Medicine development support device, operation method for medicine development support device, operation program for medicine development support device, learning device, operation method for learning device, and operation program for learning device

Abstract

This pharmaceutical development support device comprises a processor. The processor: acquires sequence information of amino acid residues constituting an amino-acid-derived substance included in a pharmaceutical product, and physical property information of the amino acid residues; inputs the sequence information and the physical property information to a substance language model in which pre-learning and fine-tuning have been performed using learning data including sequence information for learning and physical property information for learning; causes the substance language model to output a prediction result related to the substance; and presents the prediction result to a user.

Description

Drug development support device, drug development support device operation method, drug development support device operation program, learning device, learning device operation method, and learning device operation program

[0001] The technology disclosed herein relates to a drug development support device, a drug development support device operating method, a drug development support device operating program, a learning device, a learning device operating method, and a learning device operating program.

[0002] Recently, pharmaceuticals such as biopharmaceuticals, peptide drugs, and nucleic acid drugs have been attracting attention due to their high efficacy and minimal side effects. For example, biopharmaceuticals use proteins such as interferon and antibodies as their active ingredients. Biopharmaceuticals are stored in a preservative solution. To maintain the stable quality of biopharmaceuticals, it is important to ensure that the preservative solution formulation (also known as formulation) is suitable for biopharmaceuticals.

[0003] In the field of natural language processing (NLP), language models such as BERT (Bidirectional Encoder Representations from Transformers) using a transformer encoder have been attracting attention. In language models, masked language modeling (MLM) and next sentence prediction (NSP) are performed as pre-learning tasks. MLM is a so-called fill-in-the-blank problem in which a trained input sentence in which a part of a word or phrase is masked is predicted to be filled in the masked part. NSP is a process for determining whether two sentences are semantically consecutive. After this pre-training, fine-tuning (hereinafter abbreviated as FT) is performed according to the desired natural language processing task. Note that a phrase refers to a single word and / or a phrase consisting of a combination of one or more words.

[0004] Various technologies have been proposed that use language models to promote the development of biopharmaceuticals by treating sequence information of amino acid residues that make up proteins as sentences. For example, Danqing Wang et al. "On Pre-trained Language Models for Antibodies" Published as a conference paper at ICLR 2023 January 31, 2023 (hereinafter referred to as Non-Patent Document 1) describes a technology for pre-training a language model using tasks such as predicting the ancestral germline of a protein (antibody) and predicting mutation sites in the protein. In addition, Japanese Patent Laid-Open No. 2023-022060 (hereinafter referred to as Patent Document 1) describes a technology for pre-training a language model based on sequence information of amino acid residues that make up a protein, functional information of the protein, and structural information of the protein. The structural information given is based on the extraction results of a point group consisting of heavy atoms of a protein. The functional information is described only as "text description information of the protein's function," with no specific examples given. The language model to which the sequence information of the amino acid residues that make up a protein is applied is called a protein language model.

[0005] Amino acid residues that make up amino acid-derived substances such as proteins vary in physical properties, such as hydrophilicity / hydrophobicity, surface exposure area, and charge state. Furthermore, these physical properties determine, for example, appropriate formulations for stabilizing the quality of biopharmaceuticals. However, Non-Patent Document 1 and Patent Document 1 do not take into account the physical properties of amino acid residues. As a result, the prediction accuracy of the protein language model may be insufficient.

[0006] One embodiment of the technology of the present disclosure provides a pharmaceutical development support device, an operating method for the pharmaceutical development support device, an operating program for the pharmaceutical development support device, a learning device, an operating method for the learning device, and an operating program for the learning device, which are capable of improving the prediction accuracy of a substance language model to which sequence information of amino acid residues that constitute amino acid-derived substances such as proteins is applied.

[0007] The pharmaceutical development support device of the present disclosure includes a processor, which acquires sequence information of amino acid residues that constitute substances derived from amino acids contained in pharmaceuticals, and physical property information of the amino acid residues, and uses a substance language model that has been pre-trained and fine-tuned using training data including training sequence information and training physical property information, inputs the sequence information and physical property information into the substance language model, causes the substance language model to output prediction results related to substances, and presents the prediction results to a user.

[0008] The material language model is preferably a model that has been pre-trained using training data in which at least one of the sequence information portion and the physical property information portion is masked.

[0009] The physical property information is preferably at least one of a numerical value representing the hydrophilicity / hydrophobicity of an amino acid residue, a surface exposed area of ​​the amino acid residue, and data representing the charge state of the amino acid residue.

[0010] The predicted results preferably include information regarding the prescription of a stock solution of the pharmaceutical product.

[0011] Preferably, the substance is one of a protein, a peptide, and a nucleic acid.

[0012] Preferably, the protein is an antibody.

[0013] The method of operating the drug development support device disclosed herein includes obtaining sequence information of amino acid residues that constitute substances derived from amino acids contained in the drug, and physical property information of the amino acid residues; using a substance language model that has been pre-trained and fine-tuned using training data including sequence information and physical property information for training; inputting the sequence information and physical property information into the substance language model; outputting prediction results related to substances from the substance language model; and presenting the prediction results to a user.

[0014] The operating program of the drug development support device disclosed herein causes a computer to execute processes including obtaining sequence information of amino acid residues that constitute substances derived from amino acids contained in drugs, and physical property information of the amino acid residues, using a substance language model that has been pre-trained and fine-tuned using training data including sequence information and physical property information for training, inputting the sequence information and physical property information into the substance language model, outputting prediction results related to substances from the substance language model, and presenting the prediction results to a user.

[0015] The learning device disclosed herein includes a processor that acquires training data including sequence information of amino acid residues that constitute substances derived from amino acids contained in pharmaceuticals and physical property information of the amino acid residues, and uses the training data to pre-train and fine-tune a substance language model that outputs prediction results regarding substances.

[0016] It is preferable that at least one of the sequence information portion and the physical property information portion of the training data used in the pre-training is masked.

[0017] The method of operating the learning device disclosed herein includes obtaining training data including sequence information of amino acid residues that constitute substances derived from amino acids contained in pharmaceuticals and physical property information of the amino acid residues, and using the training data to pre-train and fine-tune a substance language model that outputs prediction results regarding substances.

[0018] The operating program of the learning device disclosed herein causes a computer to execute processes including acquiring training data including sequence information of amino acid residues that constitute substances derived from amino acids contained in pharmaceuticals and physical property information of the amino acid residues, and using the training data to pre-train and fine-tune a substance language model that outputs prediction results regarding substances.

