Estimation device
The estimation device predicts biopharmaceutical quality during production, addressing the challenge of lengthy production times and post-purification quality assessment, thereby reducing costs through timely process adjustments.
Patent Information
- Application Number
- JP2022579518
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Priority Date
- 2021-02-08
- Filing Date
- 2022-01-31
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2042-01-31
AI Technical Summary
Biopharmaceutical production is lengthy, typically taking several weeks to months, and quality assessment is only possible after the purification process, making it difficult to ensure uniformity and increasing costs.
An estimation device that uses a prediction model to predict the quality of biopharmaceuticals during the culture process by analyzing measurement data from culture containers, allowing for adjustments to be made before purification.
Enables quality prediction during the culture process, enabling protocol revisions and reducing manufacturing costs by allowing for timely adjustments.
Smart Images

Figure 0007589759000001 
Figure 0007589759000002 
Figure 0007589759000003
Abstract
Description
[Technical field]
[0001] This disclosure relates to a presumptive Place This is related to. [Background technology]
[0002] In recent years, research into biopharmaceuticals produced by culturing cells has been progressing. Japanese Patent No. 4496086 (Patent Document 1) discloses a cell culture method for culturing host cells to produce a target protein.
[0003] Biopharmaceuticals are produced using cellular metabolism, which is difficult to control artificially, and the drugs used may contain components derived from living cells. Therefore, even if they are manufactured according to a set protocol, it is difficult to ensure that the quality of the produced biopharmaceuticals is consistent. Therefore, generally, the drug substance of the biopharmaceutical obtained by cell culture is subjected to characteristic analysis to confirm the quality of the drug substance. [Prior art documents] [Patent documents]
[0004] [Patent Document 1] Patent No. 4496086 Summary of the Invention [Problem to be solved by the invention]
[0005] The production of biopharmaceuticals takes a very long time, from several weeks to several months, including the cultivation, purification, and processing steps. In addition, the drug substance that is the subject of characterization is obtained in the purification step that follows the completion of the cultivation step. In other words, since the results of characterization cannot be obtained at the cultivation step, it is only once the purification step is completed that it can be determined whether the desired drug substance has been obtained by cultivation.
[0006] One object of the present disclosure is to predict, at a mid-stage of culture, the quality of a drug substance obtained by proceeding with the culture. 。 [Means for solving the problem]
[0007] The estimation device of the present disclosure includes a receiving unit that receives input of measurement data, a prediction unit that generates quality prediction data indicating the quality of a drug substance by inputting the measurement data received by the receiving unit into a prediction model, and an output unit that outputs the quality prediction data generated by the prediction unit. The measurement data includes a measurement result obtained by measuring at least one substance in a culture vessel containing cells and the culture medium at least one timing after a predetermined period of time has elapsed since the cells were seeded in the culture medium. The prediction model is a model for predicting the quality of a drug substance of a biopharmaceutical produced by culturing cells. Effect of the Invention
[0013] According to the present disclosure, it is possible to predict the quality of a drug substance obtained by proceeding with the culture during the culture stage. .So Therefore, based on the predicted quality of the drug substance, users can take various measures, such as revising the protocol, extending the culture time, or even stopping the culture, at the culture stage prior to the purification process to obtain the drug substance, thereby ultimately reducing the manufacturing costs of biopharmaceuticals. [Brief description of the drawings]
[0014] [Figure 1] FIG. 1 is a diagram illustrating an overall configuration of a prediction system according to a first embodiment. [Diagram 2] FIG. 2 is a block diagram showing a hardware configuration of the estimation device. [Diagram 3] FIG. 2 is a block diagram showing a hardware configuration of the learning device. [Figure 4] FIG. 11 is a diagram for explaining measurement timing. [Diagram 5] FIG. 11 is a diagram illustrating an example of measurement data. [Figure 6] 2 is a block diagram showing an example of a configuration of a prediction unit; FIG. [Figure 7] FIG. 11 is a diagram illustrating an example of an output result output by an output unit. [Figure 8] 4 is a flowchart showing the flow of the estimation method according to the first embodiment. [Figure 9] 4 is a flowchart showing the flow of a learning method according to the first embodiment. [Figure 10] FIG. 13 is a block diagram showing a configuration of a learning device according to a modified example. [Figure 11] FIG. 13 is a diagram illustrating a modified example of the model generating unit. [Figure 12] FIG. 13 is a block diagram showing a configuration of an estimation device according to a modified example. [Figure 13] FIG. 13 is a diagram illustrating a modified example of a prediction model. [Figure 14] FIG. 11 is a diagram illustrating a schematic diagram of an overall configuration of a prediction system according to a second embodiment. [Figure 15] FIG. 2 is a block diagram showing a hardware configuration of the optimization device. [Figure 16] FIG. 1 is a block diagram showing an example of a configuration of an optimization device. [Figure 17] 13 is a flowchart showing a process flow of the optimization device. [Figure 18] FIG. 11 is a diagram showing a modified example of the prediction system according to the second embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0015] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the drawings. In the drawings, the same or corresponding parts are designated by the same reference characters and their description will not be repeated.
[0016] Embodiment 1 [Overall configuration of the prediction system] FIG. 1 is a diagram showing a schematic diagram of an overall configuration of a prediction system according to the first embodiment. 1is a system for predicting the quality of drug substances of biopharmaceuticals produced by cell culture. In the present embodiment 1, "biopharmaceuticals" refer to pharmaceuticals produced using cells, such as antibody pharmaceuticals and vaccines. In the present embodiment 1, biopharmaceuticals may also include the cells themselves used in regenerative medicine. In the present embodiment 1, "drug substances" refer to target substances obtained by cell culture, such as those obtained through a purification process for extracting target components after a culture process for culturing cells. Prediction system SYS 1 includes a measurement device 100, an estimation device 200, a learning device 300, and a server 400.
[0017] The measuring device 100 measures a substance in a culture vessel 10 including a cell 1 and a culture medium 2 as a measurement target. The substance to be measured is preferably a substance that changes during the process of culturing the cell 1, such as the cell 1 itself, a metabolite of the cell 1, or a nutrient of the cell 1. A user can select any measuring device 100 according to the measurement target. For example, when the measurement target is the cell 1, the user can select a microscope and software for analyzing an image obtained by the microscope as the measuring device 100 in order to obtain measurement results such as the number of cells in a culture solution composed of the cell 1 and the culture medium 2 in the culture vessel 10, a cell viability rate, and a shape of the cell (roundness, major axis length, minor axis length, etc.). When the measurement target is a metabolite or nutrient of the cell 1, the user can select a liquid chromatography mass spectrometer (LC-MS), an ICP (Inductively Coupled Plasma) mass spectrometer, etc. as the measuring device 100 in order to obtain measurement results such as the concentration of a metabolite in the culture solution, the concentration of a nutrient, etc.
[0018] The estimation device 200 predicts the quality of the drug substance using measurement data 510 including the measurement results for the measurement target after a predetermined period of time has elapsed since the cells 1 were seeded on the culture medium 2. For example, the measurement data 510 is obtained by sampling the culture medium in the culture container 10 at least once after the predetermined period of time has elapsed since the cells 1 were seeded on the culture medium 2, and measuring a substance in the sampled culture medium using the measurement device 100.
[0019] The estimation device 200 includes a receiving unit 210, a prediction unit 220, and an output unit 230, as an example of a software configuration realized by the processor 21 (see FIG. 2) executing an estimation program 560 (see FIG. 2). The receiving unit 210 receives input of measurement data 510.
[0020] The prediction unit 220 generates quality prediction data 540 indicating the quality of the drug substance by inputting the measurement data 510 accepted by the acceptance unit 210 into the prediction model 420. The prediction model 420 is a model for predicting the quality of the drug substance of a biopharmaceutical produced by culturing cells 1, and receives the input of the measurement data 510 and outputs a prediction result indicating the quality of the drug substance. The prediction model 420 is generated by the model generation unit 320 of the learning device 300 executing a supervised learning process.
