Information processing device, information processing system, and information processing method

Abstract

An information processing device according to one embodiment of the present disclosure comprises: a structure prediction unit that predicts a three-dimensional structure of a candidate compound on the basis of sequence information on the candidate compound, and obtains three-dimensional structure information on the candidate compound; a characteristic value prediction unit that predicts a characteristic value of the candidate compound on the basis of the sequence information and the three-dimensional structure information on the candidate compound, and obtains characteristic value information on the candidate compound; and a presentation unit that presents the characteristic value information and the three-dimensional structure information on the candidate compound to a user.

Description

Information processing device, information processing system, and information processing method

[0001] The present disclosure relates to an information processing device, an information processing system, and an information processing method.

[0002] In fields such as drug discovery, biomaterials development, and food development, there is a demand for high-throughput cycles for designing and evaluating new compounds with desired properties and functions, i.e., desired compounds. However, designing desired compounds requires repeated trial and error, including the synthesis and analysis of the compounds, which is extremely time-consuming, laborious, and costly.

[0003] Therefore, with the recent development of deep learning technology in image recognition and natural language processing, proposals have been made in the field of bioinformatics to solve the above-mentioned problems using machine learning. For example, Patent Document 1 proposes a method for predicting the three-dimensional structure of a protein using a neural network with an amino acid sequence as input.

[0004] US Patent Application Publication No. 2021 / 0166779

[0005] John Jumper et al., “Highly Accurate Protein Structure Prediction with AlphaFold”, Nature, 26 August 2021.Minkyung Beak et al., “Accurate Prediction of Protein Structures and Interactions using a Three-track Neural Network”, Science, 15 July 2021 (First Release).Hannes Stark et al., “EquiBind: Geometric Deep Learning for Drug Binding Structure Prediction”, Proceedings of the 39th International Conference on Machine Learning, 4 June 2022.

[0006] However, there remains a need for a technology that can achieve high throughput in the cycle of designing and evaluating desired compounds and that enables the efficient production of desired compounds.

[0007] Therefore, the present disclosure proposes an information processing device, an information processing system, and an information processing method that enable efficient production of desired compounds.

[0008] An information processing device according to one embodiment of the present disclosure includes a structure prediction unit that predicts a three-dimensional structure of a candidate compound based on sequence information of the candidate compound and obtains three-dimensional structure information of the candidate compound, a characteristic value prediction unit that predicts characteristic values ​​of the candidate compound based on the sequence information and three-dimensional structure information of the candidate compound and obtains characteristic value information of the candidate compound, and a presentation unit that presents the characteristic value information and three-dimensional structure information of the candidate compound to a user.

[0009] An information processing system according to one embodiment of the present disclosure includes a structure prediction unit that predicts a three-dimensional structure of a candidate compound based on sequence information of the candidate compound and obtains three-dimensional structure information of the candidate compound, a characteristic value prediction unit that predicts characteristic values ​​of the candidate compound based on the sequence information and three-dimensional structure information of the candidate compound and obtains characteristic value information of the candidate compound, and a presentation unit that presents the characteristic value information and three-dimensional structure information of the candidate compound to a user.

[0010] An information processing method according to one embodiment of the present disclosure is one in which an information processing device executes the following: predicting a three-dimensional structure of a candidate compound based on sequence information of the candidate compound, obtaining three-dimensional structure information of the candidate compound; predicting characteristic values ​​of the candidate compound based on the sequence information and three-dimensional structure information of the candidate compound, obtaining characteristic value information of the candidate compound; and presenting the characteristic value information and three-dimensional structure information of the candidate compound to a user.

[0011] FIG. 1 is a diagram showing an example of the configuration of an information processing system according to the first embodiment. FIG. 2 is a diagram showing an example of the configuration of an information processing device according to the first embodiment. FIG. 3 is a diagram showing an example of presentation of three-dimensional structure information and characteristic value information of a compound according to the first embodiment. FIG. 4 is a flowchart showing the flow of an example of a compound design process according to the first embodiment. FIG. 5 is a diagram for explaining structure / sequence prediction AI (Phase 1) according to the first embodiment. FIG. 6 is a diagram for explaining structure / sequence prediction AI (Phase 2) according to the first embodiment. FIG. 7 is a diagram for explaining characteristic value prediction AI (Phase 1) according to the first embodiment. FIG. 8 is a diagram for explaining characteristic value prediction AI (Phase 2) according to the first embodiment. FIG. 9 is a diagram for explaining characteristic value prediction AI (Phase 3) according to the first embodiment. FIG. 10 is a diagram showing an example of the configuration of an information processing device according to the second embodiment. FIG. 11 is a flowchart showing the flow of an example of a compound design process according to the second embodiment. FIG. 12 is a diagram for explaining interaction prediction AI (Phase 1) according to the second embodiment. FIG. 13 is a diagram for explaining interaction prediction AI (Phase 2) according to the second embodiment. FIG. 14 is a diagram showing an example of the configuration of an information processing device according to the third embodiment. FIG. 15 is a diagram for explaining calculation of prediction reliability according to the third embodiment. FIG. 16 is a diagram showing an example of the configuration of computer hardware.

[0012] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the drawings. Note that the embodiments do not limit the systems, devices, methods, etc. according to the present disclosure. Furthermore, in the following embodiments, essentially the same components are designated by the same reference numerals, and redundant explanations will be omitted. The following embodiments also include examples and modified examples.