[0019] According to the technology of the present disclosure, it is possible to provide a pharmaceutical development support device, an operating method for a pharmaceutical development support device, an operating program for a pharmaceutical development support device, a learning device, an operating method for a learning device, and an operating program for a learning device, which are capable of improving the prediction accuracy of a substance language model to which sequence information of amino acid residues that constitute amino acid-derived substances such as proteins is applied.

[0020] 1 is a diagram showing a drug development support system. FIG. 2 is a diagram showing sequence information. FIG. 3 is a diagram showing physical property information. FIG. 4 is a block diagram showing computers constituting a learning device, a drug development support device, and a user terminal. FIG. 5 is a block diagram showing a processing unit of a CPU of the learning device. FIG. 6 is a diagram showing a past data group. FIG. 7 is a diagram showing processing of a generation unit. FIG. 8 is a diagram showing masking processing for sequence information. FIG. 9 is a diagram showing masking processing for physical property information. FIG. 10 is a diagram showing a 1_1 training data group. FIG. 11 is a diagram showing a 1_2 training data group. FIG. 12 is a diagram showing processing of a pre-training unit. FIG. 13 is a diagram showing processing in MLM. FIG. 14 is a diagram showing processing in NSP. FIG. 15 is a diagram showing processing in FT. FIG. 16 is a block diagram showing a processing unit of a CPU of a drug development support device. FIG. 17 is a diagram showing processing of a prediction unit. FIG. 18 is a block diagram showing a processing unit of a CPU of a user terminal. FIG. 19 is a diagram showing an information input screen. FIG. 20 is a diagram showing a prediction result display screen. FIG. 21 is a flowchart showing the processing procedure of a learning device. FIG. 22 is a flowchart showing the processing procedure of a drug development support device.

[0021] As shown in FIG. 1 , a drug development support system 10 is a system for supporting the development of biopharmaceuticals, and includes a learning device 11, a drug development support device 12, and a user terminal 13. The learning device 11 generates a protein language model 14 and transmits the protein language model 14 to the drug development support device 12. The protein language model 14 is a BERT-based language model. The protein language model 14 predicts information useful for promoting the development of biopharmaceuticals by treating sequence information 16 of amino acid residues constituting proteins, which are active ingredients of biopharmaceuticals, as sentences. In this example, the protein is an antibody 15. The information useful for promoting the development of biopharmaceuticals is information regarding an appropriate formulation of a preservative solution for stabilizing the quality of the biopharmaceutical. The biopharmaceutical is an example of a "drug" according to the technology of the present disclosure. The antibody 15 is an example of an "amino acid-derived substance" and a "protein" according to the technology of the present disclosure, and the protein language model 14 is an example of a "substance language model" according to the technology of the present disclosure.

[0022] The drug development support device 12 and user terminal 13 are connected via a network 17. The user terminal 13 is installed at a pharmaceutical company developing biopharmaceuticals or at an organization contracted by a pharmaceutical company to develop biopharmaceuticals, i.e., a contract research organization (CRO). The user terminal 13 is operated by a user U involved in the development of biopharmaceuticals at the pharmaceutical company or contract research organization. The network 17 is, for example, a wide area network (WAN) such as the Internet or a public communication network. Note that while only one user terminal 13 is connected to the drug development support device 12 in FIG. 1 , in reality, multiple user terminals 13 from multiple pharmaceutical companies or contract research organizations are connected to the drug development support device 12.

[0023] The user terminal 13 transmits a prediction request 18 to the drug development support device 12. The prediction request 18 is a request to have the drug development support device 12 predict information regarding an appropriate formulation of a biopharmaceutical preservative solution. The prediction request 18 includes sequence information 16 of the amino acid residues that make up the antibody 15, and physical property information 19 of the amino acid residues. The sequence information 16 is identified through experiments. The physical property information 19 is identified through experiments or simulations. Although not shown, the prediction request 18 also includes a terminal ID (identification data) and the like for uniquely identifying the user terminal 13 that sent the prediction request 18.

[0024] When receiving the prediction request 18, the pharmaceutical development support device 12 predicts information regarding an appropriate formulation prescription for a biopharmaceutical preservative solution using the protein language model 14. The resulting prescription prediction result 20 is then distributed to the user terminal 13 that sent the prediction request 18. When receiving the prescription prediction result 20, the user terminal 13 makes the prescription prediction result 20 available for viewing by the user U. The prescription prediction result 20 is an example of a "prediction result" according to the technology of the present disclosure.

[0025] As an example, as shown in Figure 2, sequence information 16 describes the order of peptide bonds of the amino acid residues that make up antibody 15, from the amino terminus to the carboxyl terminus, using single-letter alphabetic abbreviations that represent the amino acid residues. Since antibody 15 consists of approximately 450 amino acid residues, sequence information 16 also consists of a string of approximately 450 letters. For example, "E" represents glutamic acid, "L" represents leucine, and "G" represents glycine. Such an amino acid residue sequence is also called a primary structure.

[0026] As shown in FIG. 3 as an example, the physical property information 19 is information in which the standard free energy change ΔG, the surface exposed area SASA (Solvent Accessible Surface Area), and the charge state of each amino acid residue are registered in the order of the amino acid residue sequence. A positive standard free energy change ΔG indicates that the amino acid residue is hydrophilic, and a negative standard free energy change ΔG indicates that the amino acid residue is hydrophobic. In other words, the standard free energy change ΔG is an example of a "numerical value representing the hydrophilicity / hydrophobicity of an amino acid residue" according to the technology of the present disclosure. A larger value of the surface exposed area SASA indicates a lower stability of the amino acid residue. The charge state is registered as either "+ (positive)," "- (negative)," or "N (neutral)." These "+ (positive)," "- (negative)," and "N (neutral)" are examples of "data representing the charge state of an amino acid residue" according to the technology of the present disclosure.

[0027] 4, the computers that make up the learning device 11, the drug development support device 12, and the user terminal 13 basically have the same configuration, and include a storage 25, a memory 26, a CPU (Central Processing Unit) 27, a communication unit 28, a display 29, and an input device 30. These are interconnected via a bus line 31.