[0021] The quality prediction data 540 may be the prediction result itself output from the prediction model 420, or may be generated based on the prediction result output from the prediction model 420. The prediction result output from the prediction model 420 is a result corresponding to the quality data 520 used as the correct answer data when the model generating unit 320 is generated. For example, when the quality data 520 is data itself indicating the analysis result obtained by performing an arbitrary analysis on the drug substance 3, the prediction model 420 outputs the analysis result predicted as the prediction result. When the quality data 520 is data indicating the similarity with the target drug substance obtained based on the analysis result, the prediction model 420 outputs the similarity predicted as the prediction result.
[0022] The output section 230 outputs the quality prediction data 540 generated by the prediction section 220 to an arbitrary output destination such as a display, a printer, or a server.
[0023] The learning device 300 generates a prediction model 420 using learning data 530. The learning device 300 includes a receiving unit 310 and a model generating unit 320, which are an example of a software configuration realized by the processor 31 (see FIG. 3) executing a learning program 580 (see FIG. 3).
[0024] The receiving unit 310 receives an input of learning data 530. The learning data 530 includes time-series data 512 and quality data 520.
[0025] The time series data 512 is a type of measurement data 510, and is obtained by measuring at least one substance in the culture vessel 10 at multiple timings (timing t1, . . . timing tx) after the cells 1 are seeded in the culture medium 2. The time series data 512 is data indicating the change over time of a measurement target, the substance in the culture vessel 10 including the cells 1 and the culture medium 2. The time series data 512 is obtained by collecting culture fluid in the culture vessel 10 at multiple timings (timing t1, . . . timing tx) after the cells 1 are seeded in the culture medium 2, and measuring each of the collected culture fluids using the measuring device 100. For example, the time series data 512 indicates the change over time of the number of cells, the change over time of the viability rate, the change over time of the shape of the cells, the change over time of the concentration of nutrients, the change over time of the concentration of metabolites, and the like.
[0026] The quality data 520 is obtained by analyzing the drug substance 3 of the biopharmaceutical produced from the cell 1. The quality data 520 is data obtained through a purification step and an analysis step after culturing the cell 1, and is data indicating the quality of the drug substance 3 obtained by performing an arbitrary analysis in the analysis step on the drug substance 3 obtained through the purification step. The analysis performed in the analysis step is selected arbitrarily according to the type of the drug substance 3. The quality of the drug substance 3 is evaluated from one or more viewpoints, for example, from the viewpoint of physical properties, chemical properties, biological activity, immunochemical properties, etc. The quality data 520 may be data indicating an analysis result obtained by performing an arbitrary analysis on the drug substance 3, or may be data indicating the similarity to the target drug substance obtained based on the analysis result.
[0027] For example, when the drug substance 3 is used for an antibody drug, amino acid composition analysis, amino acid sequence analysis, peptide mapping analysis, analysis of the cross-linking position of disulfide bonds, analysis of glycosylation structure, etc. are performed on the drug substance 3 in the analysis step. In this case, as an example, data indicating the degree of similarity of the amino acid composition to the target drug substance as a result of the amino acid composition analysis may be included in the quality data 520.
[0028] The model generation unit 320 executes a learning process using the quality data 520 included in the learning data 530 as correct answer data to generate a prediction model 420. For example, the model generation unit 320 inputs the time-series data 512 included in the learning data 530 to the prediction model 420, obtains an error between the output quality prediction result and the quality data 520 which is correct answer data, and optimizes the prediction model 420 so as to reduce this error.
[0029] The server 400 is an example of a storage device for storing the prediction model 420. The prediction model 420 may be stored in the storage 23 (see FIG. 2) of the estimation device 200, or in the storage 33 (see FIG. 3) of the learning device 300. The estimation device 200 and the learning device 300 may be realized by a single device. When the estimation device 200 and the learning device 300 are realized by a single device, the reception unit 210 and the reception unit 310 may be realized as a single reception unit.
[0030] In this embodiment, the culture conditions are assumed to be the same when generating the prediction model 420 and when generating the quality prediction data 540 using the generated prediction model 420. In this embodiment, the "culture conditions" being the same means that the culture is performed according to a common protocol.
[0031] Even if the cell culture is performed according to a common protocol, the drug substance obtained by cell culture may exhibit different properties from the intended drug substance because the cells themselves are different, and the drug used may contain components derived from living cells, etc. Therefore, in order to ensure the quality of biopharmaceuticals, it is common to perform characteristic analysis on the drug substance obtained by cell culture.
[0032] The drug substance 3 is obtained by the activity of the cell 1. Therefore, it is expected that there is a correlation between the characteristics of the drug substance 3 and the activity state of the cell 1. In addition, it is expected that the substance in the culture vessel 10 containing the cell 1 will show different changes over time when the activities of the cell 1, such as proliferation and metabolism, are active and when they are not active. Therefore, it is expected that there is a correlation between the state of the substance in the culture vessel 10 and the activity state of the cell 1. In other words, it is expected that there is an indirect correlation between the state of the substance in the culture vessel 10 and the characteristics of the drug substance 3.
[0033] The learning device 300 generates the prediction model 420 by executing a learning process using learning data 530 including time-series data 512 indicating the change over time of the substance in the culture vessel 10 and quality data 520 indicating the quality of the drug substance 3. In other words, the learning device 300 generates the prediction model 420 by learning the association between the change over time of the substance in the culture vessel 10 and the quality of the drug substance 3.
[0034] A user repeatedly performs a culture test including a culture step and a purification step while changing culture conditions. Quality data 520 is obtained for each culture test. The measuring device 100 outputs time-series data 512 for each culture test. By repeatedly performing the culture test, a large amount of learning data 530 corresponding to the number of culture tests is collected. The learning device 300 repeatedly optimizes the prediction model 420 based on the large amount of learning data.
[0035] In the present embodiment 1, the measurement data 510 is obtained by sampling the culture solution in the culture container 10 at least once a predetermined period of time has elapsed since the cells 1 were seeded in the culture medium 2, and measuring substances in the sampled culture solution using the measuring device 100. That is, the measurement data 510 indicates the state of substances in the culture solution at least once a predetermined period of time has elapsed since the cells 1 were seeded in the culture medium 2.
[0036] Since the measurement data 510 indicates the state of the substance after a predetermined period of time has elapsed since the cells 1 were seeded on the culture medium 2, by inputting the measurement data 510 into a prediction model 420 that has been generated by learning the relationship between the change in the substance over time and the quality of the drug substance 3, a prediction result can be obtained that indicates the quality of the drug substance 3 obtained after the culture has been continued.
[0037] As described above, the estimation device 200 according to the first embodiment generates quality prediction data 540 indicating the quality of the drug substance obtained by proceeding with the culture of the cells 1 in the culture vessel 10 based on the measurement data 510 including the measurement results obtained by measuring the measurement target in the culture vessel 10. Therefore, the quality of the drug substance obtained by proceeding with the culture can be predicted at a stage during the culture. In addition, the learning device 300 can generate a prediction model for predicting the quality of the drug substance obtained by proceeding with the culture at a stage during the culture. Based on the quality prediction data 540, the user can take various measures such as reviewing the protocol at the stage of the culture process before the purification process for obtaining the drug substance, extending the culture time, or stopping the culture, and as a result, the manufacturing cost of the biopharmaceutical can be reduced.
[0038] Furthermore, the learning device 300 according to the first embodiment can generate a prediction model 420 that has learned the association between the change over time of the substance in the culture vessel 10 and the quality of the drug substance 3 by executing a learning process using learning data 530 that includes time-series data 512 indicating the change over time of the substance in the culture vessel 10 and the quality of the drug substance 3.
