Pharmaceutical support device, method of operating the pharmaceutical support device, and operating program for the pharmaceutical support device.

Abstract

This pharmaceutical assistance device comprises a processor, wherein the processor performs prediction processing in stages by using a plurality of machine-learning models that are respectively provided for a plurality of types of preservation stability, wherein each machine-learning model outputs prediction data that indicates the preservation stability, at a future time, of a candidate preservation solution that is a candidate of a preservation solution of a biomedicine. The prediction processing is performed by inputting, to the machine-learning model, prescription information pertaining to the prescription of the candidate preservation solution to be predicted, and measurement data obtained by actually measuring the preservation stability of the candidate preservation solution that has actually been prepared, and by causing the machine-learning model to output the prediction data, wherein the prediction data obtained from the previous stage prediction processing is input to the machine-learning model in the next stage prediction processing.

Description

Technical Field

[0001] The technology of the present disclosure relates to a pharmaceutical support device, a method for operating the pharmaceutical support device, and an operation program for the pharmaceutical support device.

Background Art

[0002] Recently, biopharmaceuticals have attracted attention due to their high drug efficacy and few side effects. Biopharmaceuticals contain 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 formulation of the storage solution (also referred to as the formulation recipe) suitable for the biopharmaceuticals.

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

[0004] Traditionally, determining the formulation of a preservation solution involved preparing multiple preservation solutions by changing combinations of buffers, additives, and surfactants, and then verifying the preservation stability of the protein in each solution through actual experiments. However, this process is extremely time-consuming. Therefore, techniques have been proposed to predict the preservation stability of proteins with respect to additives without conducting experiments, using molecular dynamics (MD) methods or machine learning models based on information about the biopharmaceutical protein and the additives in the preservation solution, such as "Theresa K. Cloutier, et al., Machine Learning Models of Antibody-Excipient Preferential Interactions for Use in Computational Formulation Design, Mol. Pharmaceutics, 17, 9, 3589-3599, 2020." (hereinafter referred to as Reference 1). [Overview of the project] [Problems that the invention aims to solve]

[0005] The storage stability of a preservation solution encompasses a wide range of factors, including the storage stability against protein additives as described in Reference 1, as well as the storage stability against protein aggregation and the storage stability of proteins against temperature. Predicting these multiple types of storage stability together as a unified storage stability can lead to low prediction accuracy, potentially hindering the successful discovery of suitable preservation solution formulations for biopharmaceuticals. However, since these multiple types of storage stability influence each other, predicting them individually may also fail to lead to the successful discovery of suitable preservation solution formulations for biopharmaceuticals.

[0006] One embodiment of the technology of this disclosure provides a pharmaceutical support device, a method of operating the pharmaceutical support device, and an operating program for the pharmaceutical support device that can reduce the risk of being unable to successfully search for a suitable preservation solution formulation for a biopharmaceutical. [Means for solving the problem]

[0007] The pharmaceutical support device of this disclosure includes a processor, the processor is a machine learning model that outputs predictive data indicating the storage stability of a candidate storage solution, which is a candidate storage solution for a biopharmaceutical, at a future point in time, and uses multiple machine learning models provided for each of the multiple types of storage stability, inputs formulation information relating to the formulation of the candidate storage solution to be predicted and measurement data obtained by actually measuring the storage stability of the prepared candidate storage solution into the machine learning model, performs a prediction process in stages using multiple machine learning models to output predictive data from the machine learning model, and inputs the predictive data obtained in the preceding prediction process into the machine learning model in the subsequent prediction process.

[0008] The multiple machine learning models provided for each of the multiple types of storage stability preferably correspond to at least two of the following: storage stability against protein aggregation of the biopharmaceutical, storage stability against protein temperature, and storage stability against protein degradation over time.

[0009] The processor preferably also inputs the formulation information of the candidate storage solutions that have actually been prepared into the machine learning model.

[0010] Preferably, the machine learning model outputs the confidence level of the predicted data along with the predicted data, and the processor also inputs the confidence level obtained in the preceding prediction process into the machine learning model in the subsequent prediction process.

[0011] The processor preferably also inputs features derived based on protein information about the proteins contained in the biopharmaceutical into the machine learning model.

[0012] The feature quantities preferably include at least one of the following: the solvent-exposed surface area of ​​the protein, the spatial aggregation tendency, the spatial charge map, and an index indicating the compatibility between the protein and the additives contained in the candidate preservation solution.

[0013] The measurement data is preferably time-series data measured at at least two points in time.

[0014] The formulation information for the candidate preservation solution to be predicted preferably includes at least one of the following: the type of buffer, additive, and surfactant contained in the candidate preservation solution; the concentration of each of the buffer, additive, and surfactant; and the hydrogen ion concentration of the candidate preservation solution itself.

[0015] The measurement data preferably includes at least one of the following: aggregation analysis data of subvisible particles in a candidate preservation solution of the protein contained in the biopharmaceutical; dynamic light scattering analysis data of the protein in the candidate preservation solution; size exclusion chromatography analysis data of the protein in the candidate preservation solution; and differential scanning calorometry analysis data of the protein in the candidate preservation solution.

[0016] The proteins contained in biopharmaceuticals are preferably antibodies.

[0017] The method of operating the pharmaceutical support device of this disclosure is a machine learning model that outputs predictive data indicating the storage stability of a candidate storage solution, which is a candidate storage solution for a biopharmaceutical, at a future point in time, and includes using multiple machine learning models provided for each of the multiple types of storage stability, inputting formulation information relating to the formulation of the candidate storage solution to be predicted and measurement data obtained by actually measuring the storage stability of the prepared candidate storage solution into the machine learning model, and performing a prediction process in stages using multiple machine learning models to output predictive data from the machine learning model, and inputting the predictive data obtained in the preceding prediction process into the machine learning model in the subsequent prediction process.

[0018] The operating program for the pharmaceutical support device of this disclosure is a machine learning model that outputs predictive data indicating the storage stability of a candidate storage solution, which is a candidate storage solution for a biopharmaceutical, at a future point in time. The program uses multiple machine learning models, each provided for a different type of storage stability, and inputs formulation information relating to the formulation of the candidate storage solution to be predicted, and measurement data obtained by actually measuring the storage stability of the prepared candidate storage solution, into the machine learning model. This prediction process is performed stepwise using multiple machine learning models to output predictive data from the machine learning model, and the predictive data obtained in the preceding prediction process is input into the machine learning model in the subsequent prediction process. The program causes a computer to execute these processes. [Effects of the Invention]

[0019] The technology disclosed herein provides a pharmaceutical support device, a method for operating the pharmaceutical support device, and an operating program for the pharmaceutical support device that can reduce the risk of being unable to successfully search for a suitable preservation solution formulation for a biopharmaceutical. [Brief explanation of the drawing]

[0020] [Figure 1] This diagram shows a pharmaceutical support server and operator terminal. [Figure 2] This figure shows the information exchanged between the pharmaceutical support server and the operator terminal during the prediction of storage stability against protein aggregation in the first stage. [Figure 3] This figure shows the information exchanged between the pharmaceutical support server and the operator terminal during the second stage of predicting the temperature-dependent storage stability of the protein. [Figure 4] This figure shows the information exchanged between the pharmaceutical support server and the operator terminal during the prediction of the storage stability of the protein against degradation over time in the third stage. [Figure 5] This figure shows the first predicted prescription information. [Figure 6] This figure shows the second predicted prescription information. [Figure 7] This figure shows the third predictive prescription information. [Figure 8] It is a diagram showing the first measured prescription information. [Figure 9] It is a diagram showing the second measured prescription information. [Figure 10] It is a diagram showing the third measured prescription information. [Figure 11] It is a diagram showing a series of processes of preparing a preparation solution, adding an antibody, performing a stress test, and measuring measurement data, and the measurement data. [Figure 12] It is a diagram showing the first measurement data. [Figure 13] It is a diagram showing the second measurement data. [Figure 14] It is a diagram showing the third measurement data. [Figure 15] It is a block diagram showing a computer constituting a pharmaceutical support server. [Figure 16] It is a block diagram showing a processing unit of a CPU of a pharmaceutical support server. [Figure 17] It is a diagram showing a storage stability prediction model. [Figure 18] [[ID=3,2]]It is a diagram showing an overview of the first-stage prediction using an agglomeration storage stability prediction model. [Figure 19] It is a diagram showing details of the first-stage prediction using an agglomeration storage stability prediction model. [Figure 20] It is a table showing learning data of an agglomeration storage stability prediction model. [Figure 21] It is a diagram showing an overview of processing in the learning phase of an agglomeration storage stability prediction model. [Figure 22] It is a diagram showing an overview of the second-stage prediction using a temperature storage stability prediction model. [Figure 23] It is a diagram showing details of the second-stage prediction using a temperature storage stability prediction model. [Figure 24] It is a table showing learning data of a temperature storage stability prediction model. [Figure 25] It is a diagram showing an overview of processing in the learning phase of a temperature storage stability prediction model. [Figure 26]This figure shows an overview of the third stage of prediction using a model for predicting storage stability over time. [Figure 27] This figure shows the details of the third stage prediction using a model for predicting storage stability over time. [Figure 28] This table shows the training data for the model that predicts storage stability over time. [Figure 29] This figure shows an overview of the processing during the learning phase of the model for predicting storage stability over time. [Figure 30] This flowchart shows the processing procedure of the pharmaceutical support server in the first stage of prediction. [Figure 31] This flowchart shows the processing procedure of the pharmaceutical support server in the second stage of prediction. [Figure 32] This flowchart shows the processing procedure of the pharmaceutical support server in the third stage of prediction. [Figure 33] This diagram shows an overview of the first stage prediction, the second stage prediction, and the third stage prediction. [Figure 34] This figure shows how the first confidence level is output from the aggregation and storage stability prediction model along with the first prediction data. [Figure 35] This figure shows a configuration in which the first confidence level is input to the temperature storage stability prediction model, and the second confidence level is output from the temperature storage stability prediction model along with the second prediction data. [Figure 36] This figure shows a configuration in which the first and second confidence levels are input into a model for predicting storage stability over time, and the model outputs the third confidence level along with the third prediction data. [Figure 37] This table shows the training data for the temperature storage stability prediction model in the second embodiment. [Figure 38] This figure shows the details of the first stage prediction using the aggregation and storage stability prediction model in the second embodiment. [Figure 39] This diagram shows the feature derivation unit, which derives feature quantities based on antibody information. [Figure 40] This figure shows how feature variables are also input into the aggregation and preservation stability prediction model. [Figure 41] This figure shows how feature variables are also input into a temperature storage stability prediction model. [Figure 42] This figure shows an example in which feature quantities are also input into a model for predicting storage stability over time. [Figure 43] This figure shows how the measured formulation information of the actually prepared solution is input into the storage stability prediction model as predicted formulation information. [Figure 44] This figure shows another example of a storage stability prediction model. [Figure 45] This figure shows another example of predictive prescription information. [Figure 46] This figure shows another example of measurement data. [Modes for carrying out the invention]

[0021] [First Embodiment] As an example, as shown in Figure 1, the pharmaceutical support server 10 is connected to the operator terminal 11 via a network 12. The pharmaceutical support server 10 is an example of a "pharmaceutical support device" related to the technology disclosed herein. The operator terminal 11 is installed, for example, in a pharmaceutical facility and operated by operators such as pharmacists involved in the manufacture of biopharmaceuticals at the pharmaceutical facility. The operator terminal 11 has a display 13 and input devices 14 such as a keyboard and mouse. The network 12 is, for example, a WAN (Wide Area Network) such as the Internet or a public communication network. Although only one operator terminal 11 is connected to the pharmaceutical support server 10 in Figure 1, in reality, multiple operator terminals 11 from multiple pharmaceutical facilities are connected to the pharmaceutical support server 10.