[0028] The storage 25 is a hard disk drive built into the computers constituting the learning device 11, the drug development support device 12, and the user terminal 13, or connected via a cable or network. Alternatively, the storage 25 is a disk array consisting of multiple hard disk drives. The storage 25 stores control programs such as an operating system, various application programs (hereinafter referred to as APs (Application Programs)), and various data associated with these programs. Note that a solid state drive may be used instead of a hard disk drive.

[0029] The memory 26 is a work memory for the CPU 27 to execute processing. The CPU 27 loads programs stored in the storage 25 into the memory 26 and executes processing in accordance with the programs. In this way, the CPU 27 comprehensively controls each part of the computer. The CPU 27 is an example of a "processor" according to the technology of the present disclosure. The memory 26 may be built into the CPU 27.

[0030] The communication unit 28 is a network interface that controls the transmission of various information via the network 17, etc. The display 29 displays various screens. The various screens are equipped with operation functions using a GUI (Graphical User Interface). The computers that make up the learning device 11, the drug development support device 12, and the user terminal 13 accept input of operation instructions from the input device 30 via the various screens. The input device 30 is a keyboard, a mouse, a touch panel, a microphone for voice input, etc.

[0031] In the following explanation, the parts of the computer that make up the learning device 11 (storage 25 and CPU 27) are distinguished by adding the suffix "A" to their symbols, the parts of the computer that make up the pharmaceutical development support device 12 (storage 25 and CPU 27) are distinguished by adding the suffix "B" to their symbols, and the parts of the computer that make up the user terminal 13 (storage 25, CPU 27, display 29, and input device 30) are distinguished by adding the suffix "C" to their symbols.

[0032] 5, an operating program 35 is stored in the storage 25A of the learning device 11. The operating program 35 is an AP for causing a computer to function as the learning device 11. In other words, the operating program 35 is an example of an "operating program for a learning device" according to the technology of the present disclosure. The storage 25A also stores a protein language model 14, a past data group 36, a learning data group 37, and the like.

[0033] When the operating program 35 is started, the CPU 27A of the computer constituting the learning device 11 functions as an RW control unit 40, a generation unit 41, a pre-learning unit 42, and an FT unit 43 in cooperation with the memory 26 and the like.

[0034] The RW control unit 40 controls the storage of various data in the storage 25A and the reading of various data from the storage 25A. For example, the RW control unit 40 reads the past data group 36 from the storage 25A and outputs the read past data group 36 to the generation unit 41. The RW control unit 40 also reads a first training data group 371 of the training data group 37 from the storage 25A and outputs the read first training data group 371 to the pre-training unit 42. Similarly, the RW control unit 40 reads a second training data group 372 of the training data group 37 from the storage 25A and outputs the read second training data group 372 to the FT unit 43. The RW control unit 40 also reads the protein language model 14 from the storage 25A and outputs the protein language model 14 to the pre-training unit 42 or the FT unit 43.

[0035] The generation unit 41 generates a training data group 37, which is a collection of training data for training the protein language model 14, based on the past data group 36. The generation unit 41 outputs the training data group 37 to the RW control unit 40. The RW control unit 40 stores the training data group 37 in the storage 25A.

[0036] The pre-training unit 42 pre-trains the protein language model 14 using the first training data group 371. The pre-training unit 42 outputs the protein language model 14 after pre-training to the RW control unit 40. The RW control unit 40 stores the protein language model 14 after pre-training in the storage 25A. In the following description, the protein language model 14 before pre-training will be referred to as the protein language model 14A, and the protein language model 14 after pre-training will be referred to as the protein language model 14B.

[0037] The FT unit 43 performs FT on the protein language model 14B using the second training data group 372. The FT unit 43 outputs the protein language model 14B after FT to the RW control unit 40. The RW control unit 40 stores the protein language model 14B after FT in the storage 25A. In the following description, the protein language model 14B after FT will be referred to as the protein language model 14C. This protein language model 14C is transmitted from the learning device 11 to the pharmaceutical development support device 12. In addition to these processing units 40 to 43, the CPU 27A also includes an instruction receiving unit that receives various operation instructions from the input device 30.

[0038] As an example, as shown in FIG. 6 , the historical data group 36 is a collection of historical data 45 for multiple biopharmaceuticals developed in the past. Each piece of historical data 45 is identified by the antibody identification data (ID) of the antibody 15 contained in the biopharmaceutical. The historical data 45 may be obtained from a public database of biopharmaceuticals, or may be obtained from biopharmaceuticals previously developed by a pharmaceutical company or a contract research organization. Alternatively, the historical data 45 may be composed of both public information and information independently accumulated by a pharmaceutical company or a contract research organization.

[0039] The past data 45 includes prescription information 46 and related antibody information 47 in addition to the sequence information 16 and physical property information 19. The prescription information 46 is information on the formulation used for the preservation solution of the biopharmaceutical, and here, the hydrogen ion exponent (pH (Potential Of Hydrogen) value) of the preservation solution is shown as an example. The related antibody information 47 registers the antibody IDs of antibodies 15 related to the antibody 15 contained in the biopharmaceutical. Related antibodies 15 are antibodies 15 with similar structures, antibodies 15 targeting the same organ, disease, etc.

[0040] 7 as an example, the generation unit 41 first randomly allocates the multiple pieces of past data 45 in the past data group 36 into a first past data group 361 and a second past data group 362. At this time, the generation unit 41 allocates a larger number of pieces of past data 45 in the first past data group 361 than in the second past data group 362. For example, 80% of the multiple pieces of past data 45 in the past data group 36 are allocated to the first past data group 361, and the remaining 20% ​​are allocated to the second past data group 362. Note that the past data 45 in the first past data group 361 and the past data 45 in the second past data group 362 may partially overlap.

[0041] The generation unit 41 generates a first learning data group 371 from the first past data group 361. The generation unit 41 also generates a second learning data group 372 from the second past data group 362. The first learning data group 371 and the second learning data group 372 constitute the learning data group 37.

[0042] The first learning data group 371 is composed of a first_1 learning data group 371_1 and a first_2 learning data group 371_2. The generating unit 41 generates the first_1 learning data group 371_1 by performing a masking process on the past data 45 of the first past data group 361.

[0043] As an example, as shown in Figures 8 and 9, the masking process is a process in which the alphabet of the sequence information 16 and the standard free energy change ΔG, surface exposure area SASA, and charge state of the physical property information 19 are each considered as one token, and the alphabet of the sequence information 16 is masked (Figure 8) or the standard free energy change ΔG, surface exposure area SASA, and charge state of the physical property information 19 are masked (Figure 9).