[0039] [Hardware configuration of the estimation device] 2 is a block diagram showing a hardware configuration of the estimation device 200. The estimation device 200 includes a processor 21, a main memory 22, a storage 23, a communication interface (I / F) 24, a Universal Serial Bus (USB) interface (I / F) 25, an input unit 26, and a display unit 27. These components are connected via a processor bus 28.
[0040] The processor 21 is composed of a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), etc., and reads out a program (e.g., estimation program 560) stored in the storage 23, expands it in the main memory 22, and executes it. The processor 21 executes the program to perform a series of processes for predicting the quality of the drug substance.
[0041] The main memory 22 is configured with a volatile storage device such as a random access memory (RAM) or a dynamic RAM (DRAM). The storage 23 is configured with a non-volatile storage device such as a hard disk drive (HDD) or a solid state drive (SSD).
[0042] The storage 23 stores measurement data 510 sent via the USB I / F 25, quality prediction data 540 generated based on the measurement data 510, and an estimation program 560 for performing a series of processes to predict the quality of the active pharmaceutical ingredient.
[0043] The communication I / F 24 exchanges signals with the server 400 in which the prediction model 420 is stored, using wired or wireless communication.
[0044] A USB memory (not shown) is detachably attached to the USB I / F 25, and measurement data 510 stored in the USB memory is read out. Note that the measurement data 510 may be sent from the measurement device 100 to the estimation device 200 by connecting the measurement device 100 and the estimation device 200 so that they can communicate with each other.
[0045] The input unit 26 receives user operations and is typically configured with a touch panel, a keyboard, a mouse, etc. The display unit 27 is an example of an output destination of the quality prediction data 540, and is configured with a liquid crystal panel or the like capable of displaying images.
[0046] [Hardware configuration of the learning device] 3 is a block diagram showing the hardware configuration of the learning device 300. The learning device 300 includes a processor 31, a main memory 32, a storage 33, a communication I / F 34, a USB I / F 35, an input unit 36, and a display unit 37. These components are connected via a processor bus 38.
[0047] The processor 31 is composed of a CPU, a GPU, etc., and reads a program (for example, a learning program 580) stored in the storage 33, expands it in the main memory 32, and executes it. The processor 31 performs a series of processes for generating the prediction model 420 by executing the program.
[0048] The main memory 32 is configured with a volatile storage device such as a RAM, a DRAM, etc. The storage 33 is configured with a non-volatile storage device such as a HDD, an SSD, etc.
[0049] The storage 33 stores the learning data 530 sent via the USB I / F 35 and a learning program 580 for performing a series of processes for generating the prediction model 420.
[0050] The communication I / F 34 exchanges signals with the server 400 in which the prediction model 420 is stored, using wired or wireless communication.
[0051] A USB memory (not shown) is detachably attached to the USB I / F 35, and the learning data 530 stored in the USB memory is read out. Note that the learning device 300 may be communicably connected to an information processing device that receives input of the measurement data 510 and the quality data 520 and generates the learning data 530 by associating the measurement data 510 with the quality data 520, so that the learning data 530 may be sent from the information processing device to the estimation device 200.
[0052] The input unit 36 receives user operations and is typically configured with a touch panel, a keyboard, a mouse, etc. The display unit 37 is configured with a liquid crystal panel or the like capable of displaying images.
[0053] [An example of measurement data 510 used for quality prediction] The measurement data 510 received by the estimation device 200 to predict the quality of the drug substance 3 includes a measurement result obtained by measuring the measurement target at least one timing after a predetermined period of time has elapsed since the cells 1 were seeded on the culture medium 2. The measurement data 510 used to predict the quality of the drug substance 3 may be time-series data 512 showing a change in the measurement target over time, which is a measurement result obtained by measuring the measurement target at multiple timings after the cells 1 are seeded on the culture medium 2.
[0054] FIG. 4 is a diagram for explaining the timing of measurement. In FIG. 4, a typical cell growth curve is shown. As shown in FIG. 4, generally, immediately after the cells 1 are seeded in the medium 2, they do not grow immediately, but go through a lag phase (also called a lag phase) and then enter a logarithmic growth phase in which they grow. Thereafter, as nutrients decrease or waste products (metabolites) accumulate, the logarithmic growth phase ends and the cell enters a stationary phase (also called a static phase). In the stationary phase, a balance is maintained between the number of growing cells 1 and the number of dying cells 1. After the stationary phase continues for a while, the number of dying cells 1 increases, and the cell enters a death phase (also called a decline phase).
[0055] The measurement data 510 may include a measurement result obtained by measuring the measurement target at timing t0 when the culture of the cell 1 is started. The start of the culture of the cell 1 may include a timing before the cell 1 is seeded in the culture medium 2 and a timing immediately after the cell 1 is seeded in the culture medium 2. The measurement result at timing t0 indicates the initial state inside the culture vessel 10. When the measurement data 510 includes a measurement result indicating the initial state inside the culture vessel 10, the measurement data 510 indicates a change from the initial state.
[0056] Furthermore, the measurement data 510 may include a measurement result obtained by measuring the measurement target between timing t1 and timing t2, which are in the logarithmic growth phase. That is, at least one timing after a predetermined period of time has elapsed since the cells 1 were seeded in the culture medium 2 is a timing between timing t1 and timing t2, which are in the logarithmic growth phase. In the logarithmic growth phase, the number of cells 1 fluctuates greatly, and therefore the state of the substance in the culture vessel 10 fluctuates greatly. By including a result of measuring the measurement target at a timing when such a large state change occurs in the time-series data 512, the estimation device 200 can predict the quality of the drug substance 3 using data that more accurately indicates the state change of the cells 1.
[0057] Furthermore, the measurement data 510 may include a measurement result obtained by measuring the measurement target during the stationary phase between timing t2 and timing t3. That is, at least one timing after a predetermined period of time has elapsed since the cells 1 were seeded in the culture medium 2 is a timing between timing t2 and timing t3, which is the stationary phase. In the stationary phase, a balance is maintained between the number of proliferating cells 1 and the number of dying cells 1, and the number of cells 1 is at a maximum. In this way, by including in the time-series data 512 a result of measuring the measurement target at the timing when the number of cells 1 is at a maximum, the estimation device 200 can predict the quality of the drug substance 3 using data showing the state of the substance in the culture vessel 10 when the number of cells 1 is at a maximum.
[0058] The measurement data 510 (time series data 512) includes the measurement results measured during the death phase after timing t3. When cells die, enzymes in the cells are exposed to the medium, which may affect the quality of the drug substance. In this case, the measurement results measured during the cell death phase may be used in machine learning as an example of poor quality data.
[0059] Fig. 5 is a diagram showing an example of the measurement data 510. Note that, in the example shown in Fig. 5, time series data 512a to 512f measured at multiple timings after the cell 1 is seeded on the culture medium 2 are shown as an example, but each of the time series data 512a to 512f may include a measurement result obtained by measuring the measurement target at least at one timing after a predetermined period has elapsed after the cell 1 is seeded on the culture medium 2.
[0060] The measurement data 510 includes, as measurement results obtained by measuring the cells 1, time series data 512a of the cell count, time series data 512b of the cell viability, time series data 512c of the cell shape, and time series data 512d of the microscope image, for example.
[0061] The measurement data 510 may be, for example, time-series data 512 of metabolites as a measurement result obtained by measuring metabolites produced by the metabolism of the cell 1. e The metabolites include at least one of the metabolites secreted from the cell 1 and the metabolites contained in the cell 1 itself. nothing .
[0062] The measurement data 510 also includes, for example, time series data 512f of medium components as a measurement result obtained by measuring nutrients ingested by the cell 1. When the cell 1 metabolizes a component common to the components contained in the medium, the time series data 512f of the medium components can be time series data of metabolites.
[0063] That is, the substance to be measured is at least one of nutrients ingested by the cell 1, metabolites produced by the metabolism of the cell 1, and the cell 1. The measurement data 510 may include information such as the pH and temperature of the culture solution in the culture container 10.