[0022] As an example, as shown in Figure 2, the operator terminal 11 sends a first prediction request 15_1 to the pharmaceutical support server 10. The first prediction request 15_1 is a request to the pharmaceutical support server 10 to perform a first-stage prediction of the storage stability of candidate storage solutions, which are candidates for storage solutions of biopharmaceuticals, or more specifically, a prediction of the storage stability against aggregation of antibody 37 (see Figure 11), which is the active ingredient of the biopharmaceutical. The first prediction request 15_1 includes a first prediction prescription information group 16_1 and a first measured prescription information / first measurement data set group 17_1.

[0023] The first predicted formulation information group 16_1 includes multiple first predicted formulation information 18_1. The first predicted formulation information 18_1 is information regarding the formulation of a candidate preservation solution (hereinafter referred to as the target solution) that is the target of the first stage of prediction. In this embodiment, the target solution is not actually prepared. The first measured formulation information / first measurement data set group 17_1 includes a set of multiple first measured formulation information 19_1 and multiple first measurement data 20_1. The first measured formulation information 19_1 is information regarding the formulation of a candidate preservation solution (hereinafter referred to as the prepared solution) 35 (see Figure 11) that has actually been prepared. The first measurement data 20_1 is data obtained by actually measuring the storage stability of the prepared solution 35 against aggregation. The first predicted formulation information 18_1, the first measured formulation information 19_1, and the first measurement data 20_1 are input by the operator operating the input device 14. Although not shown in the diagram, the first prediction request 15_1 includes a terminal ID (Identification Data) and the like to uniquely identify the operator terminal 11 that sent the first prediction request 15_1.

[0024] When the first prediction request 15_1 is received, the pharmaceutical support server 10 derives first prediction data 21_1 indicating the storage stability against aggregation of the target solution at a future point in time. The pharmaceutical support server 10 derives multiple first prediction data 21_1 for each of the multiple first prediction prescription information 18_1. The pharmaceutical support server 10 delivers the first prediction data group 22_1, which consists of multiple first prediction data 21_1, to the operator terminal 11 that sent the first prediction request 15_1.

[0025] When the first prediction data set 22_1 is received, the operator terminal 11 displays multiple first prediction data 21_1 on the display 13 and makes the multiple first prediction data 21_1 available for the operator to view. Based on the multiple first prediction data 21_1, the operator narrows down the target solutions from among the multiple target solutions for the first stage of prediction to proceed to the second stage of prediction.

[0026] As an example, as shown in Figure 3, after the completion of the first stage of prediction, the operator terminal 11 sends a second prediction request 15_2 to the pharmaceutical support server 10. The second prediction request 15_2 is a request to the pharmaceutical support server 10 to perform a second stage prediction of the storage stability of the target solution narrowed down in the first stage, or more specifically, a prediction of the storage stability of the antibody 37 with respect to temperature. The second prediction request 15_2 includes a second prediction prescription information group 16_2, a second measured prescription information / second measurement data set group 17_2, and a first prediction data group 22_1.

[0027] The second predicted prescription information group 16_2 includes multiple second predicted prescription information sets 18_2. The second predicted prescription information sets 18_2 are information regarding the prescriptions of the target solutions narrowed down in the first stage. The second measured prescription information / second measurement data set group 17_2 includes sets of multiple second measured prescription information sets 19_2 and multiple second measurement data sets 20_2. The second measured prescription information sets 19_2 are information regarding the prescription of the prepared solution 35. The second measurement data sets 20_2 are data obtained by measuring the storage stability of the prepared solution 35 with respect to temperature. The second predicted prescription information sets 18_2, the second measured prescription information sets 19_2, and the second measurement data sets 20_2 are also input by the operator operating the input device 14. The first predicted data group 22_1 is attached as is, which was distributed from the pharmaceutical support server 10 in the first stage of prediction. Although not shown in the diagram, the second prediction request 15_2 also includes the terminal ID of the operator terminal 11, etc.

[0028] When the pharmaceutical support server 10 receives a second prediction request 15_2, it derives second prediction data 21_2 indicating the storage stability of the target solution with respect to temperature at a future point in time. The pharmaceutical support server 10 derives multiple second prediction data 21_2 for each of the multiple second prediction prescription information 18_2. The pharmaceutical support server 10 delivers the second prediction data group 22_2, which consists of multiple second prediction data 21_2, to the operator terminal 11 that sent the second prediction request 15_2.

[0029] When the second prediction data group 22_2 is received, the operator terminal 11 displays multiple second prediction data 21_2 on the display 13 and makes the multiple second prediction data 21_2 available for the operator to view. Based on the multiple second prediction data 21_2, the operator narrows down the target solutions from among the multiple target solutions for the second stage of prediction to proceed to the third stage of prediction.

[0030] As an example, as shown in Figure 4, after the completion of the second stage of prediction, the operator terminal 11 sends a third prediction request 15_3 to the pharmaceutical support server 10. The third prediction request 15_3 is a request to the pharmaceutical support server 10 to perform a third stage prediction of the storage stability of the target solution narrowed down in the second stage, or more specifically, a prediction of the storage stability of the antibody 37 against degradation over time. The third prediction request 15_3 includes a third prediction prescription information group 16_3, a third measured prescription information / third measurement data set group 17_3, a first prediction data group 22_1, and a second prediction data group 22_2.

[0031] The third predicted prescription information group 16_3 includes multiple third predicted prescription information 18_3. The third predicted prescription information 18_3 is information regarding the prescription of the target solution narrowed down in the second stage. The third measured prescription information / third measurement data set group 17_3 includes sets of multiple third measured prescription information 19_3 and multiple third measurement data 20_3. The third measured prescription information 19_3 is information regarding the prescription of the prepared solution 35. The third measurement data 20_3 is data obtained by measuring the storage stability of the prepared solution 35 against deterioration over time. The third predicted prescription information 18_3, the third measured prescription information 19_3, and the third measurement data 20_3 are also input by the operator operating the input device 14. The first predicted data group 22_1 and the second predicted data group 22_2 are attached as they were delivered from the pharmaceutical support server 10 in the first and second stage predictions. Although not shown in the diagram, the third prediction request 15_3 also includes the terminal ID of the operator terminal 11, etc.

[0032] When the pharmaceutical support server 10 receives a third prediction request 15_3, it derives third prediction data 21_3 indicating the storage stability of the target solution against deterioration over time at a future point in time. The pharmaceutical support server 10 derives multiple third prediction data 21_3 for each of the multiple third prediction prescription information 18_3. The pharmaceutical support server 10 delivers the third prediction data group 22_3, which consists of multiple third prediction data 21_3, to the operator terminal 11 that sent the third prediction request 15_3.

[0033] When the third prediction data group 22_3 is received, the operator terminal 11 displays multiple third prediction data 21_3 on the display 13 and makes the multiple third prediction data 21_3 available for the operator to view. Based on the multiple third prediction data 21_3, the operator selects a target solution to be adopted as a biopharmaceutical preservation solution from among multiple target solutions that are the target of the second stage prediction.

[0034] In the following, if there is no need to distinguish between the first prediction request 15_1, the second prediction request 15_2, and the third prediction request 15_3, they may be collectively referred to as prediction request 15. Similarly, the first prediction prescription information group 16_1, the second prediction prescription information group 16_2, and the third prediction prescription information group 16_3 may be collectively referred to as prediction prescription information group 16. In addition, the first measured prescription information / first measurement dataset group 17_1, the second measured prescription information / second measurement dataset group 17_2, and the third measured prescription information / third measurement dataset group 17_3 may be collectively referred to as measured prescription information / measurement dataset group 17.

[0035] The first predicted prescription information 18_1, the second predicted prescription information 18_2, and the third predicted prescription information 18_3 are sometimes collectively referred to as predicted prescription information 18. Also, the first measured prescription information 19_1, the second measured prescription information 19_2, and the third measured prescription information 19_3 are sometimes collectively referred to as measured prescription information 19. Furthermore, the first measurement data 20_1, the second measurement data 20_2, and the third measurement data 20_3 are sometimes collectively referred to as measurement data 20. Also, the first predicted data 21_1, the second predicted data 21_2, and the third predicted data 21_3 are sometimes collectively referred to as predicted data 21. In addition, the first predicted data group 22_1, the second predicted data group 22_2, and the third predicted data group 22_3 are sometimes collectively referred to as predicted data group 22.

[0036] Predictive formulation information 18 is an example of "formulation information relating to the formulation of a candidate preservation solution to be predicted" related to the technology of this disclosure. Measured formulation information 19 is an example of "formulation information of a candidate preservation solution actually prepared" related to the technology of this disclosure.

[0037] As an example, as shown in Figure 5, the first predicted formulation information 18_1 includes the type of additive 25 contained in the target solution and the hydrogen ion index 26 of the target solution. As an example, as shown in Figure 6, the second predicted formulation information 18_2 includes the storage temperature 27 in addition to the type of additive 25 and the hydrogen ion index 26. As an example, as shown in Figure 7, the third predicted formulation information 18_3 includes the storage period 28 in addition to the type of additive 25 and the hydrogen ion index 26. The predicted formulation information 18 is assigned a solution ID to uniquely identify the target solution. The predicted formulation information 18 is created by the operator based on the antibody 37 information and their own experience, etc. The type of additive 25 is quantified, for example, additive A is "1", additive B is "2", etc.

[0038] As an example, as shown in Figure 8, the first measured formulation information 19_1 includes the type of additive 25 contained in the prepared solution 35 and the hydrogen ion index 26 of the prepared solution 35. As an example, as shown in Figure 9, the second measured formulation information 19_2 includes the storage temperature 27 in addition to the type of additive 25 and the hydrogen ion index 26. As an example, as shown in Figure 10, the third measured formulation information 19_3 includes the storage period 28 in addition to the type of additive 25 and the hydrogen ion index 26. The measured formulation information 19 is also assigned a solution ID. The measured formulation information 19 is created by, for example, using two types of additives 25 and varying the hydrogen ion index 26 in increments of 0.5, for example, 6.5, 7.0, and 7.5.