[0044] The masking process is performed according to preset masking conditions. The masking conditions include randomly masking, for example, 15% of the alphabets of the sequence information 16, and the standard free energy change ΔG, surface exposed area SASA, and charge state of the physical property information 19. Consider a case where the total number of alphabets of the sequence information 16, the standard free energy change ΔG, surface exposed area SASA, and charge state of the physical property information 19 is, for example, 2000. In this case, since 2000 × 0.15 = 300, the masking condition is satisfied if at least 300 are masked. Because the masking process is random, there are cases where the sequence information 16 is not masked, and cases where the physical property information 19 is not masked. In the following description, the sequence information 16 after the masking process is referred to as sequence information 16A, and the physical property information 19 after the masking process is referred to as physical property information 19A.

[0045] As an example, as shown in FIG. 10 , the first_1 learning data 501_1 constituting the first_1 learning data group 371_1 includes sequence information 16A, physical property information 19A, and mask information 51. The mask information 51 is information indicating the alphabet of the sequence information 16 masked by the masking process, as well as the standard free energy change ΔG, the surface exposure area SASA, and the charge state of the physical property information 19. In other words, the mask information 51 is information that serves as the answer to the pre-learning MLM. The sequence information 16A and physical property information 19A of the first_1 learning data 501_1 are examples of "sequence information for learning" and "physical property information for learning" according to the technology of the present disclosure.

[0046] 11 , the first_2 learning data 501_2 constituting the first_2 learning data group 371_2 includes sequence information 16, physical property information 19, and related antibody information 47. The related antibody information 47 is information that serves as the answer to the pre-learned NSP. The sequence information 16 and physical property information 19 of the first_2 learning data 501_2 are also examples of "sequence information for learning" and "physical property information for learning" according to the technology of the present disclosure.

[0047] As an example, as shown in FIG. 12 , second learning data 502 constituting the second learning data group 372 includes sequence information 16, physical property information 19, and prescription information 46. Prescription information 46 is information that serves as the answer to FT. Sequence information 16 and physical property information 19 of second learning data 502 are also examples of "sequence information for learning" and "physical property information for learning" according to the technology of the present disclosure. In the following description, unless there is a need to particularly distinguish between first_1 learning data 501_1, first_2 learning data 501_2, and second learning data 502, they will be collectively referred to as learning data 50.

[0048] 13, the pre-training unit 42 performs MLM using a first_1 training data group 371_1 as pre-training. The pre-training unit 42 also performs NSP using a first_2 training data group 371_2 as pre-training. By performing pre-training in this manner, the pre-training unit 42 transforms the protein language model 14A into a protein language model 14B.

[0049] As an example, as shown in FIG. 14 , in the pre-training MLM, the pre-training unit 42 inputs sequence information 16A and physical property information 19A from the first_1 training data 501_1 into the protein language model 14A and causes the protein language model 14A to output a mask prediction result 55. The mask prediction result 55 is a result of predicting the alphabet of the sequence information 16 masked by the masking process, as well as the standard free energy change ΔG, surface exposure area SASA, and charged state portion of the physical property information 19. Based on the mask prediction result 55 and the mask information 51, a loss calculation is performed on the protein language model 14A using a loss function. Then, the values ​​of various parameters of the protein language model 14A are updated according to the result of the loss calculation, and the protein language model 14A is updated according to the update settings.

[0050] The pre-training unit 42 repeatedly performs the above series of processes, including inputting the sequence information 16A and the physical property information 19A to the protein language model 14A, outputting the mask prediction result 55 from the protein language model 14A, calculating the loss, setting the update, and updating the protein language model 14A, while changing the first_1 training data 501_1, until the prediction accuracy of the mask prediction result 55 reaches a preset level. Alternatively, the pre-training unit 42 ends the above series of processes when the series of processes has been repeated a preset number of times.

[0051] As an example, as shown in FIG. 15 , in the pre-training NSP, the pre-training unit 42 inputs the sequence information 16 and physical property information 19 of two first_2 training data 501_2 to the protein language model 14A and causes the protein language model 14A to output a relevance prediction result 57. The relevance prediction result 57 is a result of predicting whether or not there is a relevance between the antibodies 15 associated with the two first_2 training data 501_2. Based on this relevance prediction result 57 and the related antibody information 47, a loss calculation is performed on the protein language model 14A using a loss function. Then, the values ​​of various parameters of the protein language model 14A are updated according to the result of the loss calculation, and the protein language model 14A is updated according to the update setting.

[0052] The pre-training unit 42 repeatedly performs the above series of processes, including inputting the sequence information 16 and the physical property information 19 to the protein language model 14A, outputting the relevance prediction result 57 from the protein language model 14A, calculating the loss, setting the update, and updating the protein language model 14A, while changing the first_2 training data 501_2, until the prediction accuracy of the relevance prediction result 57 reaches a preset level. Alternatively, the pre-training unit 42 terminates the above series of processes after repeating the process a preset number of times. Note that, as pre-training, only the MLM shown in FIG. 14 may be performed, and the NSP shown in FIG. 15 may not be performed.

[0053] 16 , the FT unit 43 inputs the sequence information 16 and physical property information 19 of the second training data 502 to the protein language model 14B, and causes the protein language model 14B to output a training prescription prediction result 20L. Based on the training prescription prediction result 20L and prescription information 46, a loss calculation is performed for the protein language model 14B using a loss function. Then, the values ​​of various parameters of the protein language model 14B are updated according to the result of the loss calculation, and the protein language model 14B is updated according to the update settings.

[0054] The FT unit 43 repeatedly performs the above series of processes, including inputting the sequence information 16 and the physical property information 19 to the protein language model 14B, outputting the training prescription prediction result 20L from the protein language model 14B, calculating the loss, updating the settings, and updating the protein language model 14B, while changing the second training data 502, until the prediction accuracy of the training prescription prediction result 20L reaches a predetermined level. Alternatively, the FT unit 43 terminates the above series of processes after a predetermined number of repetitions. The protein language model 14B whose prediction accuracy of the training prescription prediction result 20L has thus reached a predetermined level, or the protein language model 14B for which the above series of processes has been repeated a predetermined number of times, is transmitted from the learning device 11 to the drug development support device 12 as the protein language model 14C after FT.