[0064] 4 and 5, the measurement data 510 accepted by the estimation device 200 has been described. Note that the time-series data 512 included in the learning data 530 accepted by the learning device 300 may also include multiple types of time-series data, similar to the measurement data 510 accepted by the estimation device 200.
[0065] For example, the time series data 512 included in the learning data 530 may indicate changes over time in at least one of the nutrients ingested by cell 1, metabolites produced by the metabolism of cell 1, and substances of cell 1.
[0066] More specifically, the time series data 512 included in the learning data 530 includes time series data of cell count, time series data of cell viability, time series data of cell shape, or time series data of microscope images, which are obtained by measuring cells. The time series data 512 included in the learning data 530 also includes time series data of metabolites obtained by measuring metabolites. The time series data 512 included in the learning data 530 also includes time series data of medium components obtained by measuring nutrients. The time series data 512 included in the learning data 530 may include information such as the pH and temperature of the culture solution in the culture vessel 10.
[0067] Furthermore, the time-series data 512 used to generate the prediction model 420 and the measurement data 510 input to the prediction model 420 when generating the quality prediction data 540 using the prediction model 420 each contain at least measurement results obtained by measuring a common substance. In other words, when time-series data showing the change over time in the number of cells is used to generate the prediction model 420, the measurement data 510 input to the prediction model 420 contains at least measurement results showing the number of cells.
[0068] [Configuration of the prediction unit 220 of the estimation device 200] 6 is a block diagram showing an example of the configuration of the prediction unit 220. The prediction unit 220 includes an estimation processing unit 222 and a determination unit 224. The estimation processing unit 222 inputs the measurement data 510 accepted by the acceptance unit 210 to the prediction model 420 to obtain a prediction result. The prediction result is, for example, a predicted value 542.
[0069] In general, the quality of the drug substance 3 is evaluated from one or more viewpoints. Thus, the prediction model 420 may be configured to output a predicted value 542 when evaluated from a specific viewpoint as a prediction result. As an example, the predicted value 542 is an estimated value of a result obtained when the drug substance 3 obtained from cells in culture is analyzed from a specific viewpoint, or a similarity indicating the degree to which the estimated value obtained is similar to the result of the drug substance of interest.
[0070] Perspectives for evaluating the quality of the drug substance 3 include, for example, physical properties, chemical properties, biological activity, and immunochemical properties. The prediction model 420 may be configured to output a plurality of predicted values obtained when the drug substance 3 is evaluated from each of a plurality of perspectives. For example, the prediction model 420 may output a first predicted value obtained when the drug substance 3 is evaluated from a first perspective (e.g., physical properties), and a second predicted value obtained when the drug substance 3 is evaluated from a second perspective (e.g., biological activity).
[0071] In addition, prediction model 420 may be configured to include a first prediction model for outputting a first predicted value when the quality of the drug substance is evaluated from a first perspective, and a second prediction model for outputting a second predicted value when the quality of the drug substance is evaluated from a second perspective.
[0072] The judgment unit 224 judges the quality of the drug substance 3 based on the predicted value 542 and generates a judgment result 544. The judgment result 544 may be a two-level evaluation result (e.g., good / bad), a multi-level evaluation result (e.g., A, B, C, etc.), or a score-based evaluation result (e.g., 50 points, 95 points, etc.).
[0073] When the prediction model 420 is configured to output a plurality of predicted values obtained when the drug substance 3 is evaluated from each of a plurality of viewpoints, the determination unit 224 determines the quality of the drug substance 3 from the plurality of predicted values and generates a determination result 544. In this case, the determination unit 224 may weight each of the plurality of predicted values according to the importance of the viewpoint.
[0074] The quality prediction data 540 includes a predicted value 542 generated by the estimation processing unit 222 and a determination result 544 generated by the determination unit 224. For example, if the prediction model 420 is configured to output a plurality of predicted values obtained when the drug substance 3 is evaluated from a plurality of perspectives, the quality prediction data 540 will include a plurality of predicted values (a first predicted value, a second predicted value, etc.).
[0075] The quality prediction data 540 includes a plurality of predicted values obtained when the drug substance 3 is evaluated from a plurality of perspectives, so that the quality of the drug substance can be evaluated from various perspectives.
[0076] Furthermore, since the quality prediction data 540 includes the judgment result 544, the prediction result for the quality of the drug substance can be shown in an easier-to-understand manner.
[0077] The output section 230 outputs the quality prediction data 540. FIG.
[0078] As shown in FIG. 7, the output unit 230 may output a determination result 544 together with a predicted value 542 obtained for each viewpoint.
[0079] It should be noted that the prediction unit 220 does not necessarily have to include the determination unit 224 that comprehensively determines the quality of the drug substance.
[0080] [Estimation method] 8 is a flowchart showing the flow of the estimation method according to the present embodiment 1. The estimation method includes a culture step (S100) of culturing cells, a measurement step (S200) of measuring a substance in a culture vessel, a prediction step (S300) of predicting the quality of the drug substance 3 based on measurement data 510 including the measurement results, and an output step (S400) of outputting quality prediction data 540 obtained in the prediction step.
[0081] The culture step (S100) includes a step (S110) of seeding the cells 1 in a medium 2. The culture step (S100) may also include a step (S112) of replacing the medium 2 and a step (S114) of adding a reagent, depending on the culture conditions (a predetermined protocol). The medium 2 may be liquid or solid.
[0082] The measuring step (S200) may be performed once or a plurality of times during the culturing step (S100). The measuring step (S200) may be performed at any timing, such as when the step (S110) of seeding the cell 1 in the medium 2 is performed, when the step (S112) of replacing the medium 2 is performed, or when the step (S114) of adding a reagent is performed.
[0083] That is, the measurement timing may be the timing of the step (S110) of seeding the cells 1 in the medium 2, the timing of the step (S112) of replacing the medium 2, the timing of the step (S114) of adding a reagent, etc. By matching the timing of the operations occurring in the culture step (S100) with the timing of the measurement in this way, contamination can be prevented. As described with reference to FIG. 4, the measurement timing may be during the logarithmic growth phase, the stationary phase, or the death phase.
[0084] By performing one or more measurement steps (S200) during the culture step (S100), the substance in the culture container 10 is measured at least once a predetermined period of time after the cells 1 are seeded in the culture medium 2, and measurement results are obtained.
[0085] The culturing step (S100) and the measuring step (S200) may be performed manually or may be automated or semi-automated by machine.
[0086] The prediction step (S300) includes a step (S310) of accepting input of measurement data 510, a step (S312) of inputting the measurement data 510 to the prediction model 420, a step (S314) of acquiring a prediction result from the prediction model 420, and a step (S316) of generating quality prediction data 540. That is, in the prediction step (S300), the quality prediction data 540 is generated by inputting the measurement data 510 including the measurement result obtained in the measurement step (S200) to the prediction model 420. Note that S310 to S316 and S400 are specifically performed by the estimation device 200.
[0087] [Learning method] 9 is a flowchart showing the flow of the learning method according to the first embodiment. The learning method includes a culture step (S10) of culturing cells, a measurement step (S20) of measuring substances in a culture vessel, a purification step (S30) of obtaining an active pharmaceutical ingredient, an analysis step (S40) of analyzing the quality of the active pharmaceutical ingredient, and a learning step (S50) of executing a learning process using learning data 530 to generate a prediction model 420. S5 0).
[0088] The culturing step (S10) is common to the culturing step (S100) included in the estimation method. Specifically, the culturing step (S10) includes a step (S11) of seeding the cells 1 in a medium 2. The culturing step (S10) may also include a step (S12) of replacing the medium 2 and a step (S13) of adding a reagent, depending on the culture conditions (predetermined protocol). The medium 2 may be liquid or solid.
[0089] The measurement step (S20) is performed multiple times during the incubation step (S10). This allows for the acquisition of time-series data 512 that indicates the change over time of the measurement target. 0) is performed at any timing, such as when the step (S11) of seeding cells 1 onto culture medium 2 is performed, when the step (S12) of replacing culture medium 2 is performed, and when the step (S13) of adding a reagent is performed.