[0039] As an example, as shown in Figure 11, the operator prepares the solution 35 in a test tube 36 according to the actual formulation information 19. Then, the antibody 37 is added to the solution 35. Subsequently, a stress test involving repeated freezing and thawing, stirring, and heating is performed for a predetermined period (e.g., 4 to 8 weeks). During this stress test, measurement data 20 is measured using a measuring instrument 38. Figure 11 shows an example in which measurement data 20(1W), 20(2W), 20(3W), 20(4W), etc. are measured in the first, second, third, fourth, ... weeks of the stress test. The measuring instrument 38 transmits the measurement data 20 to the operator terminal 11.

[0040] The measurement data 20, as shown in the measurement data 20 (1W) for week 1, includes agglutination analysis data 40 of sub-visible particles (SVP) of antibody 37 in the prepared solution 35 (hereinafter referred to as SVP agglutination analysis data), and differential scanning calorimetry (DSC) analysis data 41 of antibody 37 in the prepared solution 35 (hereinafter referred to as DSC analysis data). The SVP agglutination analysis data 40 represents the amount of SVP of antibody 37 in the prepared solution 35 that causes a decrease in the efficacy of the biopharmaceutical. A lower value for the SVP agglutination analysis data 40 indicates higher storage stability of antibody 37 against agglutination. The DSC analysis data 41 represents values ​​related to thermophysical properties of antibody 37 in the prepared solution 35, such as the glass transition temperature and crystallization temperature. A lower value for the DSC analysis data 41 indicates higher storage stability of antibody 37 against temperature. The SVP agglutination analysis data 40 and DSC analysis data 41 for each week are distinguished by adding labels such as "1W" (indicating week 1), "2W" (indicating week 2), etc.

[0041] In Figure 11, for convenience, only one measuring instrument 38 is depicted; however, in reality, various measuring instruments 38 are prepared to measure the SVP agglutination analysis data 40 and DSC analysis data 41. Furthermore, while Figure 11 only illustrates the preparation of one prepared solution 35, the addition of antibody 37, the stress test, and the measurement of measurement data 20, in reality, these preparations, additions, stress tests, and measurement of measurement data 20 are performed for each of the multiple prepared solutions 35.

[0042] As an example, as shown in Figure 12, the first measurement data 20_1 includes the SVP agglutination analysis data 40(1W) for week 1 and the SVP agglutination analysis data 40(2W) for week 2. The SVP agglutination analysis data 40(1W) for week 1 and the SVP agglutination analysis data 40(2W) for week 2 are examples of "time-series data measured at at least two points in time" relating to the technology of this disclosure. As an example, as shown in Figure 13, the second measurement data 20_2 includes the DSC analysis data 41(1W) for week 1 and the DSC analysis data 41(2W) for week 2. The DSC analysis data 41(1W) for week 1 and the DSC analysis data 41(2W) for week 2 are examples of "time-series data measured at at least two points in time" relating to the technology of this disclosure. As an example, as shown in Figure 14, the third measurement data 20_3 includes the SVP agglutination analysis data 40(4W) for week 4. The measurement data 20 is assigned a solution ID to uniquely identify the prepared solution, similar to the predicted formulation information 18.

[0043] As an example, as shown in Figure 15, the computer comprising the pharmaceutical support server 10 includes storage 50, memory 51, CPU (Central Processing Unit) 52, communication unit 53, display 54, and input device 55. These are interconnected via a bus line 56.

[0044] Storage 50 is a hard disk drive built into the computer constituting the pharmaceutical support server 10, or connected via cable or network. Alternatively, storage 50 is a disk array consisting of multiple hard disk drives installed in series. Storage 50 stores control programs such as the operating system, various application programs, and various data associated with these programs. A solid-state drive may be used instead of a hard disk drive.

[0045] Memory 51 is work memory for the CPU 52 to execute processing. The CPU 52 loads the program stored in storage 50 into memory 51 and executes processing according to the program. In this way, the CPU 52 comprehensively controls each part of the computer. CPU 52 is an example of a "processor" related to the technology of this disclosure. Note that memory 51 may be built into the CPU 52.

[0046] The communication unit 53 controls the transmission of various information to external devices such as the operator terminal 11. The display 54 displays various screens. These screens are equipped with GUI (Graphical User Interface) operation functions. The computer constituting the pharmaceutical support server 10 accepts operation instructions from input devices 55 through the various screens. The input devices 55 include keyboards, mice, touch panels, and microphones for voice input.

[0047] As an example, as shown in Figure 16, the storage 50 of the pharmaceutical support server 10 stores an operating program 60. The operating program 60 is an application program that causes the computer to function as the pharmaceutical support server 10. In other words, the operating program 60 is an example of an "operating program for a pharmaceutical support device" related to the technology of this disclosure. The storage 50 also stores a storage stability prediction model 61.

[0048] When the operating program 60 is started, the CPU 52 of the computer constituting the pharmaceutical support server 10 works in cooperation with the memory 51 and the like to function as a reception unit 65, a read / write (hereinafter abbreviated as RW (Read Write)) control unit 66, a prediction unit 67, and a distribution control unit 68.

[0049] The reception unit 65 receives a prediction request 15 from the operator terminal 11. As described above, the prediction request 15 includes a prediction prescription information group 16 and an actual prescription information / measurement data set group 17. In the case of the second prediction request 15_2 and the third prediction request 15_3, it also includes the first prediction data group 22_1 obtained in the first stage prediction and the second prediction data group 22_2 obtained in the second stage prediction (hereinafter, these will be collectively referred to as the preceding prediction data group 22PS). Therefore, by receiving the prediction request 15, the reception unit 65 acquires the prediction prescription information group 16, the actual prescription information / measurement data set group 17, and the preceding prediction data group 22PS (if included). The reception unit 65 outputs the prediction prescription information group 16, the actual prescription information / measurement data set group 17, and the preceding prediction data group 22PS (if included) to the RW control unit 66. Furthermore, the reception unit 65 outputs the terminal ID of the operator terminal 11 (not shown) to the distribution control unit 68.

[0050] The RW control unit 66 controls the storage of various data to the storage 50 and the reading of various data from the storage 50. For example, the RW control unit 66 stores the predicted prescription information group 16, the measured prescription information / measurement data set group 17, and the preceding predicted data group 22PS (if included) from the reception unit 65 in the storage 50. The RW control unit 66 also reads the predicted prescription information group 16, the measured prescription information / measurement data set group 17, and the preceding predicted data group 22PS (if included) from the storage 50 and outputs the predicted prescription information group 16, the measured prescription information / measurement data set group 17, and the preceding predicted data group 22PS (if included) to the prediction unit 67. Furthermore, the RW control unit 66 reads the storage stability prediction model 61 from the storage 50 and outputs the storage stability prediction model 61 to the prediction unit 67.

[0051] The prediction unit 67 uses the storage stability prediction model 61 to derive the prediction data group 22 based on the prediction prescription information group 16, the actual prescription information / measurement data set group 17, and the previous prediction data group 22PS (if included). The prediction unit 67 outputs the derived prediction data group 22 to the distribution control unit 68.

[0052] The distribution control unit 68 controls the distribution of the prediction data set 22 to the operator terminal 11 that sent the prediction request 15. At this time, the distribution control unit 68 identifies the operator terminal 11 that sent the prediction request 15 based on the terminal ID from the reception unit 65.

[0053] As an example, as shown in Figure 17, the storage stability prediction model 61 consists of an aggregation storage stability prediction model 61A, a temperature storage stability prediction model 61B, and a time-degradation storage stability prediction model 61C. The aggregation storage stability prediction model 61A, the temperature storage stability prediction model 61B, and the time-degradation storage stability prediction model 61C are constructed, for example, using a neural network or a support vector regression (SVR) algorithm. The aggregation storage stability prediction model 61A, the temperature storage stability prediction model 61B, and the time-degradation storage stability prediction model 61C are examples of "machine learning models" related to the technology of this disclosure.

[0054] As an example, as shown in Figure 18, the prediction unit 67 inputs the first predicted formulation information 18_1, the first measured formulation information 19_1, and the first measured data 20_1 into the aggregation and storage stability prediction model 61A in the first stage of prediction. The aggregation and storage stability prediction model 61A then outputs the first predicted data 21_1. The first predicted data 21_1 is also assigned a solution ID, similar to the first predicted formulation information 18_1, etc.

[0055] As an example, as shown in Tables 80 and 81 of Figure 19, in the first stage of prediction, the prediction unit 67 inputs the first prediction formulation information 18_1 and one of the sets of first measured formulation information 19_1 and first measurement data 20_1 for multiple types of prepared solutions 35 into the agglutination and storage stability prediction model 61A, and outputs provisional first prediction data from the agglutination and storage stability prediction model 61A. The prediction unit 67 repeats this process for multiple types of prepared solutions 35, and outputs multiple provisional first prediction data from the agglutination and storage stability prediction model 61A. The prediction unit 67 calculates the average value of the multiple provisional first prediction data, and the calculated average value is set as the first prediction data 21_1. The provisional first prediction data, and thus the first prediction data 21_1, is data that predicts the SVP agglutination analysis data 40 (4W) for the target solution at week 4, represented by the first prediction formulation information 18_1. Note that the first prediction data 21_1 may be the median, maximum, or minimum of multiple provisional first prediction data, instead of the mean value, or it may be a weighted mean calculated by assigning appropriate weights to multiple provisional first prediction data.

[0056] Figure 19 illustrates the case where five provisional first prediction data points are derived for five prepared solutions 35, No. 1 to No. 5, where the additive type 25 is either L-arginine hydrochloride or L-histidine hydrochloride, and the hydrogen ion concentration is one of 6.5, 7.0, or 7.5. The provisional first prediction data points are 320 for No. 1, 250 for No. 2, 3, and 5, and 280 for No. 4. Therefore, the first prediction data point 21_1 is (320 + 250 + 250 + 280 + 250) / 5 = 270.

[0057] As an example, as shown in Table 85 in Figure 20, multiple sets of the following are stored in storage 50 as training data 90 for the aggregation storage stability prediction model 61A: training first predicted formulation information 18_1L, training first measured formulation information 19_1L, training first measurement data 20_1L, and first correct answer data 21_1CA (see Figure 21 for all). The same content is registered in training first predicted formulation information 18_1L and training first measured formulation information 19_1L. Training first measurement data 20_1L is data obtained by measuring the storage stability of the prepared solution 35 in which training first measured formulation information 19_1L is registered, and consists of SVP aggregation analysis data 40(1W) for week 1 and SVP aggregation analysis data 40(2W) for week 2.