[0055] The sequence information 16 and the physical property information 19 (sequence information 16A and physical property information 19A in the case of FIG. 14 ) are converted into vector data in the protein language model 14. The vector data is data in which the alphabet of the sequence information 16 and the standard free energy change ΔG, the surface exposure area SASA, and the charge state of the physical property information 19 are each expressed as a multidimensional, for example, 64-dimensional vector.

[0056] 17 , an operating program 60 is stored in storage 25B of drug development support device 12. Operating program 60 is an AP for causing a computer to function as drug development support device 12. In other words, operating program 60 is an example of an "operating program for a drug development support device" according to the technology of the present disclosure. Storage 25B also stores a protein language model 14C, sequence information 16, physical property information 19, etc.

[0057] When the operating program 60 is started, the CPU 27B of the computer constituting the pharmaceutical development support device 12 works in cooperation with the memory 26 etc. to function as a request receiving unit 65, an RW control unit 66, a prediction unit 67, and a screen distribution control unit 68.

[0058] The request receiving unit 65 receives various requests from the user terminal 13. In particular, the request receiving unit 65 receives a prediction request 18 from the user terminal 13. The prediction request 18 includes the sequence information 16 and the physical property information 19, as described above. Therefore, by receiving the prediction request 18, the request receiving unit 65 acquires the sequence information 16 and the physical property information 19. When the prediction request 18 is received, the request receiving unit 65 outputs the sequence information 16 and the physical property information 19 included in the prediction request 18 to the RW control unit 66. Furthermore, the request receiving unit 65 outputs the terminal ID of the user terminal 13 included in the prediction request 18 to the screen distribution control unit 68.

[0059] The RW control unit 66 controls the storage of various data in the storage 25B and the reading of various data from the storage 25B. For example, the RW control unit 66 stores the protein language model 14C transmitted from the learning device 11 in the storage 25B. The RW control unit 66 also stores the sequence information 16 and the physical property information 19 from the request receiving unit 65 in the storage 25B. The RW control unit 66 reads the sequence information 16 and the physical property information 19 from the storage 25B and outputs the read sequence information 16 and the physical property information 19 to the prediction unit 67. The RW control unit 66 also reads the protein language model 14C from the storage 25B and outputs the read protein language model 14C to the prediction unit 67.

[0060] 18 , as an example, the prediction unit 67 inputs the sequence information 16 and the physical property information 19 into the protein language model 14C, and causes the protein language model 14C to output a prescription prediction result 20. The prediction unit 67 outputs the prescription prediction result 20 to the screen delivery control unit 68. Here, the prescription prediction result 20 is the hydrogen ion exponent of the preservative solution.

[0061] The screen delivery control unit 68 controls the delivery of various screens to the user terminal 13. Specifically, the screen delivery control unit 68 delivers and outputs various screens to the user terminal 13 that has sent the various requests in the form of screen data for web delivery created using a markup language such as XML (Extensible Markup Language). At this time, the screen delivery control unit 68 identifies the user terminal 13 that has sent the various requests based on the terminal ID from the request receiving unit 65. Note that instead of XML, other data description languages ​​such as JSON (Javascript (registered trademark) Object Notation) may be used.

[0062] The various screens include an information input screen 75 (see FIG. 20) for inputting sequence information 16 and physical property information 19, and a prediction result display screen 80 (see FIG. 21) for displaying prescription prediction results 20. In addition to these processing units 65 to 68, the CPU 27B also includes an instruction receiving unit that receives various operation instructions from the input device 30.

[0063] As an example, as shown in FIG. 19 , a prediction AP 70 is stored in the storage 25C of the user terminal 13. The prediction AP 70 is installed in the user terminal 13 by the user U. The prediction AP 70 is an AP for predicting information regarding an appropriate formulation prescription for a biopharmaceutical preservative solution. When the prediction AP 70 is launched, the CPU 27C of the user terminal 13 functions as a browser control unit 72 in cooperation with the memory 26 and the like. The browser control unit 72 controls the operation of a dedicated web browser for the prediction AP 70.

[0064] The browser control unit 72 reproduces various screens based on various screen data from the drug development support device 12 and displays the reproduced various screens on the display 29C. The browser control unit 72 also accepts various operation instructions input by the user U from the input device 30C via the various screens. The browser control unit 72 transmits various requests, including a prediction request 18, to the drug development support device 12 in response to the operation instructions.

[0065] When the prediction AP 70 is launched, an information input screen 75 shown in FIG. 20 is displayed on the display 29C under the control of the browser control unit 72. The information input screen 75 has an input box 76 for sequence information 16 and an input box 77 for physical property information 19. In the input box 76, the sequence information 16 can be written or a file for the sequence information 16 can be dropped. Similarly, in the input box 77, the physical property information 19 can be written or a file for the physical property information 19 can be dropped.

[0066] User U inputs desired sequence information 16 and physical property information 19 into input boxes 76 and 77, and then selects prediction button 78. When prediction button 78 is selected, browser control unit 72 generates prediction request 18 including sequence information 16 and physical property information 19 input into input boxes 76 and 77, and transmits generated prediction request 18 to drug development support device 12.

[0067] Furthermore, when the pharmaceutical development support device 12 predicts information regarding an appropriate formulation prescription for a biopharmaceutical preservative solution, a prediction result display screen 80, as an example shown in FIG. 21 , is displayed on the display 29C under the control of the browser control unit 72. The prediction result display screen 80 displays the prescription prediction result 20, or more precisely, a message representing the prescription prediction result 20. In this way, the prescription prediction result 20 is presented to the user U in the form of screen data distribution.

[0068] At the top of the prediction result display screen 80, there are provided a sequence information display button 81 and a physical property information display button 82. When the sequence information display button 81 is selected, a display screen for the sequence information 16 is displayed as a pop-up. Similarly, when the physical property information display button 82 is selected, a display screen for the physical property information 19 is displayed as a pop-up.

[0069] In addition, a save button 83 and an OK button 84 are provided at the bottom of the prediction result display screen 80. When the save button 83 is selected, the sequence information 16, the physical property information 19, and the prescription prediction result 20 are stored in association with each other in the storage 25C of the user terminal 13. When the OK button 84 is selected, the display of the prediction result display screen 80 is cleared.