[0090] That is, the measurement timing is the timing of the step of seeding the cells 1 on the medium 2 (S11), the step of replacing the medium 2 ( S1 The timing of performing the step of adding the reagent (S 13 In this way, the incubation step (S1 0) By matching the timing of the measurement with the timing of the operations performed in the above, contamination can be prevented. As described with reference to FIG. 4, the measurement may be performed during the logarithmic growth phase, the stationary phase, or the death phase.
[0091] Measurement process (S2 0) The culture process (S1 0) By carrying out the measurement multiple times during the culture, the substance in the culture vessel 10 is measured at multiple times after the cells 1 are seeded in the medium 2, and measurement results are obtained.
[0092] The culturing step (S10) and the measuring step (S20) may be performed manually, or may be automated or semi-automated by machine.
[0093] The purification step (S30) is a task of extracting only the target component from the culture solution in the culture vessel 10. Through the purification step (S30), it is possible to obtain a drug substance containing the target component at a high purity.
[0094] In the analysis step (S40), one or more types of analyses are performed to evaluate the drug substance from one or more perspectives, and quality data 520 is obtained. When multiple types of analyses are performed, quality data 520 may include multiple types of analysis results (analysis results).
[0095] The learning step (S50) includes a step (S51) of accepting input of learning data 530, a step (S52) of performing a learning process using the learning data 530, and a step (S53) of outputting the prediction model 420 generated by the learning process. In the first embodiment, the prediction model 420 is output to the server 400. The prediction model 420 may be stored in the storage 33 of the learning device 300 without being output. S51 to S53 are specifically performed by the learning device 300.
[0096] <Modification> [Modifications of the learning device] In the above-mentioned first embodiment, the learning device 300 acquires the learning data 530 from outside. The learning device may have a function of generating the learning data 530. Furthermore, in the above-mentioned first embodiment, the measurement timing for obtaining the measurement data 510 accepted by the estimation device 200 has been described with reference to FIG. 4. The proliferation curve of the cell 1 differs depending on the type of the cell 1. Therefore, the learning device may have a function of identifying the measurement timing at the time of prediction for generating the quality prediction data 540.
[0097] Fig. 10 is a block diagram showing a configuration of a learning device according to a modified example. Fig. 10 shows a configuration that is not included in the learning device 300 according to the first embodiment. The learning device 300a differs from the learning device 300 in that it further includes a learning data generating unit 312 and a timing identifying unit 330.
[0098] The learning data generation unit 312 generates learning data 530 based on the time-series data 512 and the quality data 520. Although not shown, the learning data generation unit 312 transmits the generated learning data 530 to the reception unit 310. The reception unit 310 receives an input of the transmitted learning data 530.
[0099] The timing specifying unit 330 specifies the measurement timing at the time of prediction for generating the quality prediction data 540, based on the time series data 512a of the cell count. More specifically, the timing specifying unit 330 specifies the lag phase, logarithmic growth phase, stationary phase, and death phase based on the time series data 512a (see FIG. 4), and generates information capable of specifying the timings t1 to t3. The information capable of specifying the timings t1 to t3 includes, for example, the time from when the cells 1 are seeded in the culture medium 2. The user can determine the measurement timing at the time of prediction, based on the information capable of specifying the timings t1 to t3.
[0100] [Culture conditions] In the above-mentioned first embodiment, the culture conditions are the same when generating the prediction model 420 and when generating the quality prediction data 540 using the generated prediction model 420. Note that the culture conditions do not have to be common when generating the prediction model 420 and when generating the quality prediction data 540 using the generated prediction model 420. The learning method and estimation method in this case will be described with reference to Figs. 11 to 13.
[0101] 11 is a diagram showing a modified example of the model generating unit 320a according to the modified example, which generates a prediction model 420a based on a plurality of types of learning data 530a and 530b.
[0102] Each of the learning data 530a, 530b includes condition data 550. The condition data 550 is information indicating the culture conditions. For example, by changing the culture conditions and repeating the culture step, the measurement step, the purification step, and the analysis step (see FIG. 9), the learning data 530a, 530b, etc., generated under different culture conditions can be obtained.
[0103] Different culture conditions may include different components of medium 2, different types of chemicals used, such as different types of reagents added, as well as different temperature conditions and different timing of adding reagents.
[0104] The condition data 550 may include information on the chemicals used, temperature conditions, timing of adding reagents, etc. Note that the condition data 550 may not be information on a specific culture technique, but may be information that merely indicates that the learning data 530a, 530b were obtained under culture conditions different from each other (for example, condition A, condition B, etc.).
[0105] The model generation unit 320a generates a prediction model 420a by learning the correlation between the change over time of the substance in the culture vessel 10 and the quality of the active pharmaceutical ingredient 3, as well as the correlation between differences in culture conditions and the change over time of the substance in the culture vessel 10, or the correlation between differences in culture conditions and the quality of the active pharmaceutical ingredient 3.
[0106] Fig. 12 is a block diagram showing a configuration of an estimation device according to a modified example. The configuration of the output unit according to the first embodiment is omitted in Fig. 12. The estimation device 200a differs from the estimation device 200 according to the first embodiment in that it includes a reception unit 210a instead of the reception unit 210 and a prediction unit 220a instead of the prediction unit 220.
[0107] The receiving unit 210a receives input of condition data 550 indicating culture conditions in addition to the measurement data 510. The prediction unit 220a inputs the condition data 550 in addition to the measurement data 510 to the prediction model 420a generated by the model generation unit 320a shown in Fig. 13 to obtain a prediction result indicating the quality of the drug substance.
[0108] The prediction model 420a is a model that takes into account the relationship between the difference in culture conditions and the change over time of the substance in the culture vessel 10, or the relationship between the difference in culture conditions and the quality of the drug substance 3, and therefore can predict the quality of the drug substance 3 more accurately than the prediction model 420. Even if the culture conditions are different when the prediction model 420a is generated and when the quality prediction data 540 is generated using the generated prediction model 420a, the prediction unit 220a can obtain a prediction result indicating the quality of the drug substance by inputting the condition data 550 in addition to the measurement data 510 to the prediction model 420a. That is, the prediction unit 220a can obtain a prediction result that takes into account the relationship between the difference in culture conditions and the quality.
[0109] Fig. 13 is a diagram showing a modified example of the prediction model. For example, as shown in Fig. 13, the model generating unit may generate a prediction model for each culture condition, such as a prediction model 420-1 dedicated to culture condition A and a prediction model 420-2 dedicated to culture condition B.
[0110] In this case, the estimation device may use a prediction model suited to the culture conditions based on the condition data 550.
[0111] Embodiment 2 [Overall configuration of the prediction system] Fig. 14 is a diagram showing a schematic diagram of the overall configuration of the prediction system according to the second embodiment. Like the prediction system SYS1, the prediction system SYS2 has a function of predicting the quality of an active pharmaceutical ingredient of a biopharmaceutical produced by cell culture. In addition to the functions of the prediction system SYS1, the prediction system SYS2 has a function of optimizing culture conditions. The prediction system SYS2 has an optimization device 600 that optimizes the culture conditions.
[0112] The optimization device 600 estimates optimal culture conditions for obtaining a high-quality drug substance 3 based on the time-series data 512 and quality data 520 included in the learning data 530. The optimization device 600 outputs the estimated optimal culture conditions. The user performs a new culture test based on the culture conditions output from the optimization device 600. The culture test includes a culture step (step S10), a purification step (step S30), and an analysis step (step S40). Each of these steps is similar to each of the steps described as the first embodiment. The details of each step are shown in FIG. 9, and therefore the description thereof will not be repeated here.