[0058] Similarly, the first correct answer data 21_1CA is data obtained by actually measuring the storage stability of the prepared solution 35 for which the first measured formulation information 19_1L for learning purposes was registered. Specifically, the first correct answer data 21_1CA is the SVP agglutination analysis data 40 (4W) from week 4. If the first measurement data 20_1L for learning purposes was not measured, a pre-prepared value or a mask indicating that no data is available will be registered.

[0059] As an example, as shown in Figure 21, in the learning phase of the aggregation preservation stability prediction model 61A, learning data 90 is provided. The learning data 90 is a set of the first learning prediction prescription information 18_1L, the first learning measurement prescription information 19_1L, the first learning measurement data 20_1L, and the first correct answer data 21_1CA, which are registered in Table 85 shown in Figure 20. The first learning prediction prescription information 18_1L, the first learning measurement prescription information 19_1L, and the first learning measurement data 20_1L from the learning data 90 are input to the aggregation preservation stability prediction model 61A, and as a result, the first learning prediction data 21_1L is output from the aggregation preservation stability prediction model 61A.

[0060] Based on the first training prediction data 21_1L and the first ground truth data 21_1CA, a loss calculation is performed on the agglomeration preservation stability prediction model 61A using a loss function. Then, based on the result of the loss calculation, various coefficients of the agglomeration preservation stability prediction model 61A are updated, and the agglomeration preservation stability prediction model 61A is updated according to the update settings.

[0061] In the learning phase of the agglomeration preservation stability prediction model 61A, the above series of processes—input of the first prediction prescription information 18_1L for learning, the first measured prescription information 19_1L for learning, and the first measurement data 20_1L for learning into the agglomeration preservation stability prediction model 61A, output of the first prediction data 21_1L for learning from the agglomeration preservation stability prediction model 61A, loss calculation, update settings, and updating the agglomeration preservation stability prediction model 61A—are repeated while the learning data 90 is exchanged. The repetition of the above series of processes ends when the prediction accuracy of the first prediction data 21_1L for learning with respect to the first ground truth data 21_1CA reaches a predetermined set level. The agglomeration preservation stability prediction model 61A, whose prediction accuracy has reached the set level, is stored in the storage 50 and used by the prediction unit 67. Note that learning may be terminated after the above series of processes has been repeated a set number of times, regardless of the prediction accuracy of the first prediction data 21_1L for learning with respect to the first ground truth data 21_1CA.

[0062] As an example, as shown in Figure 22, in the second stage of prediction, the prediction unit 67 inputs the second prediction formulation information 18_2, the storage temperature 27 of the second measured formulation information 19_2, the second measurement data 20_2, and the first prediction data 21_1 into the temperature storage stability prediction model 61B. The temperature storage stability prediction model 61B then outputs the second prediction data 21_2. The second prediction data 21_2 is also assigned a solution ID, similar to the second prediction formulation information 18_2, etc. The second stage prediction by the temperature storage stability prediction model 61B is an example of the "later prediction processing" related to the technology of this disclosure. Furthermore, the first stage prediction by the aggregation storage stability prediction model 61A is an example of the "preceding prediction processing" related to the technology of this disclosure, and the first prediction data 21_1 is an example of the "prediction data obtained in the preceding prediction processing" related to the technology of this disclosure.

[0063] As an example, as shown in Tables 95 and 96 of Figure 23, in the second stage of prediction, the prediction unit 67 inputs the second prediction formulation information 18_2, one of the pairs of storage temperature 27 and second measurement data 20_2 from the second measured formulation information 19_2 for multiple types of prepared solutions 35, and the first prediction data 21_1 into the temperature storage stability prediction model 61B, and causes the temperature storage stability prediction model 61B to output provisional second prediction data. The prediction unit 67 repeats this process for multiple types of prepared solutions 35, causing the temperature storage stability prediction model 61B to output multiple provisional second prediction data. The prediction unit 67 calculates the average value of the multiple provisional second prediction data, and the calculated average value is set as the second prediction data 21_2. The provisional second prediction data, and thus the second prediction data 21_2, is data that predicts the DSC analysis data 41 (4W) at the fourth week when the target solution represented by the second prediction formulation information 18_2 is heated at the storage temperature 27. Furthermore, the second prediction data 21_2 may also be the median, maximum, or minimum of multiple provisional second prediction data, instead of the mean value, or it may be a weighted mean calculated by assigning appropriate weights to multiple provisional second prediction data.

[0064] Figure 23 illustrates the case where, as in Figure 19, five types of prepared solutions 35, No. 1 to No. 5, are derived, with the additive type 25 being either L-arginine hydrochloride or L-histidine hydrochloride, the hydrogen ion concentration being either 6.5, 7.0, or 7.5, and the storage temperature 27 being either 40°C, 45°C, or 50°C. The provisional second prediction data for No. 1 to No. 5 are 50 for No. 1 and 2, 48 for No. 3, 62 for No. 4, and 60 for No. 5. Therefore, the second prediction data 21_ 2 The answer is (50+50+48+62+60) / 5 = 54.

[0065] As an example, as shown in Table 100 in Figure 24, multiple sets of the following are stored in storage 50 as training data 105 for the temperature storage stability prediction model 61B: training second predicted formulation information 18_2L, storage temperature 27 for training second measured formulation information 19_2L, training first measurement data 20_1L, training second measurement data 20_2L, and second correct answer data 21_2CA (see Figure 25 for all). The same content is registered for the storage temperature 27 of training second predicted formulation information 18_2L and training second measured formulation information 19_2L. The training first measurement data 20_1L is data obtained by measuring the storage stability of the prepared solution 35 in which training second measured formulation information 19_2L is registered, and is the SVP agglutination analysis data 40 (4W) for the fourth week. This first learning measurement data 20_1L corresponds to the first prediction data 21_1 in the operational phase of the temperature storage stability prediction model 61B shown in Figures 22 and 23. The second learning measurement data 20_2L is data obtained by measuring the storage stability of the prepared solution 35 to which the second learning measurement formulation information 19_2L was registered, and consists of the DSC analysis data 41(1W) for week 1 and the DSC analysis data 41(2W) for week 2.

[0066] Similarly, the second correct answer data 21_2CA is data obtained by actually measuring the storage stability of the prepared solution 35 to which the second measured formulation information 19_2L for learning purposes was registered. Specifically, the second correct answer data 21_2CA is the DSC analysis data 41(4W) from week 4. If the first measurement data 20_1L for learning purposes has not been measured, a pre-prepared value or a mask indicating that there is no data will be registered. Alternatively, the first predicted data 21_1 predicted by the aggregation storage stability prediction model 61A may be registered as the first measurement data 20_1L for learning purposes.

[0067] As an example, as shown in Figure 25, in the learning phase of the temperature storage stability prediction model 61B, learning data 105 is provided. The learning data 105 is a set of the second prediction prescription information for learning 18_2L, the second measured prescription information for learning 19_2L, the first measurement data for learning 20_1L, the second measurement data for learning 20_2L, and the second ground truth data 21_2CA, which are registered in Table 100 shown in Figure 24. The second prediction prescription information for learning 18_2L, the second measured prescription information for learning 19_2L, the first measurement data for learning 20_1L, and the second measurement data for learning 20_2L from the learning data 105 are input to the temperature storage stability prediction model 61B, and as a result the temperature storage stability prediction model 61B outputs the second prediction data for learning 21_2L.

[0068] Based on the second training prediction data 21_2L and the second ground truth data 21_2CA, a loss calculation is performed on the temperature storage stability prediction model 61B using a loss function. Then, based on the result of the loss calculation, various coefficients of the temperature storage stability prediction model 61B are updated, and the temperature storage stability prediction model 61B is updated according to the update settings.

[0069] In the learning phase of the temperature storage stability prediction model 61B, the above series of processes—input of the second prediction prescription information 18_2L for learning, the second measured prescription information 19_2L for learning, the first measurement data 20_1L for learning, and the second measurement data 20_2L for learning into the temperature storage stability prediction model 61B, output of the second prediction data 21_2L for learning from the temperature storage stability prediction model 61B, loss calculation, update settings, and updating the temperature storage stability prediction model 61B—are repeated while the learning data 105 is exchanged. The repetition of the above series of processes ends when the prediction accuracy of the second prediction data 21_2L for learning relative to the second correct answer data 21_2CA reaches a predetermined set level. The temperature storage stability prediction model 61B, whose prediction accuracy has thus reached the set level, is stored in the storage 50 and used by the prediction unit 67. Furthermore, regardless of the prediction accuracy of the second prediction data 21_2L for training relative to the second correct answer data 21_2CA, the training may be terminated after the above series of processes has been repeated a set number of times.

[0070] As an example, as shown in Figure 26, the prediction unit 67 inputs the third predicted formulation information 18_3, the storage period 28 of the third measured formulation information 19_3, the third measurement data 20_3, the first predicted data 21_1, and the second predicted data 21_2 into the time-dependent deterioration storage stability prediction model 61C in the third stage of prediction. The time-dependent deterioration storage stability prediction model 61C then outputs the third predicted data 21_3. The third predicted data 21_3 is also assigned a solution ID, similar to the third predicted formulation information 18_3, etc. The third stage prediction by the time-dependent deterioration storage stability prediction model 61C is an example of the "subsequent prediction processing" related to the technology of this disclosure. Furthermore, the first-stage prediction by the aggregation storage stability prediction model 61A and the second-stage prediction by the temperature storage stability prediction model 61B are examples of "preliminary prediction processing" related to the technology of this disclosure, and the first prediction data 21_1 and the second prediction data 21_2 are examples of "prediction data obtained in the preliminary prediction processing" related to the technology of this disclosure. In other words, the second-stage prediction by the temperature storage stability prediction model 61B is a "subsequent prediction processing" for the first-stage prediction and a "preliminary prediction processing" for the third-stage prediction.

[0071] As an example, as shown in Tables 110 and 111 of Figure 27, in the third stage of prediction, the prediction unit 67 inputs the third prediction formulation information 18_3, one of the sets of storage period 28 and third measurement data 20_3 from the third measured formulation information 19_3 for multiple types of prepared solutions 35, the first prediction data 21_1, and the second prediction data 21_2 into the time-dependent deterioration storage stability prediction model 61C, and outputs provisional third prediction data from the time-dependent deterioration storage stability prediction model 61C. The prediction unit 67 repeats this process for multiple types of prepared solutions 35, and outputs multiple provisional third prediction data from the time-dependent deterioration storage stability prediction model 61C. The prediction unit 67 calculates the average value of the multiple provisional third prediction data, and sets the calculated average value as the third prediction data 21_3. The provisional third prediction data, and by extension the third prediction data 21_3, is data that predicts the SVP agglutination analysis data 40 for the storage period 28 (in this case, week 8) of the target solution represented by the third prediction formulation information 18_3. Note that the third prediction data 21_3 may also be the median, maximum, or minimum of multiple provisional third prediction data, or a weighted average calculated by assigning appropriate weights to multiple provisional third prediction data.