[0070] Next, the operation of the above configuration will be described with reference to the flowcharts shown in Figures 22 and 23 as an example. When the operating program 35 is launched in the learning device 11, the CPU 27A of the learning device 11 functions as the RW control unit 40, the generation unit 41, the pre-learning unit 42, and the FT unit 43, as shown in Figure 5. When the operating program 60 is launched in the drug development support device 12, the CPU 27B of the drug development support device 12 functions as the request receiving unit 65, the RW control unit 66, the prediction unit 67, and the screen distribution control unit 68, as shown in Figure 17. When the prediction AP 70 is launched in the user terminal 13, the CPU 27C of the user terminal 13 functions as the browser control unit 72, as shown in Figure 19.

[0071] 22 , in the learning device 11, as shown in FIGS. 7 to 12 , the generation unit 41 generates a learning data group 37 from the past data group 36 (step ST100). The learning data 50 constituting the learning data group 37 includes the sequence information 16 and the physical property information 19. Therefore, by generating the learning data group 37 using the generation unit 41, the learning device 11 acquires the learning data 50 including the sequence information 16 and the physical property information 19. The learning data group 37 is output from the generation unit 41 to the RW control unit 40 and stored in the storage 25A under the control of the RW control unit 40.

[0072] The RW control unit 40 reads the protein language model 14A and the first training data group 371 from the storage 25A and outputs the read protein language model 14A and the first training data group 371 to the pre-training unit 42. As shown in FIGS. 13 to 15 , the pre-training unit 42 performs MLM as pre-training on the protein language model 14A using the first_1 training data group 371_1 (step ST110). Furthermore, the pre-training unit 42 performs NSP as pre-training on the protein language model 14A using the first_2 training data group 371_2 (step ST120). The protein language model 14B after pre-training is output from the pre-training unit 42 to the RW control unit 40 and stored in the storage 25A under the control of the RW control unit 40.

[0073] The RW control unit 40 reads the protein language model 14B and the second training data group 372 from the storage 25A and outputs the read protein language model 14B and second training data group 372 to the FT unit 43. As shown in FIG. 16 , the FT unit 43 performs FT on the protein language model 14B using the second training data group 372 (step ST130). The protein language model 14C after FT is output from the FT unit 43 to the RW control unit 40 and stored in the storage 25A under the control of the RW control unit 40. The protein language model 14C is then transmitted from the learning device 11 to the drug development support device 12 and stored in the storage 25B of the drug development support device 12.

[0074] 20 is displayed on display 29C of user terminal 13 under the control of browser control unit 72. When user U inputs desired sequence information 16 and physical property information 19 into input boxes 76 and 77 on information input screen 75 and selects prediction button 78, a prediction request 18 is sent from browser control unit 72 to pharmaceutical development support device 12. As shown in FIG. 1, prediction request 18 includes sequence information 16, physical property information 19, the terminal ID of user terminal 13, etc.

[0075] 23 , in pharmaceutical development support device 12, prediction request 18 is received by request receiving unit 65, whereby sequence information 16 and physical property information 19 included in prediction request 18 are acquired (YES in step ST200). Sequence information 16 and physical property information 19 included in prediction request 18 are output from request receiving unit 65 to RW control unit 66 and stored in storage 25B under the control of RW control unit 66 (step ST210). In addition, the terminal ID of user terminal 13 included in prediction request 18 is output from request receiving unit 65 to screen distribution control unit 68.

[0076] The sequence information 16 and the physical property information 19 are read from the storage 25B by the RW control unit 66 (step ST220). The sequence information 16 and the physical property information 19 are output from the RW control unit 66 to the prediction unit 67. The RW control unit 66 also reads the protein language model 14C from the storage 25B and outputs the read protein language model 14C to the prediction unit 67.

[0077] 18, in the prediction unit 67, the sequence information 16 and the physical property information 19 are input to the protein language model 14C. As a result, the prescription prediction result 20 is output from the protein language model 14C (step ST230). The prescription prediction result 20 is output from the prediction unit 67 to the screen delivery control unit 68.

[0078] The screen distribution control unit 68 generates screen data for the prediction result display screen 80 shown in Fig. 21 based on the prescription prediction result 20. Under the control of the screen distribution control unit 68, the screen data for the prediction result display screen 80 is distributed to the user terminal 13 that sent the prediction request 18 (step ST240).

[0079] In the user terminal 13, under the control of the browser control unit 72, the screen data of the prediction result display screen 80 is reproduced, and the reproduced prediction result display screen 80 is displayed on the display 29C. In this way, the prescription prediction result 20 is presented to the user U.

[0080] As described above, the CPU 27A of the learning device 11 includes a generation unit 41, a pre-learning unit 42, and an FT unit 43. The generation unit 41 generates and acquires training data 50 including sequence information 16 of amino acid residues constituting an antibody 15 contained in a biopharmaceutical and physical property information 19 of the amino acid residues. The pre-learning unit 42 and the FT unit 43 use the training data 50 to perform pre-learning and fine-tuning of a protein language model 14 that outputs a prescription prediction result 20.

[0081] The CPU 27B of the pharmaceutical development support device 12 also includes a request receiving unit 65, a prediction unit 67, and a screen distribution control unit 68. The request receiving unit 65 receives a prediction request 18 and thereby acquires sequence information 16 and physical property information 19. The prediction unit 67 uses a protein language model 14C that has been pre-trained and fine-tuned using training data 50 including training sequence information 16 and training physical property information 19. The prediction unit 67 inputs the sequence information 16 and the physical property information 19 into the protein language model 14C and causes the protein language model 14C to output a prescription prediction result 20. The screen distribution control unit 68 presents the prescription prediction result 20 to the user U by distributing screen data of a prediction result display screen 80 including the prescription prediction result 20 to the user terminal 13. Because not only the sequence information 16 but also the physical property information 19 is taken into consideration, the prediction accuracy of the protein language model 14C can be improved compared to when only the sequence information 16 is input to the protein language model 14C.

[0082] 8, 9, 13, and 14, the protein language model 14C is a model that has been subjected to MLM pre-training using the first_1 training data 501_1 in which the sequence information 16 and the physical property information 19 have been masked to produce sequence information 16A and physical property information 19A. This makes it possible to generate a protein language model 14C that can refer to not only the sequence information 16 but also the physical property information 19.

[0083] The physical property information 19 includes data representing the standard free energy change ΔG, a numerical value representing the hydrophilicity / hydrophobicity of the amino acid residue, the surface exposed area SASA of the amino acid residue, and the charge state of the amino acid residue. All of these greatly influence the appropriate formulation for stabilizing the quality of biopharmaceuticals. This can greatly improve the prediction accuracy of the protein language model 14C. The physical property information 19 may include at least one of the numerical value representing the hydrophilicity / hydrophobicity of the amino acid residue, the surface exposed area SASA of the amino acid residue, and data representing the charge state of the amino acid residue.