[0113] When a new culture test is performed, new learning data 530 is input to the optimization device 600. The optimization device 600 re-estimates the optimal culture conditions using the learning data 530 that has already been input and the newly input learning data 530. The optimization device 600 outputs the newly estimated culture conditions. The user performs the next culture test based on the culture conditions output from the optimization device 600.
[0114] Therefore, the culture conditions are further optimized with each increase in the number of culture tests. As the optimization of the culture conditions progresses, the properties of the training data 530 change in a direction toward the production of higher quality drug substance 3. In other words, as the optimization of the culture conditions progresses, the quality of the training data 530 improves. As the quality of the training data 530 improves, the accuracy of optimizing the prediction model 420 increases.
[0115] The optimization device 600 outputs optimized culture conditions in response to a user's instruction. When predicting the quality of a drug substance using the estimation device 200, the user can culture cells based on the culture conditions output from the optimization device 600. Therefore, the user can predict the quality of the drug substance in an environment where a high-quality drug substance can be produced.
[0116] Therefore, the optimization device 600 functions extremely effectively in both the process of generating the prediction model 420 and the process of predicting the quality of the drug substance by utilizing the prediction model 420. According to the prediction system SYS2 according to the second embodiment, the possibility of efficiently obtaining a drug substance of higher quality is increased.
[0117] [Hardware configuration of the optimization device] 15 is a block diagram showing a hardware configuration of an optimization device 600. The optimization device 600 includes a processor 61, a main memory 62, a storage 63, a communication I / F 64, a USB I / F 65, an input unit 66, and a display unit 67. These components are connected via a processor bus 68.
[0118] The processor 61 is composed of a CPU, a GPU, etc., and reads out a program (for example, an optimization program 630) stored in the storage 63, expands it in the main memory 62, and executes it. The processor 61 executes the program to perform a series of processes for searching for optimal values of parameters of the culture conditions.
[0119] The main memory 62 is configured with a volatile storage device such as a RAM, a DRAM, etc. The storage 63 is configured with a non-volatile storage device such as a HDD, an SSD, etc.
[0120] The storage 63 stores learning (estimation) data 530 sent via the USB I / F 65, an optimization program 630 for searching for optimal values 620 of the parameters of the culture conditions, and optimal values 620 of the parameters of the culture conditions newly determined by the processor 61.
[0121] The communication I / F 64 exchanges signals with other communication devices using wired or wireless communication.
[0122] A USB memory (not shown) is detachably attached to the USB I / F 65, and learning data 530 stored in the USB memory is read out.
[0123] The input unit 66 receives user operations and is typically configured with a touch panel, a keyboard, a mouse, etc. The display unit 67 is configured with a liquid crystal panel or the like capable of displaying images.
[0124] [Configuration of optimization device 600] Fig. 16 is a block diagram showing an example of a configuration of an optimization device 600. The optimization device 600 includes a receiving unit 601, an estimation unit 602, an output unit 603, and a storage unit 604, as an example of a software configuration realized by a processor 61 (see Fig. 15) executing an optimization program 630 (see Fig. 15).
[0125] The optimization device 600 searches for culture conditions (hereinafter also referred to as optimal solutions) that improve the quality level of the pharmaceutical active ingredient 3 by Bayesian optimization. Here, the culture conditions are a combination of values of multiple parameters. Bayesian optimization is a method of estimating an optimal event from an observed event using a statistical approach based on Bayesian probability. In Bayesian optimization, a trial is performed based on the set optimal solution, and another optimal solution is searched for based on the result of the trial. The searched optimal solution is set as the optimal solution to be used in the next trial.
[0126] The reception unit 601 receives estimation data. The estimation data is included in the learning data 530 shown in FIG. 14. The estimation data includes at least a part of the time-series data 512 and the quality data 520. The estimation unit 602 searches for an optimal solution of the culture conditions by Bayesian optimization by executing an estimation program. The storage unit 604 includes a plurality of memories for storing the estimation program, the input estimation data, and the optimal solution searched for. The output unit 603 is, for example, the display unit 67 (FIG. 15) that displays the optimal solution of the culture conditions. The output unit 603 may be an interface for outputting the optimal solution of the culture conditions to a display or a printer.
[0127] As shown in window W10, one estimation data set consists of the culture condition data and quality data used in one culture test. The culture condition data consists of multiple parameters p1, p2, ... pn that define the culture conditions. The various parameters are, for example, any of the following: nutrient concentration, pH, temperature, humidity, medium components, type of reagent added, timing of reagent addition, etc.
[0128] The various parameters are, for example, medium conditions collected at the timing when cells are seeded in the medium. This timing corresponds to the measurement timing t0 shown in FIG. 4. That is, the values of the various parameters are the values of the parameters measured at the measurement timing t0 among the time-series data 512 shown in FIG. 14. However, the reception unit 601 may further input the values of the various parameters measured at the various measurement timings shown in FIG. 4. For example, values representing changes in nutrient concentration over time, changes in metabolite concentration over time, changes in pH of the culture medium in the culture vessel, changes in temperature, changes in humidity, etc. may be input to the reception unit 601 as estimation data. The estimation unit 602 may search for optimal culture conditions based on the estimation data.
[0129] [How to optimize culture conditions] Fig. 17 is a flowchart showing the flow of processing by the optimization device 600. Hereinafter, the flow of processing by the optimization device 600 to estimate optimal culture conditions will be described with reference to Figs.
[0130] First, the estimation unit 602 determines whether estimation data has been input to the reception unit 601 (step S600). When estimation data has been input to the reception unit 601, the estimation unit 602 stores the input estimation data in the storage unit 604 (step S601). By repeatedly executing step S601, the estimation data input to the reception unit 601 is accumulated in the storage unit 604. Next, the estimation unit 602 determines the values of parameters p1, p2, ... pn of the optimal culture conditions using all estimation data stored in the storage unit 604 (step S602). In step S602, the estimation unit 602 searches for culture conditions that provide optimal values for the quality level of the drug substance 3 by Bayesian optimization.
[0131] It is preferable to search for culture conditions each time a plurality of pieces of estimation data corresponding to a plurality of culture tests are input, rather than each time a single piece of estimation data is input. For example, it is possible to perform the process of step S602 each time 4 to 10 pieces of estimation data corresponding to 4 to 10 culture tests are input.
[0132] Next, the estimation unit 602 stores the newly determined optimal parameter values in the storage unit 604. If the optimal parameter values determined in the previous search are stored in the storage unit 604, the estimation unit 602 updates those values with the newly determined optimal parameter values (step S603). Next, the estimation unit 602 outputs the newly determined parameter values from the output unit 603 (step S604).
[0133] The parameter values output from the output unit 603 are the latest parameter values indicating the optimal culture conditions. The user performs a new culture test once or multiple times based on the output parameter values. When estimation data corresponding to the new culture test is input to the optimization device 600, the estimation unit 602 searches again for parameter values of the optimal culture conditions. The search results are stored in the memory unit 604 as the optimal parameter values at the current time. Therefore, by repeating the culture test and the optimization of the culture conditions by the optimization device 600, the optimization of the parameter values of the culture conditions stored in the optimization device 600 progresses.
[0134] If the estimation unit 602 judges NO in step S600, it judges whether or not there has been a request to output the parameter values (step S605). For example, if a user wants to predict the quality of a drug substance obtained by a certain incubation test using the estimation device 200, it is necessary to determine the incubation conditions for the incubation test. When determining the incubation conditions for the incubation test, it is useful to refer to the parameter values stored in the optimization device 600. If it is judged in step S600 that there has been a request to output the parameter values, the estimation unit 602 outputs the parameter values stored in the memory unit 604 from the output unit 603 (step S604), and ends the processing based on this flowchart.