[0072] Figure 27 illustrates the case where, as in Figures 19 and 23, five types of prepared solutions 35, No. 1 to No. 5, are derived, with the additive type 25 being either L-arginine hydrochloride or L-histidine hydrochloride, the hydrogen ion concentration being either 6.5, 7.0, or 7.5, and the storage period 28 being either 6 weeks, 7 weeks, or 8 weeks. The provisional third prediction data for No. 1 to No. 5 are 320, No. 2 and 3 are 380, No. 4 is 290, and No. 5 is 420. Therefore, the third prediction data 21_ 3 The answer is (320+380+380+290+420) / 5 = 358.

[0073] As an example, as shown in Table 115 in Figure 28, multiple sets of the following are stored in storage 50 as training data 120 for the time-dependent deterioration storage stability prediction model 61C: training third predicted formulation information 18_3L, storage period 28 for the third measured formulation information 19_3L for training, first measurement data 20_1L for training, second measurement data 20_2L for training, third measurement data 20_3L for training, and third correct answer data 21_3CA (see Figure 29 for all). The same content is registered for the storage period 28 for the third predicted formulation information 18_3L for training and the storage period 28 for the third measured formulation information 19_3L for training. The first measurement data 20_1L for training is data obtained by measuring the storage stability of the prepared solution 35 in which the third measured formulation information 19_3L for training is registered, and is the SVP agglutination analysis data 40 (4W) for the fourth week. The first measurement data for learning, 20_1L, corresponds to the first prediction data 21_1 in the operational phase of the time-dependent deterioration storage stability prediction model 61C shown in Figures 26 and 27. The second measurement data for learning, 20_2L, is data obtained by actually measuring the storage stability of the prepared solution 35 to which the third measurement formulation information for learning, 19_3L, is registered, and is the DSC analysis data 41(4W) for the fourth week. This second measurement data for learning, 20_2L, corresponds to the second prediction data 21_2 in the operational phase of the time-dependent deterioration storage stability prediction model 61C shown in Figures 26 and 27.

[0074] Similarly, the third correct answer data 21_3CA is data obtained by actually measuring the storage stability of the prepared solution 35 to which the third measured formulation information 19_3L for learning purposes is registered. Specifically, the third correct answer data 21_3CA is the SVP agglutination analysis data 40 for the storage period 28 of the third measured formulation information 19_3L for learning purposes. If the first measurement data 20_1L for learning purposes has not been measured, a pre-prepared value or a mask indicating that there is no data will be registered. Alternatively, the first predicted data 21_1 predicted by the agglutination storage stability prediction model 61A may be registered as the first measurement data 20_1L for learning purposes. Similarly, if the second measurement data 20_2L for learning purposes has not been measured, a pre-prepared value or a mask indicating that there is no data will be registered, or the second predicted data 21_2 predicted by the temperature storage stability prediction model 61B may be registered as the second measurement data 20_2L for learning purposes.

[0075] As an example, as shown in Figure 29, in the learning phase of the aging degradation storage stability prediction model 61C, learning data 120 is provided. The learning data 120 is a set of the third prediction prescription information for learning 18_3L, the third measured prescription information for learning 19_3L, the first measurement data for learning 20_1L, the second measurement data for learning 20_2L, the third measurement data for learning 20_3L, and the third correct answer data 21_3CA, which are registered in Table 115 shown in Figure 28. The third prediction prescription information for learning 18_3L, the third measured prescription information for learning 19_3L, the first measurement data for learning 20_1L, the second measurement data for learning 20_2L, and the third measurement data for learning 20_3L from the learning data 120 are input to the aging degradation storage stability prediction model 61C, and as a result, the third prediction data for learning 21_3L is output from the aging degradation storage stability prediction model 61C.

[0076] Based on the third training prediction data 21_3L and the third ground truth data 21_3CA, a loss calculation is performed on the time-degradation storage stability prediction model 61C using a loss function. Then, based on the result of the loss calculation, the various coefficients of the time-degradation storage stability prediction model 61C are updated, and the time-degradation storage stability prediction model 61C is updated according to the update settings.

[0077] In the learning phase of the aging degradation storage stability prediction model 61C, the above series of processes—input of the third prediction prescription information 18_3L for learning, the third measured prescription information 19_3L for learning, the first measurement data 20_1L for learning, the second measurement data 20_2L for learning, and the third measurement data 20_3L for learning into the aging degradation storage stability prediction model 61C, output of the third prediction data 21_3L for learning from the aging degradation storage stability prediction model 61C, loss calculation, update settings, and updating the aging degradation storage stability prediction model 61C—are repeated while the learning data 120 is exchanged. The repetition of the above series of processes ends when the prediction accuracy of the third prediction data 21_3L for learning relative to the third correct answer data 21_3CA reaches a predetermined set level. The aging degradation storage stability prediction model 61C, whose prediction accuracy has thus reached the set level, is stored in the storage 50 and used by the prediction unit 67. Furthermore, regardless of the prediction accuracy of the third prediction data 21_3L used for training relative to the third correct answer data 21_3CA, training may be terminated after the above series of processes has been repeated a set number of times.

[0078] Next, the operation of the above configuration will be explained with reference to the flowcharts in Figures 30, 31, and 32. First, when the operating program 60 is started in the pharmaceutical support server 10, as shown in Figure 16, the CPU 52 of the pharmaceutical support server 10 functions as the reception unit 65, the RW control unit 66, the prediction unit 67, and the distribution control unit 68.

[0079] The operator inputs desired first predicted prescription information 18_1, first measured prescription information 19_1, and first measurement data 20_1 via the input device 14 in order to predict the storage stability of the first-stage antibody 37 against aggregation. Then, the operator terminal 11 sends a first prediction request 15_1, including the first predicted prescription information group 16_1 and the first measured prescription information / first measurement data set group 17_1, to the pharmaceutical support server 10.

[0080] As an example, as shown in Figure 30, in the pharmaceutical support server 10, the first prediction request 15_1 is received by the reception unit 65. As a result, the first predicted prescription information group 16_1 and the first measured prescription information / first measurement data set group 17_1 are acquired (step ST100). The first predicted prescription information group 16_1 and the first measured prescription information / first measurement data set group 17_1 are output from the reception unit 65 to the RW control unit 66 and stored in the storage 50 under the control of the RW control unit 66 (step ST110).

[0081] The RW control unit 66 reads out the first predicted prescription information group 16_1 and the first measured prescription information / first measurement data set group 17_1 from the storage 50 (step ST120). The first predicted prescription information group 16_1 and the first measured prescription information / first measurement data set group 17_1 are output from the RW control unit 66 to the prediction unit 67.

[0082] As shown in Figures 18 and 19, in the prediction unit 67, the first predicted formulation information 18_1, the first measured formulation information 19_1, and the first measured data 20_1 are input to the aggregation and storage stability prediction model 61A, and as a result, the aggregation and storage stability prediction model 61A outputs the first predicted data 21_1 (step ST130). The first predicted data 21_1 is output from the prediction unit 67 to the distribution control unit 68. The process in step ST130 is repeated until all of the first predicted data 21_1 for multiple types of target solutions is output (NO in step ST140).

[0083] If all first prediction data 21_1 for multiple types of target solutions are output (YES in step ST140), the first prediction data group 22_1 is distributed to the operator terminal 11 that sent the first prediction request 15_1 under the control of the distribution control unit 68 (step ST150).

[0084] The display 13 of the operator terminal 11 shows the first prediction data group 22_1. Based on the first prediction data group 22_1, the operator narrows down the target solutions from among the multiple target solutions for the first stage of prediction to those that will proceed to the second stage of prediction.

[0085] Next, the operator inputs the desired second predicted formulation information 18_2, second measured formulation information 19_2, and second measurement data 20_2 via the input device 14 in order to predict the temperature storage stability of the second-stage antibody 37. Then, the operator terminal 11 sends a second prediction request 15_2, which includes the second predicted formulation information group 16_2, the second measured formulation information / second measurement data set group 17_2, and the first predicted data group 22_1 obtained in the first stage prediction, to the pharmaceutical support server 10.

[0086] As an example, as shown in Figure 31, in the pharmaceutical support server 10, the second prediction request 15_2 is received by the reception unit 65. As a result, the second predicted prescription information group 16_2, the second measured prescription information / second measurement data set group 17_2, and the first predicted data group 22_1 are acquired (step ST200). The second predicted prescription information group 16_2, the second measured prescription information / second measurement data set group 17_2, and the first predicted data group 22_1 are output from the reception unit 65 to the RW control unit 66 and stored in the storage 50 under the control of the RW control unit 66 (step ST210).

[0087] The RW control unit 66 reads out the second predicted prescription information group 16_2, the second measured prescription information / second measurement data set group 17_2, and the first predicted data group 22_1 from the storage 50 (step ST220). The second predicted prescription information group 16_2, the second measured prescription information / second measurement data set group 17_2, and the first predicted data group 22_1 are output from the RW control unit 66 to the prediction unit 67.

[0088] As shown in Figures 22 and 23, in the prediction unit 67, the second predicted formulation information 18_2, the storage temperature 27 of the second measured formulation information 19_2, the second measurement data 20_2, and the first predicted data 21_1 are input to the temperature storage stability prediction model 61B, and the second predicted data 21_2 is output from the temperature storage stability prediction model 61B (step ST230). The second predicted data 21_2 is output from the prediction unit 67 to the distribution control unit 68. The process in step ST230 is repeated until all of the second predicted data 21_2 for multiple types of target solutions is output (NO in step ST240).

[0089] If all second prediction data 21_2 for multiple target solutions are output (YES in step ST240), the second prediction data group 22_2 is distributed to the operator terminal 11 that sent the second prediction request 15_2 under the control of the distribution control unit 68 (step ST250).

[0090] The operator terminal 11's display 13 shows the second prediction data group 22_2. Based on the second prediction data group 22_2, the operator narrows down the target solutions from among the multiple target solutions for the second stage of prediction to those that will proceed to the third stage of prediction.

[0091] Next, the operator inputs the desired third-stage predicted formulation information 18_3, third-stage measured formulation information 19_3, and third-stage measurement data 20_3 via the input device 14 in order to predict the storage stability of the third-stage antibody 37 against degradation over time. Then, the third-stage predicted formulation information group 16_3, the third-stage measured formulation information / third-stage measurement data set group 17_3, and the first-stage predicted data obtained in the first-stage prediction are input. group 2 2 _1, and the second prediction data obtained in the second stage of prediction group 2 2 The operator terminal 11 sends a third prediction request 15_3, including _2, to the pharmaceutical support server 10.

[0092] As an example, as shown in Figure 32, in the pharmaceutical support server 10, the third prediction request 15_3 is received by the reception unit 65. As a result, the third prediction prescription information group 16_3, the third measured prescription information / third measurement data set group 17_3, the first prediction data group 22_1, and the second prediction data group 22_2 are acquired (step ST300). The third prediction prescription information group 16_3, the third measured prescription information / third measurement data set group 17_3, the first prediction data group 22_1, and the second prediction data group 22_2 are output from the reception unit 65 to the RW control unit 66 and stored in the storage 50 under the control of the RW control unit 66 (step ST310).

