AlphaFold2", a three-dimensional structure prediction program for proteins developed by DeepMind. Antibody information 96 is an example of "protein information" involved in the technology of the present invention.

[0113] The feature quantity derivation unit 95 derives the hydrophobic solvent accessible surface area (hereinafter referred to as SASA (Solvent Accessible Surface Area)) 100 of the antibody 37, the spatial aggregation tendency (hereinafter referred to as SAP (Spatial Aggregation Propensity)) 101 of the antibody 37, and the spatial charge map (hereinafter referred to as SCM (Spatial Charge Map)) 102 of the antibody 37 as the feature quantity 97. The larger the SASA100 and the larger the SAP101, the lower the stability of the antibody 37. SCM102 is a value representing the degree of charge of the antibody 37 for each region of the antibody 37. These SASA100, SAP101, and SCM102 are derived using the molecular dynamics method. Moreover, SASA100, SAP101, and SCM102 are derived only based on the antibody information 96 without referring to the prescription information 17. Therefore, SASA100, SAP101, and SCM102 are the same in each candidate storage solution 35.

[0114] In addition, the feature quantity derivation unit 95 further derives a preferential interaction coefficient (hereinafter referred to as PIC (Preferential Interaction Coefficient)) 103 described in a non-patent document as a feature quantity 97. PIC103 is a value indicating the ease of binding between the antibody 37 and the additive for each region of the antibody 37, more specifically, the degree to which the additive covers the surface of the antibody 37, and is an index indicating the compatibility between the antibody 37 and the additive. According to PIC103, the possibility of causing aggregation of the antibody 37, which causes the decrease in the efficacy of the biopharmaceutical, can be known. PIC103 is different from the above-mentioned SASA100, SAP101, and SCM102, and is derived based on both the prescription information 17 and the antibody information 96. Therefore, PIC103 is different in each candidate storage solution 35. Therefore, with respect to the plurality of feature quantities 97 of each candidate storage solution 35, SASA100, SAP101, and SCM102 are the same, but PIC103 is different. The PIC 103 may be derived using a machine learning model such as a support vector machine (SVM) instead of the molecular dynamics method. In addition, similarly to the prescription information 17 and the like, the candidate storage solution ID is assigned to the feature amount 97 .

[0115] As an example, Fig.19 As shown, in the third embodiment, the prediction unit 67 inputs the feature value 97 into the storage stability prediction model 105 in addition to the prescription information 17 and the measurement data 18_0 to 18_2. Then, the prediction data 19 is output from the storage stability prediction model 105. The storage stability prediction model 105 may be any one of the SVP aggregation analysis data prediction model 61A, the DLS analysis data prediction model 61B, and the SEC analysis data prediction model 61C of the first embodiment, or the storage stability prediction model 90 of the second embodiment.

[0116] As described above, in the third embodiment, the feature amount 97 derived from the antibody information 96 on the antibody 37 contained in the biopharmaceutical is also input to the storage stability prediction model 105. Therefore, the prediction accuracy of the prediction data 19 can be further improved.

[0117] The feature quantity 97 includes SASA100, SAP101, SCM102, and PIC103. Therefore, the prediction data 19 that more accurately represents the storage stability of the candidate storage solution 35 can be obtained. In addition, the feature quantity 97 only needs to include at least one of SASA100, SAP101, SCM102, and PIC103. Instead of the feature quantity 97 or in addition to the feature quantity 97, the antibody information 96 can be digitized and input into the storage stability prediction model 105.

[0118] The measurement data may be any of the SVP aggregation analysis data 40, the DLS analysis data 41, the SEC analysis data 42, and the DSC analysis data 43. As a storage stability prediction model, a model for outputting the DSC analysis data 43 of the fourth week as the prediction data 19 may be prepared based on the input of the prescription information 17 and the measurement data 18_0 to 18_2.

[0119] Proteins are not limited to the exemplified antibody 37. They may also be cytokines (interferons, interleukins, etc.), hormones (insulin, glucagon, follicle-stimulating hormone, erythropoietin, etc.), growth factors (IGF (Insulin-Like Growth Factor: insulin-like growth factor)-1, bFGF (Basic Fibroblast Growth Factor: basic fibroblast growth factor)), coagulation factors (factor 7, factor 8, factor 9, etc.), enzymes (lysosomal enzymes, DNA (Deoxyribonucleic Acid: deoxyribonucleic acid) degradation enzymes, etc.), Fc (Fragment Crystallizable: crystallizable fragment) fusion proteins, receptors, albumin, protein vaccines. In addition, antibodies also include bispecific antibodies, antibody-drug conjugates, low molecular weight antibodies, sugar chain modified antibodies, etc.