[0135] [Modifications of the prediction system] FIG. 18 is a diagram showing a modified example of the prediction system according to the second embodiment. In the modified example, the learning device 300 includes an optimization unit 6000 having the functions of the optimization device 600 shown in FIG. 16. The optimization unit 6000 includes a receiving unit 601, an estimating unit 602, an output unit 603, and a storage unit 604, similar to the optimization device 600. The learning device 300 optimizes the prediction model 420 based on the learning data 530 and searches for optimal values of the parameters of the culture conditions. Therefore, according to the modified example, the added value of the learning device 300 can be increased. Furthermore, according to the modified example, it is not necessary to provide the optimization device 600 separately from the learning device 300, so that costs can be reduced.
[0136] The optimization device 600 or the optimization unit 6000 may optimize the parameter values of the culture conditions using a grid search technique instead of Bayesian optimization.
[0137] [Aspects] It will be understood by those skilled in the art that the above-described embodiments are specific examples of the following aspects.
[0138] (Item 1) An estimation device according to one embodiment includes a receiving unit that receives input of measurement data, a prediction unit that generates quality prediction data indicating the quality of a drug substance by inputting the measurement data received by the receiving unit into a prediction model, and an output unit that outputs the quality prediction data generated by the prediction unit. The measurement data includes a measurement result obtained by measuring at least one substance in a culture vessel containing cells and the culture medium at least one timing after a predetermined period of time has elapsed since the cells were seeded in the culture medium. The prediction model is a model for predicting the quality of a drug substance of a biopharmaceutical produced by culturing cells.
[0139] The estimation device described in paragraph 1 is capable of predicting the quality of the drug substance obtained by proceeding with the culture during the culture step. Therefore, based on the predicted quality of the drug substance, the user can take various measures such as reviewing the protocol at the culture step before the purification step for obtaining the drug substance, extending the culture time, or stopping the culture, thereby reducing the manufacturing cost of biopharmaceuticals.
[0140] (Item 2) In the estimation device according to item 1, the measurement data includes measurement results obtained by measuring at least one substance at multiple times after the cells are seeded in the culture medium.
[0141] (Item 3) In the estimation device according to item 1 or 2, the measurement data includes a measurement result obtained by measuring the at least one substance when starting cell culture.
[0142] (Item 4) In the estimation device according to any one of items 1 to 3, the measurement data includes a measurement result obtained by measuring at least one substance during the logarithmic growth phase of cells.
[0143] During the logarithmic growth phase, the number of cells fluctuates greatly, so it is expected that the state of the material in the culture vessel will fluctuate greatly. Since the measurement data includes the results of measuring the measurement target at the time when a large change in state occurs, the estimation device described in Section 4 can predict the quality of the drug substance using data that more accurately indicates the change in cell state.
[0144] (Item 5) In the estimation device according to any one of items 1 to 4, the measurement data includes a measurement result obtained by measuring at least one substance during the stationary phase of the cells.
[0145] In the stationary phase, a balance is maintained between the number of proliferating cells and the number of dying cells, and the number of cells is at its maximum. Since the result of measuring the measurement target at the time when the number of cells is at its maximum is included in the measurement data, the estimation device described in paragraph 5 can predict the quality of the drug substance using data showing the state of the material in the culture vessel when the number of cells is at its maximum.
[0146] (Item 6) In the estimation device described in any one of items 1 to 4, the measurement data includes a measurement result obtained by measuring the at least one substance during the death phase of the cell (Figure 5).
[0147] During the death phase, the cells die, exposing intracellular enzymes to the medium, which may affect the quality of the drug substance. In this case, the measurement results during the cell death phase may be used in machine learning as an example of poor quality data.
[0148] (Item 7) In the estimation device described in any one of items 1 to 6, the at least one substance is at least one of nutrients ingested by the cells, metabolites produced by the metabolism of the cells, and cells.
[0149] (Item 8) In the estimation device according to any one of items 1 to 7, the prediction model outputs a first predicted value when the quality of the drug substance is evaluated from a first perspective and a second predicted value when the quality of the drug substance is evaluated from a second perspective, by inputting the measurement data received by the receiving unit. The quality prediction data includes the first predicted value and the second predicted value.
[0150] The estimation device described in Section 8 can evaluate the quality of the drug substance from various viewpoints.
[0151] (Item 9) In the estimation device according to any one of items 1 to 7, the prediction model outputs a predicted value when the quality of the drug substance is evaluated from a predetermined viewpoint by inputting the measurement data received by the receiving unit. The prediction unit includes a determination unit that determines the quality of the drug substance based on the predicted value. The quality prediction data includes a determination result of the quality of the drug substance.
[0152] The estimation device described in item 9 can display the predicted results for the quality of the drug substance in an easier-to-understand manner.
[0153] (Item 10) In the estimation device according to any one of items 1 to 8, the receiving unit further receives an input of condition data indicating a culture condition of the cells. The prediction unit generates quality prediction data by inputting the measurement data and condition data received by the receiving unit into a prediction model.
[0154] In the estimation device described in item 10, a prediction result is obtained that takes into account the correlation between differences in culture conditions and quality, thereby obtaining a more accurate prediction result.
[0155] (Item 11) A learning device according to one embodiment includes a receiving unit that receives learning data, and a model generating unit that generates a predictive model by executing a learning process using the learning data received by the receiving unit. The learning data includes measurement data including measurement results obtained by measuring at least one substance in a culture vessel containing the cells and the culture medium at multiple times after the cells are seeded in the culture medium, and quality data obtained by analyzing a drug substance of a biopharmaceutical produced from the cells. The predictive model is a model for generating quality prediction data indicating the quality of a drug substance of a biopharmaceutical produced from cells contained in the culture vessel during the course of culturing the cells, based on measurement results obtained by measuring at least one substance in the culture vessel during the course of culturing the cells.
[0156] The learning device described in paragraph 11 can generate a prediction model for predicting the quality of the drug substance obtained by proceeding with the culture during the culture stage. Therefore, based on the predicted quality of the drug substance, the user can take various measures such as reviewing the protocol at the culture stage before the purification process for obtaining the drug substance, extending the culture time, or stopping the culture, thereby reducing the manufacturing cost of biopharmaceuticals.
[0157] (Item 12) In one embodiment of the learning device, the learning data further includes condition data indicating cell culture conditions, the condition data including values of a plurality of parameters that define the culture conditions, and the learning device further includes an optimization unit that optimizes a combination of values of the plurality of parameters, and the optimization unit estimates the optimal combination of values of the plurality of parameters using as input the values of the plurality of parameters and quality data obtained when cells are cultured based on the values of the plurality of parameters.
[0158] The learning device described in paragraph 12 can estimate a combination of multiple parameter values that define the cell culture conditions during the course of the culture. This allows the user to optimize the conditions during the course of the culture to improve the quality of the drug substance obtained by proceeding with the culture.
[0159] (Item 13) In the learning device described in Item 11, the learning data further includes condition data indicating conditions for culturing the cells contained in the culture vessel. The learning data accepted by the accepting unit includes first learning data obtained when the cells are cultured under first conditions, and second learning data obtained when the cells are cultured under second conditions different from the first conditions. The model generating unit generates a prediction model by executing a learning process using the first learning data and the second learning data.
[0160] The learning device described in paragraph 13 can generate a prediction model that takes into account the relationship between differences in culture conditions and changes over time of substances in the culture vessel, or the relationship between differences in culture conditions and the quality of the drug substance. Therefore, the learning device can generate a prediction model for obtaining more accurate prediction results.
[0161] (Item 14) An estimation method according to one embodiment includes the steps of placing cells and a culture medium in a culture vessel and culturing the cells, measuring at least one substance in the culture vessel at least one time point after a predetermined period of time has elapsed since the cells were seeded in the culture medium, generating quality prediction data indicating the quality of the drug substance by inputting measurement data including the measurement results obtained in the measuring step into a prediction model, and outputting the quality prediction data. The prediction model is a model for predicting the quality of the drug substance of a biopharmaceutical produced by culturing cells.