[0093] The RW control unit 66 reads out the third predicted prescription information group 16_3, the third measured prescription information / third measurement data set group 17_3, the first predicted data group 22_1, and the second predicted data group 22_2 from the storage 50 (step ST320). The third predicted prescription information group 16_3, the third measured prescription information / third measurement data set group 17_3, the first predicted data group 22_1, and the second predicted data group 22_2 are output from the RW control unit 66 to the prediction unit 67.

[0094] As shown in Figures 26 and 27, in the prediction unit 67, the storage period 28 of the third predicted formulation information 18_3, the storage period 28 of the third measured formulation information 19_3, the third measurement data 20_3, the first predicted data 21_1, and the second predicted data 21_2 are input to the time-dependent deterioration storage stability prediction model 61C, and the time-dependent deterioration storage stability prediction model 61C outputs the third predicted data 21_3 (step ST330). The third predicted data 21_3 is output from the prediction unit 67 to the distribution control unit 68. The process in step ST330 is repeated until all of the third predicted data 21_3 for multiple types of target solutions is output (NO in step ST340).

[0095] If all third prediction data 21_3 for multiple types of target solutions are output (YES in step ST340), the third prediction data group 22_3 is distributed to the operator terminal 11 that sent the third prediction request 15_3 under the control of the distribution control unit 68 (step ST350).

[0096] The operator terminal 11's display 13 shows the third prediction data group 22_3. Based on the third prediction data group 22_3, the operator selects a target solution to be adopted as a biopharmaceutical preservation solution from among multiple target solutions targeted for prediction in the second stage.

[0097] As described above, the prediction unit 67 of the CPU 52 of the pharmaceutical support server 10 is a storage stability prediction model 61 that outputs prediction data 21 indicating the storage stability of candidate storage solutions, which are candidates for storage solutions of biopharmaceuticals, at a future point in time, and uses storage stability prediction models 61 provided for each of multiple types of storage stability. Specifically, as shown in Figure 33, it uses an agglutination storage stability prediction model 61A that outputs first prediction data 21_1 indicating the storage stability of antibody 37 against agglutination, a temperature storage stability prediction model 61B that outputs second prediction data 21_2 indicating the storage stability of antibody 37 against temperature, and a time-degradation storage stability prediction model 61C that outputs third prediction data 21_3 indicating the storage stability of antibody 37 against time-degradation.

[0098] The prediction unit 67 inputs predicted formulation information 18 regarding the formulation of the target solution and measurement data 20 obtained by actually measuring the storage stability of the prepared solution 35 into the storage stability prediction model 61, and performs a stepwise prediction process to output predicted data 21 from the storage stability prediction model 61. Specifically, as shown in Figure 33, in the first stage of prediction, the first predicted formulation information 18_1 and the first measurement data 20_1 are input into the agglutination storage stability prediction model 61A, and the agglutination storage stability prediction model 61A outputs the first predicted data 21_1. Next, in the second stage of prediction, the second predicted formulation information 18_2 and the second measurement data 20_2 are input into the temperature storage stability prediction model 61B, and the temperature storage stability prediction model 61B outputs the second predicted data 21_2. Finally, in the third stage of prediction, the third prediction prescription information 18_3 and the third measurement data 20_3 are input into the time-dependent deterioration storage stability prediction model 61C, and the time-dependent deterioration storage stability prediction model 61C outputs the third prediction data 21_3.

[0099] Furthermore, the prediction unit 67 inputs the prediction data 21 obtained in the preceding prediction process into the storage stability prediction model 61 in the subsequent prediction process. Specifically, as shown in Figure 33, the first prediction data 21_1 obtained in the first prediction stage is input into the temperature storage stability prediction model 61B in the second prediction stage. In addition, the first prediction data 21_1 obtained in the first prediction stage and the second prediction data 21_2 obtained in the second prediction stage are input into the time-dependent deterioration storage stability prediction model 61C in the third prediction stage.

[0100] Therefore, when predicting multiple types of storage stability together in a form like integrated storage stability, compared to predicting multiple types of storage stability individually, it becomes possible to reduce the risk of being unable to successfully find a suitable storage solution formulation for biopharmaceuticals.

[0101] The training data 90 shown in Table 85 of Figure 20, the training data 105 shown in Table 100 of Figure 24, and the training data 120 shown in Table 115 of Figure 28 cannot be prepared in large quantities. Furthermore, as mentioned above, the storage stability of storage solutions varies widely. For this reason, it is difficult to predict multiple types of storage stability together with a single storage stability prediction model. Therefore, by using the technology disclosed herein, which predicts the storage stability in stages by dividing it into categories, it becomes possible to predict multiple types of storage stability more stably and accurately with less training data.

[0102] Furthermore, a stepwise approach can be used to narrow down candidate preservation solutions. For example, in the first stage, 1000 target solutions can be narrowed down to 100; in the second stage, 100 target solutions can be narrowed down to 10; and in the third stage, one target solution can be selected from the 10 target solutions to be adopted as a preservation solution for biopharmaceuticals. This approach increases the probability of formulating a preservation solution suitable for biopharmaceuticals compared to selecting one target solution from 1000 target solutions using a single evaluation criterion.

[0103] As shown in Figure 18, the prediction unit 67 also inputs the measured formulation information 19 of the prepared solution 35 into the storage stability prediction model 61. This improves the prediction accuracy of the prediction data 21. Furthermore, as shown in Figure 19, by inputting the measured formulation information 19 of multiple types of prepared solutions 35, even if unknown information (such as a hydrogen ion concentration of 6.8 in Figure 19 and a storage temperature of 42°C in Figure 23) is input as prediction formulation information 18 for a target solution that has not actually been prepared, it is possible to obtain prediction data 21 with high prediction accuracy. This eliminates the need to actually prepare the target solution, and as a result, can accelerate the development of biopharmaceuticals.

[0104] As shown in Figure 12, the SVP agglutination analysis data 40(1W) for week 1 and SVP agglutination analysis data 40(2W) for week 2 of the first measurement data 20_1, and as shown in Figure 13, the DSC analysis data 41(1W) for week 1 and DSC analysis data 41(2W) for week 2 of the second measurement data 20_2, the measurement data 20 is time-series data measured at at least two points in time. Therefore, the prediction accuracy of the first prediction data 21_1 and the second prediction data 21_2 can be improved compared to the case where the measurement data 20 is not time-series data.

[0105] As shown in Figure 5, the predicted formulation information 18 includes the type of additive 25 and the hydrogen ion concentration 26. Therefore, it is possible to select a preservation solution suitable for biopharmaceuticals, with at least the type of additive 25 and the hydrogen ion concentration 26 being relevant.

[0106] As shown in Figure 12, the measurement data 20 includes SVP agglutination analysis data 40 and DSC analysis data 41 of the antibody 37 in the prepared solution 35. Therefore, predictive data 21 that more accurately represents the storage stability of the prepared solution 35 can be output from the storage stability prediction model 61.

[0107] Biopharmaceuticals containing antibody 37 as a protein are called antibody drugs and are widely used to treat chronic diseases such as cancer, diabetes, and rheumatoid arthritis, as well as rare diseases such as hemophilia and Crohn's disease. Therefore, this example, in which antibody 37 is used as the protein, can further accelerate the development of antibody drugs that are widely used to treat various diseases.

[0108] The training of the aggregation storage stability prediction model 61A shown in Figure 21, the temperature storage stability prediction model 61B shown in Figure 25, and the aging degradation storage stability prediction model 61C shown in Figure 29 may be performed on the pharmaceutical support server 10 or on another device other than the pharmaceutical support server 10. Furthermore, the aggregation storage stability prediction model 61A, the temperature storage stability prediction model 61B, and the aging degradation storage stability prediction model 61C may continue to be trained even after being stored in the storage 50.

[0109] [Second Embodiment] As an example, in the second embodiment shown in Figures 34-37, the reliability of the prediction data 21 is also input into the storage stability prediction model.

[0110] As an example, as shown in Figure 34, the aggregation and storage stability prediction model 130A of the second embodiment outputs the first prediction data 21_1 along with the first confidence score 131_1, which is the confidence score of the prediction of the first prediction data 21_1, when the first prediction formula information 18_1, the first measured formula information 19_1, and the first measurement data 20_1 are input. The first confidence score 131_1 takes a value between 0 and 1. A first confidence score 131_1 of 0 indicates that the confidence score of the prediction of the first prediction data 21_1 is the lowest, and a first confidence score 131_1 of 1 indicates that the confidence score of the prediction of the first prediction data 21_1 is the highest.

[0111] As an example, as shown in Figure 35, in the second embodiment, the prediction unit 67 inputs the second prediction formula information 18_2, the second measured formula information 19_2, the second measurement data 20_2, and the first prediction data 21_1 obtained in the first stage of prediction, as well as the first confidence level 131_1 obtained in the first stage of prediction, into the temperature storage stability prediction model 130B. When the temperature storage stability prediction model 130B receives the second prediction formula information 18_2, the second measured formula information 19_2, the second measurement data 20_2, the first prediction data 21_1, and the first confidence level 131_1, it outputs the second prediction data 21_2 along with the second confidence level 131_2, which is the confidence level of the prediction of the second prediction data 21_2. The second confidence level 131_2, like the first confidence level 131_1, takes a value between 0 and 1. A value of 0 indicates the lowest confidence level of the prediction for the second prediction data 21_2, and a value of 1 indicates the highest confidence level of the prediction for the second prediction data 21_2. The first confidence level 131_1 is an example of the "confidence level obtained in the preceding prediction processing" related to the technology of this disclosure.

[0112] Furthermore, as shown in Figure 36 as an example, in the second embodiment, the prediction unit 67 inputs the third prediction formula information 18_3, the third measured formula information 19_3, the third measurement data 20_3, the first prediction data 21_1 obtained in the first stage prediction, and the second prediction data 21_2 obtained in the second stage prediction, as well as the first confidence level 131_1 obtained in the first stage prediction and the second confidence level 131_2 obtained in the second stage prediction, into the time-dependent deterioration storage stability prediction model 130C. When the time-dependent deterioration storage stability prediction model 130C receives the third prediction formula information 18_3, the third measured formula information 19_3, the third measurement data 20_3, the first prediction data 21_1, the first confidence level 131_1, the second prediction data 21_2, and the second confidence level 131_2 as inputs, it outputs the third prediction data 21_3 along with the third confidence level 131_3, which is the confidence level of the prediction of the third prediction data 21_3. The third confidence level 131_3 takes a value between 0 and 1, similar to the first confidence level 131_1 and the second confidence level 131_2. A value of 0 indicates the lowest confidence level of the prediction of the third prediction data 21_3, and a value of 1 indicates the highest confidence level of the prediction of the third prediction data 21_3. The first confidence level 131_1 and the second confidence level 131_2 are examples of "confidence levels obtained in the preceding prediction processing" related to the technology of this disclosure.