[0084] The prediction result is a prescription prediction result 20 that includes information about the formulation of the biopharmaceutical preservation solution. This can contribute to maintaining the stable quality of the biopharmaceutical. Note that the information about the formulation may include, instead of or in addition to the exemplified hydrogen ion exponent, the temperature of the preservation solution, the type of additive added to the preservation solution, or the concentration of the additive. Furthermore, the prediction result is not limited to information about the formulation. For example, the degree of aggregation of the antibody 15 may also be used.

[0085] Pharmaceuticals containing antibody 15 as a protein are called antibody drugs, and are widely used in the treatment of 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 15 is used as the protein, can further promote the development of antibody drugs that are widely used in the treatment of various diseases.

[0086] The protein is not limited to the exemplified antibody 15. It may be a cytokine (interferon, interleukin, etc.), a hormone (insulin, glucagon, follicle-stimulating hormone, erythropoietin, etc.), a growth factor (IGF (insulin-like growth factor)-1, bFGF (basic fibroblast growth factor), etc.), a blood coagulation factor (factor 7, factor 8, factor 9, etc.), an enzyme (lysosomal enzyme, DNA (deoxyribonucleic acid) degrading enzyme, etc.), an Fc (fragment crystalline) fusion protein, a receptor, albumin, or a protein vaccine. Antibodies also include bispecific antibodies, antibody-drug conjugates, low molecular weight antibodies, glycosylated antibodies, and the like.

[0087] Furthermore, substances derived from amino acids are not limited to proteins. They can also be peptides, nucleic acids, etc. Therefore, pharmaceuticals are not limited to biopharmaceuticals that require biotechnology such as genetic recombination technology and cell culture technology, but can also be peptide pharmaceuticals, nucleic acid pharmaceuticals, etc. that can be produced using only chemical synthesis technology without the need for biotechnology.

[0088] The physical property information 19 may include data on the acidity / basicity of amino acid residues, dissociation constants, etc. Furthermore, the data representing the charge state of amino acid residues may include a spatial charge map (SCM).

[0089] In addition to sequence information 16 and physical property information 19, information regarding the secondary, tertiary, and quaternary structure of the protein may also be input to protein language model 14.

[0090] The protein language model 14C may continue to be trained even after being stored in the storage 25B of the drug development support device 12.

[0091] Although an example has been shown in which the training data group 37 is generated by the generation unit 41 of the learning device 11, the present invention is not limited to this. The training data group 37 may be generated by a device other than the learning device 11, and the training data group 37 may be transmitted from the other device to the learning device 11. Furthermore, some or all of the functions of the learning device 11 may be performed by the drug development support device 12. Similarly, some or all of the functions of the drug development support device 12 may be performed by the user terminal 13.

[0092] The learning device 11 and the drug development support device 12 may be installed in a pharmaceutical company or a drug development contract organization, or may be installed in a data center independent of the pharmaceutical company or the drug development contract organization.

[0093] Instead of delivering screen data of the prediction result display screen 80 including the prescription prediction result 20 to the user terminal 13, the prescription prediction result 20 itself may be delivered to the user terminal 13. In this case, the user terminal 13 generates the prediction result display screen 80 based on the prescription prediction result 20 under the control of the browser control unit 72.

[0094] The method of presenting the prescription prediction result 20 to the user U is not limited to the example of delivering screen data. The prescription prediction result 20 may be presented to the user U by printing it on a paper medium, or by attaching it to an email and sending it to the user terminal 13.

[0095] The hardware configuration of the computers constituting the learning device 11 and the drug development support device 12 according to the technology of the present disclosure can be modified in various ways. For example, the learning device 11 and the drug development support device 12 can be configured as multiple computers separated as hardware to improve processing power and reliability. For example, the functions of the RW control unit 40 and the generation unit 41 and the functions of the pre-learning unit 42 and the FT unit 43 can be distributed across two computers. In this case, the learning device 11 is configured with two computers. Alternatively, the functions of the request reception unit 65 and the RW control unit 66 and the functions of the prediction unit 67 and the screen distribution control unit 68 can be distributed across two computers. In this case, the drug development support device 12 is configured with two computers.

[0096] In this way, the hardware configuration of the computers of learning device 11 and drug development support device 12 can be changed as appropriate depending on the required performance, such as processing power, safety, and reliability. Furthermore, not only the hardware, but also APs such as operating programs 35 and 60 can be duplicated or stored in multiple storage devices in order to ensure safety and reliability.

[0097] In the above embodiment, for example, the hardware structure of the processing units that perform various processes, such as the RW control units 40 and 66, the generation unit 41, the pre-learning unit 42, the FT unit 43, the request reception unit 65, the prediction unit 67, the screen distribution control unit 68, and the browser control unit 72, can be any of the various processors shown below. As described above, the various processors include the CPUs 27A, 27B, and 27C, which are general-purpose processors that execute software (operation programs 35 and 60, and prediction AP 70) and function as various processing units, as well as programmable logic devices (PLDs) that are processors whose circuit configuration can be changed after manufacture, such as FPGAs (Field Programmable Gate Arrays), and dedicated electrical circuits that are processors having a circuit configuration designed specifically for executing specific processing, such as ASICs (Application Specific Integrated Circuits).

[0098] A single processing unit may be configured with one of these various processors, or may be configured with a combination of two or more processors of the same or different types (e.g., a combination of multiple FPGAs and / or a combination of a CPU and an FPGA).Furthermore, multiple processing units may be configured with a single processor.

[0099] Examples of configuring multiple processing units with a single processor include, first, a form in which one processor is configured with a combination of one or more CPUs and software, as typified by computers such as client and server, and this processor functions as multiple processing units. Second, a form in which a processor is used to realize the functions of the entire system including multiple processing units with a single IC (Integrated Circuit) chip, as typified by systems on chips (SoCs). In this way, various processing units are configured using one or more of the above-mentioned various processors as a hardware structure.

[0100] Furthermore, more specifically, the hardware structure of these various processors can be an electric circuit (circuitry) that combines circuit elements such as semiconductor elements.

[0101] From the above description, the technology described in the following supplementary paragraphs can be understood.