[0162] In the estimation method described in item 14, the quality of the drug substance obtained by proceeding with the culture can be predicted during the culture step. Therefore, based on the predicted quality of the drug substance, the user can take various measures such as reviewing the protocol at the culture step before the purification step for obtaining the drug substance, extending the culture time, or stopping the culture, thereby reducing the manufacturing cost of the biopharmaceutical.
[0163] (Item 15) A learning method according to one embodiment includes the steps of placing cells and a culture medium in a culture vessel and culturing the cells, measuring at least one substance in the culture vessel at multiple times after the cells are seeded in the culture medium, analyzing the quality of an active pharmaceutical ingredient of a biopharmaceutical produced from the cells, and generating a predictive model by executing a learning process using the learning data. The learning data includes measurement data including the measurement results obtained in the measuring step, and quality data obtained in the analyzing step. The predictive model is a model for generating quality prediction data indicating the quality of the active pharmaceutical ingredient of a biopharmaceutical produced from cells contained in the culture vessel during the culture, based on the measurement results obtained by measuring at least one substance in the culture vessel during the culture of the cells.
[0164] In the learning method described in paragraph 15, a prediction model is generated midway through the culture to predict the quality of the drug substance obtained by proceeding with the culture. Therefore, based on the predicted quality of the drug substance, the user can take various measures such as reviewing the protocol at the culture step prior to the purification step for obtaining the drug substance, extending the culture time, or stopping the culture, thereby reducing the manufacturing cost of biopharmaceuticals.
[0165] (Item 16) Moreover, an optimization device according to one embodiment includes a receiving unit that receives values of a plurality of parameters that define culture conditions when cells are seeded in a culture medium and quality data obtained by analyzing an active pharmaceutical ingredient of a biopharmaceutical produced by culturing cells, an estimation unit that estimates an optimal combination of values of the plurality of parameters using the values of the plurality of parameters and the quality data received by the receiving unit as input, and an output unit that outputs the combination of values of the plurality of parameters estimated by the estimation unit.
[0166] In the optimization device described in paragraph 16, a combination of values of a plurality of parameters defining the cell culture conditions can be estimated during the course of the culture. Therefore, a user can optimize conditions for improving the quality of the drug substance obtained by proceeding with the culture during the course of the culture. The estimation unit may estimate an optimal combination of values of the plurality of parameters by performing Bayesian optimization using the values of the plurality of parameters and the quality data obtained by the analyzing step as input.
[0167] (Item 17) An optimization method according to one embodiment includes the steps of placing cells and a culture medium in a culture vessel and culturing the cells under culture conditions defined by values of a plurality of parameters, analyzing the quality of an active pharmaceutical ingredient of a biopharmaceutical produced by culturing the cells, and estimating an optimal combination of values of the plurality of parameters using the values of the plurality of parameters and the quality data obtained by the analyzing step as inputs. The cell culturing step involves culturing the cells under new culture conditions that are the optimal combination of values of the plurality of parameters estimated by the estimating step.
[0168] In the estimation method described in item 17, the user can optimize the conditions during the cultivation process to improve the quality of the drug substance obtained by proceeding with the cultivation.
[0169] (Item 18) A prediction system according to one embodiment includes an estimation device according to any one of items 1 to 10, a learning device according to any one of items 11 to 13, and a measurement device that measures at least one substance in a culture container.
[0170] In the prediction system described in paragraph 18, the estimation device and the measurement device may be realized by a single information processing device. In this case, the reception unit of the estimation device and the reception unit of the measurement device may be realized as a single reception unit. In other words, the reception unit receives measurement data obtained by the measurement device measuring a substance.
[0171] (Item 19) In the prediction system according to item 18, the measurement device includes a first measurement device that measures a first substance and a second measurement device that measures a second substance.
[0172] The embodiments disclosed herein are also intended to be combined as appropriate within the scope of technical inconsistency. The embodiments disclosed herein should be considered to be illustrative and not restrictive in all respects. The scope of the present invention is defined by the claims, not the description of the above embodiments, and is intended to include all modifications within the scope and meaning equivalent to the claims. [Explanation of symbols]
[0173] 1 cell, 2 culture medium, 3 drug substance, 10 culture vessel, 21, 31, 61 processor, 22, 32, 62 main memory, 23, 33, 63 storage, 24, 34, 64 communication I / F, 25, 35, 65 USB I / F, 26, 36, 66 input unit, 27, 37, 67 display unit, 28, 38, 68 processor bus, 100 measurement device, 200, 200a estimation device, 210, 210a, 310 reception unit, 220, 220a prediction unit, 222 estimation processing unit, 224 judgment unit, 230 output unit, 300, 300a learning device, 312 learning data generation unit, 320, 320a model generation unit, 330 timing identification unit, 400 server, 420, 420a prediction model, 510 Measurement data, 512 time Series data, 520 quality data, 530, 530a, 530b learning data, 540 quality prediction data, 542 predicted value, 544 judgment result, 550 condition data, 560 estimation program, 580 learning program, 600 optimization device, 601 reception unit, 602 estimation unit, 603 output unit, 604 memory unit, 620 optimal parameter value, 630 optimization program, 6000 optimization unit, SYS1, SYS2 prediction system, W10 window.
Claims
1. a receiving unit that receives an input of measurement data including a measurement result obtained by measuring at least one substance in a culture vessel containing the cells and the culture medium at least one time point after a predetermined period has elapsed since the cells were seeded in the culture medium; a prediction unit that generates quality prediction data indicating a predicted quality of the drug substance by inputting the measurement data received by the reception unit into a prediction model for predicting the quality of the drug substance of the biopharmaceutical produced by culturing the cells; and an output unit that outputs the quality prediction data generated by the prediction unit, The measurement data includes a measurement result obtained by measuring the at least one substance during the logarithmic growth phase of the cells, The measurement data includes measurement results obtained by measuring the at least one substance in the culture vessel at multiple times after the cells are seeded in the culture medium.
2. The estimation device according to claim 1 , wherein the measurement data includes a measurement result obtained by measuring the at least one substance when starting to culture the cells.
3. The estimation device according to claim 1 , wherein the measurement data includes a measurement result obtained by measuring the at least one substance during the stationary phase of the cell.
4. The estimation device according to claim 1 , wherein the measurement data includes a measurement result obtained by measuring the at least one substance during the death phase of the cell.
5. The estimation device according to claim 1 , wherein the at least one substance is at least one of a nutrient ingested by the cell, a metabolite produced by metabolism of the cell, and the cell itself.
6. a first predicted value when the quality of the drug substance is evaluated from a first perspective and a second predicted value when the quality of the drug substance is evaluated from a second perspective are output from the prediction model by inputting the measurement data accepted by the accepting unit; The method of claim 1 , wherein the quality prediction data includes the first prediction value and the second prediction value. Estimation device.
7. a prediction value when the quality of the drug substance is evaluated from a predetermined viewpoint by inputting the measurement data accepted by the accepting unit, from the prediction model, is output; The prediction unit includes a determination unit that determines the quality of the drug substance based on the predicted value, The estimation device according to claim 1 , wherein the quality prediction data includes a result of judging the quality of the pharmaceutical ingredient.
8. The receiving unit further receives input of condition data indicating culture conditions of the cells; The estimation device according to claim 1 , wherein the prediction unit generates the quality prediction data by inputting the measurement data and the condition data accepted by the acceptance unit into the prediction model.
9. The estimation device of claim 1, wherein the multiple timings include multiple timings during the logarithmic growth phase, multiple timings during the stationary phase of the cell, or multiple timings during the death phase of the cell.
Citation Information
Patent Citations
Method for constructing prediction model for predicting quality of cell, program for constructing prediction model, recording medium recording the program, and device for constructing prediction model
JP2009044974A
Control method and system of cell culture device
JP2019110767A
Improving product quality using mammalian cell culture methods for protein production
JP4496086B2
Affinity-Based Analytical Purification Of Biotherapeutics For Bioprocess Monitoring
US20140322699A1
Multivariate Spectral Analysis and Monitoring for Biomanufacturing
US20190272894A1