[0113] As an example, as shown in Table 135 in Figure 37, the training data for the temperature storage stability prediction model 130B shown in Figure 35 includes an additional item, the first confidence level for training 131_1L, in addition to the items such as the second prediction prescription information for training 18_2L in the training data 105 of the temperature storage stability prediction model 61B of the first embodiment described above. For the measured first measurement data for training 20_1L, the highest value of 1.0 is registered as the first confidence level for training 131_1L. As enclosed by the dashed line, if the first measurement data for training 20_1L has not been measured, and the first prediction data 21_1 predicted by the aggregation storage stability prediction model 130A is registered as the first measurement data for training 20_1L, then the first confidence level 131_1 output by the aggregation storage stability prediction model 130A is registered as the first confidence level for training 131_1L. Although not shown in the diagram, the learning data for the time-dependent deterioration storage stability prediction model 130C, shown in Figure 36, also includes items for the first and second confidence levels for learning. When the first prediction data 21_1 predicted by the agglomeration storage stability prediction model 130A is registered as the first measurement data 20_1L for learning, the first confidence level 131_1 is registered as the first confidence level for learning. Similarly, when the second prediction data 21_2 predicted by the temperature storage stability prediction model 130B is registered as the second measurement data 20_2L for learning, the second confidence level 131_2 is registered as the second confidence level for learning. Note that the first prediction data 21_1, etc., predicted by the agglomeration storage stability prediction model 130A may all be registered with a uniform value such as 0.5 for the first confidence level for learning, etc.

[0114] Thus, in the second embodiment, the aggregation storage stability prediction model 130A and the temperature storage stability prediction model 130B output the first confidence level 131_1 and the second confidence level 131_2 together with the first prediction data 21_1 and the second prediction data 21_2. The prediction unit 67 inputs the first confidence level 131_1 to the temperature storage stability prediction model 130B, and inputs the first confidence level 131_1 and the second confidence level 131_2 to the time-dependent deterioration storage stability prediction model 130C. Therefore, prediction data 21 that takes into account the confidence level 131 of the prediction data 21 obtained in the preceding prediction process can be output in the subsequent prediction process.

[0115] As an example, as shown in Figure 38, a weighted average may be calculated using the first confidence level 131_1 as the weight for multiple provisional first prediction data shown in Table 140 output from the agglomeration storage stability prediction model 130A, and the calculated weighted average may be used as the first prediction data 21_1. In this way, the first prediction data 21_1 that reflects the first confidence level 131_1 can be derived. Although not shown in the figure, a weighted average may also be calculated using the second confidence level 131_2 as the weight for multiple provisional second prediction data output from the temperature storage stability prediction model 130B, and the calculated weighted average may be used as the second prediction data 21_2. Similarly, a weighted average may be calculated using the third confidence level 131_3 as the weight for multiple provisional third prediction data output from the aging deterioration storage stability prediction model 130C, and the calculated weighted average may be used as the third prediction data 21_3.

[0116] Instead of using a weighted average, you could simply use the hypothetical prediction data with the highest confidence level of 131 as prediction data 21.

[0117] [Third Embodiment] In the third embodiment shown in Figures 39-42, the storage stability prediction model 61 is configured to take antibody information 146 relating to antibody 37 and feature quantities 147 derived based on the antibody information 146 as input. In this case, when training the storage stability prediction model 61, the antibody information 146 and feature quantities 147 are added as training data along with training prediction prescription information 18L, training actual prescription information 19L, etc.

[0118] As an example, as shown in Figure 39, the CPU of the pharmaceutical support server in the third embodiment functions as a feature extraction unit 145 in addition to the processing units 65 to 68 of the first embodiment. The feature extraction unit 145 derives feature quantities 147 based on the first predicted prescription information 18_1 and antibody information 146.

[0119] Antibody information 146 includes the amino acid sequence 148 of antibody 37 and the three-dimensional structure 149 of antibody 37. Antibody information 146 is obtained by analyzing antibody 37 using well-known techniques such as mass spectrometry, X-ray crystallography, and electron microscopy. The amino acid sequence 148 describes the order of peptide bonds of the amino acids constituting antibody 37, such as asparagine (abbreviation ASn), glutamine (abbreviation Glu), and arginine (abbreviation Arg), from the amino terminus to the carboxyl terminus. The amino acid sequence 148 is also called the primary structure. The three-dimensional structure 149 shows the secondary, tertiary, and quaternary structures of the amino acids constituting antibody 37. Secondary structures include the α-helix as an example, as well as β-sheets, β-turns, etc. Tertiary structures include the dimeric coiled-coil structure as an example. Quaternary structures are dimers, trimers, and tetramers. Some antibodies 37 do not have a quaternary structure (i.e., they are monomers). In that case, the quaternary structure will not be registered, as shown in the example. Antibody information 146 is an example of "protein information" relating to the technology of this disclosure.

[0120] The feature extraction unit 145 derives the hydrophobic solvent-exposed surface area (hereinafter abbreviated as SASA (Solvent Accessible Surface Area)) 150, spatial aggregation tendency (hereinafter abbreviated as SAP (Spatial Aggregation Propensity)) 151, and spatial charge map (hereinafter abbreviated as SCM (Spatial Charge Map)) 152 of the antibody 37 as feature quantities 147. The larger the SASA 150 and the larger the SAP 151, the lower the stability of the antibody 37. The SCM 152 represents the degree of charge of the antibody 37 for each region of the antibody 37. These SASA 150, SAP 151, and SCM 152 are derived using molecular dynamics. Furthermore, SASA 150, SAP 151, and SCM 152 are derived based only on the antibody information 146 without referring to the first predicted prescription information 18_1. Therefore, SASA150, SAP151, and SCM152 are common to each of the target solutions.

[0121] Furthermore, the feature derivation unit 145 also derives the Preferred Interaction Coefficient (PIC) 153 described in Reference 1 as a feature 147. PIC 153 represents the ease of binding between the antibody 37 and the additive, more specifically, the degree to which the additive covers the surface of the antibody 37, for each region of the antibody 37, and is an indicator of the compatibility between the antibody 37 and the additive. According to PIC 153, it is possible that aggregation of the antibody 37, which causes a decrease in the efficacy of the biopharmaceutical, may occur. Unlike SASA 150, SAP 151, and SCM 152 mentioned above, PIC 153 is derived based on both the first predicted prescription information 18_1 and the antibody information 146. Therefore, PIC 153 differs for each target solution. Consequently, the multiple feature quantities 147 for each target solution are common to SASA 150, SAP 151, and SCM 152, while PIC 153 differs for each. PIC153 may be derived using machine learning models such as Support Vector Machines (SVM) instead of molecular dynamics. Note that feature 147 is assigned a solution ID, similar to the first predicted prescription information 18_1.

[0122] As an example, as shown in Figures 40-42, in the third embodiment, feature quantity 147 is also input to the aggregation storage stability prediction model 61A, the temperature storage stability prediction model 61B, and the time-dependent deterioration storage stability prediction model 61C.

[0123] Thus, in the third embodiment, the feature quantities 147 derived based on antibody information 146 regarding the antibody 37 contained in the biopharmaceutical are also input into the storage stability prediction model 61. This further improves the prediction accuracy of the prediction data 21.

[0124] Feature vector 147 includes SASA150, SAP151, SCM152, and PIC153. Therefore, the storage stability prediction model 61 can output prediction data 21 that more accurately represents the storage stability of the target solution. Note that feature vector 147 only needs to include at least one of SASA150, SAP151, SCM152, and PIC153.

[0125] Figures 40-42 show examples in which feature quantity 147 is input to the aggregation storage stability prediction model 61A, temperature storage stability prediction model 61B, and time-degradation storage stability prediction model 61C of the first embodiment, but the invention is not limited to these examples. Feature quantity 147 may also be input to the aggregation storage stability prediction model 130A, temperature storage stability prediction model 130B, and time-degradation storage stability prediction model 130C of the second embodiment.

[0126] Alternatively, instead of or in addition to feature vector 147, antibody information 146 may be quantified and input into the storage stability prediction model 61 or 130.

[0127] In each of the above embodiments, in practice, Made Information regarding the formulation of the target solution that is not subjected to stress testing is input to the storage stability prediction model 61, etc., as predicted formulation information 18, but this is not limited to this. As an example, as shown in Table 160 of Figure 43, information regarding the formulation of the prepared solution 35 that is actually prepared and subjected to stress testing, i.e., measured formulation information 19, may be input to the storage stability prediction model 61, etc., as predicted formulation information 18. In this case, the measured data 20 is obtained by actually measuring the storage stability of the prepared solution 35 in which the measured formulation information 19 was input as predicted formulation information 18. The operator then decides whether to continue the stress test of the prepared solution 35 as is or to cut their losses and stop based on the predicted data 21 output from the storage stability prediction model 61, etc.

[0128] Figure 43 illustrates a case in which the first measured formulation information 19_1, which is the first predicted formulation information 18_1, and the first measurement data 20_1 are input to the aggregation and storage stability prediction model 61A, and the first predicted data 21_1 is output from the aggregation and storage stability prediction model 61A. Thus, in the technology of this disclosure, the predicted formulation information 18 and the measured formulation information 19 may be information relating to the formulation of the actually prepared solution 35. As a result, the technology of this disclosure can also be used to determine whether or not to continue stress testing of the actually prepared solution 35.

[0129] In the embodiments described above, three types of storage stability prediction models were exemplified: an agglutination storage stability prediction model, a temperature storage stability prediction model, and a time-degradation storage stability prediction model. However, the model is not limited to these. For example, there may be two types: an agglutination storage stability prediction model and a temperature storage stability prediction model, or two types: an agglutination storage stability prediction model and a time-degradation storage stability prediction model. Additional storage stability prediction models may be added to predict other storage stability, such as the storage stability of antibody 37 to additives.

[0130] In each of the above embodiments, one type of storage stability is predicted by one storage stability prediction model, but this is not limited to this. As an example, as shown in Figure 44, the storage stability prediction model 165 may include an aggregation / temperature storage stability prediction model 165AB that predicts both the storage stability against aggregation and the storage stability against temperature of the antibody 37, and a time-dependent degradation storage stability prediction model 165C that predicts the storage stability against degradation over time of the antibody 37. The aggregation / temperature storage stability prediction model 165AB is a model that integrates the aggregation storage stability prediction model 61A or 130A and the temperature storage stability prediction model 61B or 130B of each of the above embodiments. The aggregation / temperature storage stability prediction model 165AB simultaneously outputs first prediction data 21_1 showing the storage stability against aggregation of the antibody 37 and second prediction data 21_2 showing the storage stability against temperature of the antibody 37.