[0120] The method of providing the operator with the prediction data 19 is not limited to the method of displaying it on the display 13. A printed material of the prediction data 19 may be provided to the operator, or an email with the prediction data 19 may be sent to the operator's mobile terminal. A score indicating the storage stability of the candidate storage solution 35 may be derived from the prediction data 19 using a machine learning model or the like. Then, the derived score may be distributed to the operator terminal 11 instead of the prediction data 19 or in addition to the prediction data 19.

[0121] The pharmaceutical manufacturing support server 10 may be installed in each pharmaceutical manufacturing facility or in a data center independent of the pharmaceutical manufacturing facility. Furthermore, the operator terminal 11 may assume part or all of the functions of the processing units 65 to 68 of the pharmaceutical manufacturing support server 10 .

[0122] In each of the above-mentioned embodiments, for example, as a hardware structure of a processing unit (Processing Unit) that performs various processes such as a receiving unit 65, an RW control unit 66, a prediction unit 67, a distribution control unit 68, and a feature quantity derivation unit 95, various processors (Processor) shown below can be used. As described above, various processors include a general-purpose processor, i.e., a CPU 52, that executes software (a work program 60) to function as various processing units, and processors such as FPGA (Field Programmable Gate Array) whose circuit structure can be changed after manufacturing, i.e., programmable logic devices (PLD), ASIC (Application Specific Integrated Circuit), etc., which have a circuit structure specially designed for performing specific processing, i.e., a dedicated circuit, etc.

[0123] A processing unit may be composed of one of these various processors, or a combination of two or more processors of the same or different types (for example, a combination of multiple FPGAs and / or a combination of a CPU and an FPGA). Furthermore, multiple processing units may be composed of one processor.

[0124] As an example of a processor constituting multiple processing units, there is the following method: as represented by computers such as client terminals and servers, a processor is constituted by a combination of one or more CPUs and software, and the processor functions as multiple processing units. The second method is as represented by a system on chip (SoC), etc., a processor that uses an IC (Integrated Circuit) chip to implement the functions of the entire system including multiple processing units. In this way, various processing units are constituted using one or more of the above-mentioned various processors as a hardware structure.

[0125] Furthermore, as the hardware configuration of these various processors, more specifically, a circuit (circuitry) formed by combining circuit elements such as semiconductor elements can be used.

[0126] From the above contents, the techniques described in the following supplementary notes can be understood.

[0127] [Supplementary Item 1]

[0128] A pharmaceutical preparation support device comprising a processor,

[0129] The processor performs the following processing:

[0130] obtaining prescription information related to a candidate for a storage solution of a biopharmaceutical containing a protein, ie, a prescription of a candidate storage solution;

[0131] acquiring a plurality of measurement data actually measured in a test for confirming the storage stability of the candidate storage solution;

[0132] Using a machine learning model that outputs prediction data of a future time point of at least one of the plurality of measurement data, namely, first measurement data; and

[0133] The prescription information, the first measurement data, and at least one of the plurality of measurement data other than the first measurement data, namely, the second measurement data, are input into the machine learning model, and the prediction data is output from the machine learning model.

[0134] [Supplementary Item 2]

[0135] The pharmaceutical preparation support device according to Supplementary Item 1, wherein:

[0136] The measurement data is time series data measured at at least two time points.

[0137] [Supplementary Item 3]

[0138] According to the pharmaceutical preparation support device according to Supplementary item 1 or 2,

[0139] The prescription information includes at least one of the types of buffer, additive, and surfactant contained in the candidate storage solution, the concentrations of the buffer, additive, and surfactant, and the hydrogen ion index of the candidate storage solution itself.

[0140] [Supplementary Item 4]

[0141] The pharmaceutical preparation support device according to any one of Supplementary Items 1 to 3, wherein

[0142] The measurement data includes data for confirming the storage stability of the protein against aggregation and data for confirming the storage stability of the protein against temperature.

[0143] [Supplementary Item 5]

[0144] The pharmaceutical preparation support device according to Supplementary item 4, wherein:

[0145] The measurement data is any one of aggregation analysis data of subvisible particles of the protein in the candidate storage solution, analysis data of the protein in the candidate storage solution based on dynamic light scattering, analysis data of the protein in the candidate storage solution based on size exclusion chromatography, and analysis data of the protein in the candidate storage solution based on differential scanning calorimetry.

[0146] [Supplementary Item 6]

[0147] The pharmaceutical preparation support device according to any one of Supplementary Items 1 to 5, wherein

[0148] The machine learning model outputs one type of prediction data of the first measurement data.

[0149] [Supplementary Note 7]

[0150] The pharmaceutical preparation support device according to any one of Supplementary Items 1 to 5, wherein

[0151] The machine learning model outputs the prediction data of two or more types of the first measurement data.

[0152] [Supplementary Item 8]

[0153] The pharmaceutical preparation support device according to any one of Supplementary Items 1 to 7, wherein

[0154] The processor also inputs feature quantities derived from protein information related to the protein into the machine learning model.