[0102] [Supplementary Item 1] A drug development support device comprising a processor, which acquires sequence information of amino acid residues constituting a substance derived from amino acids contained in a drug and physical property information of the amino acid residues, uses a substance language model that has been pre-trained and fine-tuned using training data including training sequence information and training physical property information, inputs the sequence information and the physical property information into the substance language model, causes the substance language model to output a prediction result regarding the substance, and presents the prediction result to a user. [Supplementary Item 2] The drug development support device according to Supplementary Item 1, wherein the substance language model is a model that has been pre-trained using training data in which at least one of a portion of the sequence information and a portion of the physical property information is masked. [Supplementary Item 3] The drug development support device according to Supplementary Item 1 or Supplementary Item 2, wherein the physical property information is at least one of a numerical value representing the hydrophilicity / hydrophobicity of the amino acid residue, a surface exposed area of ​​the amino acid residue, and data representing the charge state of the amino acid residue. [Supplementary Item 4] The drug development support device according to any one of Supplementary Items 1 to 3, wherein the prediction result includes information about a prescription for a preservative solution of the drug. [Supplementary Item 5] The drug development support device according to any one of Supplementary Items 1 to 4, wherein the substance is any one of a protein, a peptide, and a nucleic acid. [Supplementary Item 6] The drug development support device according to Supplementary Item 5, wherein the protein is an antibody. [Supplementary Item 7] A learning device comprising a processor, wherein the processor acquires training data including sequence information of amino acid residues constituting a substance derived from amino acids contained in the drug and physical property information of the amino acid residues, and performs pre-training and fine-tuning of a substance language model that outputs a prediction result for the substance using the training data. [Supplementary Item 8] The learning device according to Supplementary Item 7, wherein the training data used for the pre-training is data in which at least one of the sequence information portion and the physical property information portion is masked.

[0103] The technology of the present disclosure can be appropriately combined with the various embodiments and / or various modified examples described above. Furthermore, it is not limited to the above embodiments, and various configurations can be adopted without departing from the spirit of the present disclosure. Furthermore, the technology of the present disclosure extends not only to programs, but also to storage media that non-temporarily store programs, and computer program products that include programs.

[0104] The above-described description and illustrations are a detailed explanation of the parts related to the technology of the present disclosure and are merely an example of the technology of the present disclosure. For example, the above description of the configuration, functions, actions, and effects is an explanation of an example of the configuration, functions, actions, and effects of the parts related to the technology of the present disclosure. Therefore, it goes without saying that unnecessary parts may be deleted, new elements may be added, or replacements may be made to the above-described description and illustrations within the scope of the gist of the technology of the present disclosure. Furthermore, to avoid confusion and facilitate understanding of the parts related to the technology of the present disclosure, the above-described description and illustrations omit explanations of common technical knowledge that do not require particular explanation to enable the implementation of the technology of the present disclosure.

[0105] In this specification, "A and / or B" is synonymous with "at least one of A and B." In other words, "A and / or B" means that it may be only A, only B, or a combination of A and B. Furthermore, in this specification, the same concept as "A and / or B" is also applied when three or more things are expressed by connecting them with "and / or."

[0106] All publications, patent applications, and technical standards mentioned in this specification are herein incorporated by reference to the same extent as if each individual publication, patent application, or technical standard was specifically and individually indicated to be incorporated by reference.

Claims

1. A pharmaceutical development support device comprising a processor, which acquires sequence information of amino acid residues constituting a substance derived from amino acids contained in a pharmaceutical, and physical property information of the amino acid residues, uses a substance language model that has been pre-trained and fine-tuned using learning data including learning sequence information and learning physical property information, inputs the sequence information and the physical property information to the substance language model, outputs a prediction result regarding the substance from the substance language model, and presents the prediction result to a user.

2. A pharmaceutical development support device as described in claim 1, wherein the substance language model is a model in which the pre-training has been performed using the training data in which at least one of the sequence information portion and the physical property information portion has been masked.

3. The pharmaceutical development support device of claim 1, wherein the physical property information is at least one of a numerical value representing the hydrophilicity / hydrophobicity of the amino acid residue, a surface exposed area of ​​the amino acid residue, and data representing the charge state of the amino acid residue.

4. A pharmaceutical development support device as described in claim 1, wherein the prediction result includes information regarding a prescription for a preservation solution of the pharmaceutical.

5. The pharmaceutical development support device according to claim 1, wherein the substance is any one of a protein, a peptide, and a nucleic acid.

6. The pharmaceutical development support device according to claim 5, wherein the protein is an antibody.

7. A method for operating a pharmaceutical development support device, comprising: acquiring sequence information of amino acid residues constituting a substance derived from amino acids contained in a pharmaceutical, and physical property information of the amino acid residues; using a substance language model that has been pre-trained and fine-tuned using training data including training sequence information and training physical property information; inputting the sequence information and the physical property information into the substance language model, and outputting a prediction result regarding the substance from the substance language model; and presenting the prediction result to a user.

8. An operating program for a pharmaceutical development support device that causes a computer to execute processes including: obtaining sequence information of amino acid residues that constitute substances derived from amino acids contained in pharmaceuticals, and physical property information of the amino acid residues; using a substance language model that has been pre-trained and fine-tuned using training data including sequence information for training and physical property information for training; inputting the sequence information and the physical property information into the substance language model, and outputting prediction results regarding the substance from the substance language model; and presenting the prediction results to a user.

9. A learning device comprising a processor, which acquires learning data including sequence information of amino acid residues constituting a substance derived from amino acids contained in a pharmaceutical and physical property information of the amino acid residues, and performs pre-learning and fine-tuning of a substance language model that outputs prediction results regarding the substance using the learning data.

10. The learning device according to claim 9, wherein the learning data used in the pre-learning is data in which at least one of the sequence information portion and the physical property information portion is masked.

11. A method for operating a learning device, comprising: acquiring learning data including sequence information of amino acid residues constituting a substance derived from amino acids contained in a pharmaceutical and physical property information of the amino acid residues; and performing pre-learning and fine-tuning of a substance language model that outputs prediction results regarding the substance using the learning data.

12. An operating program for a learning device that causes a computer to execute processes including: acquiring learning data including sequence information of amino acid residues that constitute substances derived from amino acids contained in pharmaceuticals and physical property information of said amino acid residues; and performing pre-learning and fine-tuning of a substance language model that outputs prediction results regarding said substances using said learning data.

Images