[0131] In the embodiments described above, the predicted formulation information 18 is exemplified by the types of additives 25 contained in the target solution and the hydrogen ion concentration 26 of the target solution itself, but is not limited to these. As an example, the predicted formulation information 170 shown in Figure 45 may include the types 171 and concentrations 172 of the buffer, the concentrations 173 of the additives, and the types 174 and concentrations 175 of the surfactants contained in the target solution. The same applies to the measured formulation information 19. Multiple types of buffers, additives, and surfactants may be used, as can be seen from the example additives L-arginine hydrochloride and refined sucrose. In that case, the type and concentration are registered for each of the multiple types.

[0132] Thus, the predictive formulation information only needs to include at least one of the following: the types of buffers, additives, and surfactants contained in the candidate preservation solution (171, 25, and 174), the concentrations of the buffers, additives, and surfactants contained in the candidate preservation solution (172, 173, and 175), and the hydrogen ion concentration of the candidate preservation solution itself (26). The molecular formulas of the buffers, additives, and surfactants may also be added to the predictive formulation information.

[0133] Furthermore, while SVP agglutination analysis data 40 and DSC analysis data 41 are given as examples of measurement data 20, the data is not limited to these. As an example, as shown in Figure 46, measurement data 180 may include dynamic light scattering (DLS) analysis data (hereinafter referred to as DLS analysis data) 181 of the antibody 37 in the prepared solution 35, and size exclusion chromatography (SEC) analysis data (hereinafter referred to as SEC analysis data) 182 of the antibody 37 in the prepared solution 35. DLS analysis data 181 represents the amount of nanoparticles of antibody 37 in the prepared solution 35, similar to SVP agglutination analysis data 40. SEC analysis data 182 represents the molecular weight distribution of antibody 37 in the prepared solution 35. DLS analysis data 181 and SEC analysis data 182 are used as measurement data to indicate the storage stability of antibody 37 against agglutination or degradation over time.

[0134] Thus, the measurement data only needs to include at least one of the following: SVP agglutination analysis data 40, DSC analysis data 41, DLS analysis data 181, and SEC analysis data 182.

[0135] Proteins are not limited to the example antibody 37. They may also include cytokines (interferon, interleukin, etc.), hormones (insulin, glucagon, follicle-stimulating hormone, erythropoietin, etc.), growth factors (IGF (Insulin-Like Growth Factor)-1, bFGF (Basic Fibroblast Growth Factor), etc.), blood coagulation factors (factor VII, factor VIII, factor IX, etc.), enzymes (lysosomal enzymes, DNA (deoxyribonucleic acid) degrading enzymes, etc.), Fc (Fragment Crystallizable) fusion proteins, receptors, albumin, and protein vaccines. Furthermore, antibodies may also include bispecific antibodies, antibody-drug conjugates, small molecule antibodies, and glycosylated antibodies.

[0136] The method of making the forecast data 21 available for operator viewing is not limited to displaying it on the display 13. The forecast data 21 may also be provided to the operator as a printed copy, or the forecast data 21 may be attached to an email and sent to the operator's mobile terminal.

[0137] The pharmaceutical support server 10 may be installed at each pharmaceutical facility, or it may be installed in a data center independent of the pharmaceutical facilities. Furthermore, the operator terminal 11 may handle some or all of the functions of the processing units 65-67 of the pharmaceutical support server 10.

[0138] The hardware configuration of the computer constituting the pharmaceutical support server 10 according to the technology disclosed herein can be modified in various ways. For example, the pharmaceutical support server 10 can be configured with multiple computers separated as hardware, in order to improve processing power and reliability. For example, the functions of the reception unit 65 and RW control unit 66 and the functions of the prediction unit 67 and distribution control unit 68 can be distributed among two computers. In this case, the pharmaceutical support server 10 is configured with two computers.

[0139] Thus, the hardware configuration of the pharmaceutical support server 10 can be appropriately changed according to the required performance, such as processing power, safety, and reliability. Furthermore, not only the hardware, but also application programs such as the operating program 60 can, of course, be duplicated or distributed and stored on multiple storage devices for the purpose of ensuring safety and reliability.

[0140] In each of the above embodiments, the hardware structure of the Processing Unit that performs various processes, such as the reception unit 65, RW control unit 66, prediction unit 67, distribution control unit 68, and feature quantity derivation unit 145, can be the following types of processors. As mentioned above, the types of processors include a CPU 52, which is a general-purpose processor that executes software (operation program 60) and functions as various processing units, as well as programmable logic devices (PLDs), such as FPGAs (Field Programmable Gate Arrays), which have circuit configurations that can be changed after manufacturing, and dedicated electrical circuits, such as ASICs (Application Specific Integrated Circuits), which have circuit configurations specifically designed to perform specific processes.

[0141] A single processing unit may consist of one of these various processors, or it may consist of 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). Alternatively, multiple processing units may be composed of a single processor.

[0142] Examples of configuring multiple processing units with a single processor include, firstly, a configuration where one or more CPUs and software combine to form a single processor, which then functions as multiple processing units, as exemplified by client and server computers. Secondly, a configuration using a processor that realizes the functions of the entire system, including multiple processing units, on a single IC (Integrated Circuit) chip, as exemplified by System-on-a-Chip (SoC). Thus, various processing units are configured, in terms of hardware structure, using one or more of the above-mentioned processors.

[0143] Furthermore, the hardware structure of these various processors can more specifically utilize electrical circuits, which are combinations of circuit elements such as semiconductor devices.

[0144] The technology of this disclosure can be appropriately combined with the various embodiments and / or variations described above. Furthermore, it is understood that various configurations can be adopted without departing from the spirit of the invention, and the invention is not limited to the embodiments described above. Moreover, the technology of this disclosure extends not only to programs but also to storage media for storing programs non-temporarily.

[0145] The descriptions and illustrations presented above are detailed explanations of the technical aspects of this disclosure and are merely examples of the technical aspects. For example, the above descriptions of the structure, function, operation, and effect are examples of the structure, function, operation, and effect of the technical aspects of this disclosure. Therefore, it goes without saying that you may delete unnecessary parts, add new elements, or replace elements in the descriptions and illustrations presented above, as long as you do not deviate from the essence of the technical aspects of this disclosure. Furthermore, in order to avoid confusion and facilitate understanding of the technical aspects of this disclosure, explanations of common technical knowledge and the like that do not require special explanation to enable the implementation of the technical aspects of this disclosure have been omitted from the descriptions and illustrations presented above.

[0146] In this specification, "A and / or B" is synonymous with "at least one of A and B." That is, "A and / or B" means that it may be A alone, or B alone, or a combination of A and B. Furthermore, in this specification, the same concept as "A and / or B" applies when expressing three or more things linked by "and / or."

[0147] All documents, patent applications, and technical standards described herein are incorporated by reference to the same extent as if each individual document, patent application, and technical standard were specifically and individually noted to be incorporated by reference.

Claims

1. Equipped with a processor, The aforementioned processor, A machine learning model that outputs predictive data showing the storage stability of candidate storage solutions for biopharmaceuticals at a future point in time, using multiple machine learning models applied to multiple types of storage stability, The prediction process, in which formulation information regarding the formulation of the candidate preservation solution to be predicted and measurement data obtained by actually measuring the storage stability of the prepared candidate preservation solution are input into the machine learning model and the machine learning model outputs the predicted data, is performed stepwise by multiple machine learning models. The prediction data obtained in the preceding prediction process is input to the machine learning model in the subsequent prediction process. Pharmaceutical manufacturing support device.

2. The pharmaceutical support device according to claim 1, wherein the plurality of machine learning models provided for each of the plurality of types of storage stability are models corresponding to at least two of the following: storage stability against aggregation of proteins contained in the biopharmaceutical, storage stability of the protein against temperature, and storage stability of the protein against degradation over time.

3. The aforementioned processor, The pharmaceutical support device according to claim 1, wherein the formulation information of the candidate storage solution actually prepared is also input to the machine learning model.

4. The machine learning model outputs the reliability of the prediction data along with the prediction data. The aforementioned processor, The pharmaceutical support device according to claim 1, wherein the confidence level obtained in the preceding prediction process is also input to the machine learning model in the subsequent prediction process.

5. The aforementioned processor, The pharmaceutical support device according to claim 1, wherein feature quantities derived based on protein information relating to the proteins contained in the biopharmaceutical are also input to the machine learning model.

6. The pharmaceutical support device according to claim 5, wherein the characteristic quantity includes at least one of the following: the solvent-exposed surface area of ​​the protein, the spatial aggregation tendency, the spatial charge map, and an index indicating the compatibility between the protein and the additive contained in the candidate preservation solution.

7. The pharmaceutical support device according to claim 1, wherein the measurement data is time-series data measured at at least two points in time.

8. The pharmaceutical support device according to claim 1, wherein the formulation information relating to the formulation of the candidate preservation solution to be predicted includes at least one of the following: the types of buffer, additives, and surfactants contained in the candidate preservation solution; the concentrations of the buffer, additives, and surfactants; and the hydrogen ion concentration of the candidate preservation solution itself.

9. The pharmaceutical support device according to claim 1, wherein the measurement data includes at least one of the following: aggregation analysis data of subvisible particles of the protein contained in the biopharmaceutical in the candidate preservation solution; dynamic light scattering analysis data of the protein in the candidate preservation solution; size exclusion chromatography analysis data of the protein in the candidate preservation solution; and differential scanning calorometry analysis data of the protein in the candidate preservation solution.

10. The pharmaceutical support device according to claim 1, wherein the protein contained in the biopharmaceutical is an antibody.

11. A machine learning model that outputs predictive data showing the storage stability of candidate storage solutions for biopharmaceuticals at a future point in time, using multiple machine learning models applied to multiple types of storage stability, The prediction process involves inputting formulation information regarding the formulation of the candidate preservation solution to be predicted, measurement data obtained by actually measuring the storage stability of the prepared candidate preservation solution, and outputting the predicted data from the machine learning model, and performing this prediction process stepwise using multiple machine learning models, and further, The prediction data obtained in the preceding prediction process is input into the machine learning model in the subsequent prediction process. A method for operating a pharmaceutical support device, including one.

12. A machine learning model that outputs predictive data showing the storage stability of candidate storage solutions for biopharmaceuticals at a future point in time, using multiple machine learning models applied to multiple types of storage stability, The prediction process involves inputting formulation information regarding the formulation of the candidate preservation solution to be predicted, measurement data obtained by actually measuring the storage stability of the prepared candidate preservation solution, and outputting the predicted data from the machine learning model, and performing this prediction process stepwise using multiple machine learning models, and further, The prediction data obtained in the preceding prediction process is input into the machine learning model in the subsequent prediction process. An operating program for a pharmaceutical support device that causes a computer to perform a process including [specific details].

Images