[0155] [Supplementary Item 9]

[0156] The pharmaceutical preparation support device according to any one of Supplementary Items 1 to 8, wherein

[0157] The protein is an antibody.

[0158] The technology of the present invention can also appropriately combine the above-mentioned various embodiments and / or various variations. Moreover, it is not limited to the above-mentioned embodiments, and various structures can of course be adopted as long as it does not deviate from the main purpose. In addition, the technology of the present invention also relates to a storage medium that stores the program non-temporarily in addition to the program.

[0159] The recorded contents and illustrated contents shown above are detailed descriptions of the parts involved in the technology of the present invention, which are only an example of the technology of the present invention. For example, the descriptions related to the above-mentioned structure, function, action and effect are descriptions related to an example of the structure, function, action and effect of the parts involved in the technology of the present invention. Therefore, without departing from the scope of the technical purpose of the present invention, it is of course possible to delete unnecessary parts of the recorded contents and illustrated contents shown above, or to add or replace new elements. In addition, in order to avoid complicated situations and facilitate understanding of the parts involved in the technology of the present invention, descriptions related to technical common sense that does not particularly need to be explained in terms of enabling the implementation of the technology of the present invention are omitted in the recorded contents and illustrated contents shown above.

[0160] In this specification, "A and / or B" has the same meaning as "at least one of A and B". That is, "A and / or B" may mean only A, only B, or a combination of A and B. Furthermore, in this specification, the same idea as "A and / or B" is also applicable to the case where three or more items are expressed by connecting them with "and / or".

[0161] All documents, patent applications, and technical standards described in this specification are incorporated herein by reference to the same extent as if each document, patent application, or technical standard was specifically and individually described as being incorporated by reference.

Claims

1. A pharmaceutical preparation support device comprising a processor, The processor performs the following processing: obtaining prescription information related to a candidate for a storage solution of a biopharmaceutical containing a protein, ie, a prescription of a candidate storage solution; acquiring a plurality of measurement data actually measured in a test for confirming the storage stability of the candidate storage solution; using a machine learning model that outputs predicted data of a future time point of at least one of the plurality of measurement data, namely, first measurement data; and The prescription information, the first measurement data, and at least one of the plurality of measurement data other than the first measurement data, namely, the second measurement data, are input into the machine learning model, and the prediction data is output from the machine learning model.

2. The pharmaceutical preparation support device according to claim 1, in, The measurement data is time series data measured at at least two time points.

3. The pharmaceutical preparation support device according to claim 1, in, The prescription information includes at least one of the types of buffer, additive, and surfactant contained in the candidate storage solution, the concentrations of the buffer, additive, and surfactant, and the hydrogen ion index of the candidate storage solution itself.

4. The pharmaceutical preparation support device according to claim 1, in, The measurement data includes data for confirming the storage stability of the protein against aggregation and data for confirming the storage stability of the protein against temperature.

5. The pharmaceutical preparation support device according to claim 4, in, The measurement data is any one of aggregation analysis data of subvisible particles of the protein in the candidate storage solution, analysis data of the protein in the candidate storage solution based on dynamic light scattering, analysis data of the protein in the candidate storage solution based on size exclusion chromatography, and analysis data of the protein in the candidate storage solution based on differential scanning calorimetry.

6. The pharmaceutical preparation support device according to claim 1, in, The machine learning model outputs the predicted data of the first measurement data.

7. The pharmaceutical preparation support device according to claim 1, in, The machine learning model outputs the prediction data of two or more types of the first measurement data.

8. The pharmaceutical preparation support device according to claim 1, in, The processor also inputs feature quantities derived from protein information related to the protein into the machine learning model.

9. The pharmaceutical preparation support device according to claim 1, in, The protein is an antibody.

10. A method for operating a pharmaceutical support device, comprising the following steps: obtaining prescription information related to a candidate for a storage solution of a biopharmaceutical containing a protein, ie, a prescription of a candidate storage solution; acquiring a plurality of measurement data actually measured in a test for confirming the storage stability of the candidate storage solution; using a machine learning model that outputs predicted data of a future time point of at least one of the plurality of measurement data, namely, first measurement data; and The prescription information, the first measurement data, and at least one of the plurality of measurement data other than the first measurement data, namely, the second measurement data, are input into the machine learning model, and the prediction data is output from the machine learning model.

11. An operating program of a pharmaceutical preparation support device, which is used to cause a computer to execute a process comprising the following steps: obtaining prescription information related to a candidate for a storage solution of a biopharmaceutical containing a protein, ie, a prescription of a candidate storage solution; acquiring a plurality of measurement data actually measured in a test for confirming the storage stability of the candidate storage solution; using a machine learning model that outputs predicted data of a future time point of at least one of the plurality of measurement data, namely, first measurement data; and The prescription information, the first measurement data, and at least one of the plurality of measurement data other than the first measurement data, namely, the second measurement data, are input into the machine learning model, and the prediction data is output from the machine learning model.

Images