[0013] One or more of the following embodiments can be implemented independently. However, at least a portion of the following embodiments may be implemented in appropriate combination with at least a portion of another embodiment. These embodiments may include novel features that are different from one another. Therefore, each embodiment may contribute to solving different purposes or problems, and may provide different effects. The effects of each embodiment are merely examples and are not intended to be limiting, and other effects may also be provided.

[0014] The present disclosure will be described in the following order: 1. First embodiment 1-1. Configuration example of information processing system 1-2. Configuration example of information processing device 1-3. Processing example of compound design 1-4. Neural network training method 2. Second embodiment 2-1. Configuration example of information processing device 2-2. Processing example of compound design 2-3. Neural network training method 3. Third embodiment 3-1. Configuration example of information processing device 3-2. Neural network training method 4. Actions and effects 5. Other embodiments 6. Configuration example of hardware 7. Supplementary notes

[0015] <1. First embodiment> <1-1. Configuration example of information processing system> An example of the configuration of an information processing system 1 according to a first embodiment will be described with reference to Fig. 1. Fig. 1 is a diagram showing an example of the configuration of the information processing system 1 according to the first embodiment. The information processing system 1 according to the first embodiment functions as a compound design system.

[0016] 1, an information processing system 1 according to the first embodiment includes an information processing device 10, a sequence / structure information DB 20, and a property information DB 30. The information processing device 10, the sequence / structure information DB 20, and the property information DB 30 are configured to be able to communicate (transmit / receive) various types of information via a network 40.

[0017] The information processing device 10 is a device for designing compounds such as organic compounds, and is used by a user such as a designer who designs compounds. The information processing device 10 executes a process for designing compounds using, for example, machine learning such as a neural network (e.g., various machine learning models). The information processing device 10 also generates, for example, various machine learning models for designing compounds. In generating the machine learning models, the information processing device 10 uses various pieces of information stored in the sequence / structure information DB 20 and the property information DB 30. The compound design process and the generation of the machine learning models will be described in detail below.

[0018] The information processing device 10 may be, for example, a local PC, a physical server, a cloud server, etc. Examples of the local PC include a desktop PC (personal computer), a notebook PC, a tablet device, and a smartphone. The information processing device 10 can access the sequence / structure information DB 20, the property information DB 30, etc. via a network 40.

[0019] The sequence / structure information DB 20 records and manages sequence information and three-dimensional structure information for each type of compound. This sequence / structure information DB 20 may be, for example, an on-premise database or a cloud-based database. For example, the database of the World Protein Data Bank (wwPDB) can be used as the sequence / structure information DB 20. The sequence information and three-dimensional structure information in the sequence / structure information DB 20 function as template information for the sequence and three-dimensional structure of a compound.

[0020] The sequence information of a compound is information about the sequence of the compound. This sequence information of a compound includes, for example, an alphabetic character string that indicates the sequence. Furthermore, the three-dimensional structure information of a compound is information about the three-dimensional structure of the compound (for example, the specific three-dimensional structure and function of the compound). The sequence information and three-dimensional structure information of a compound are paired pieces of information for one compound.

[0021] The property information DB 30 records and manages property information for each type of compound. Similar to the sequence / structure information DB 20, the property information DB 30 may be, for example, either an on-premise database or a cloud-based database. The property information in the property information DB 30 functions as template information for the properties of the compound.

[0022] The compound characteristic information is information about the characteristics of the compound. This compound characteristic information includes, for example, characteristic value information and experimental condition information. The characteristic value information is information about the characteristic values ​​of the compound. For example, the characteristic value information includes various characteristic values ​​corresponding to the compound. Examples of the various characteristic values ​​include denaturation temperature, solubility, pH characteristics, etc. The experimental condition information is information about the experimental conditions for compound production. The experimental conditions for compound production are the conditions of the experiment for producing the compound, and include, for example, various condition values ​​related to the experiment. Examples of the various condition values ​​include the temperature and humidity during the experiment, the experimental procedure, the drop volume, and the possibility of foreign matter contamination.

[0023] The network 40 may be either wireless or wired, or may be either a dedicated line or a public line, or may be any network that allows communication between the information processing device 10, the sequence / structure information DB 20, and the property information DB 30.

[0024] For example, the network 40 may be a communication network such as a local area network (LAN), a wide area network (WAN), a cellular network, a fixed telephone network, a regional Internet Protocol (IP) network, or the Internet. The network 40 may also include a core network. The core network may be, for example, an evolved packet core (EPC) or a 5G core network (5GC). The network 40 may also include a data network other than the core network. For example, the data network may be a service network of a telecommunications carrier, such as an IP Multimedia Subsystem (IMS) network. The data network may also be a private network, such as an in-house network. The network 40 may also be configured as a software-defined network (SDN).

[0025] 1, there is one information processing device 10, but there may be multiple devices, for example, one for each designer. Also, in the example of FIG. 1, there is one sequence / structure information DB 20 and one characteristic information DB 30, but there may be multiple devices.

[0026] Furthermore, although the generation of the machine learning model is executed by the information processing device 10, it may also be executed, for example, by a device other than the information processing device 10. In this case, the machine learning model generated by the other device is stored in a storage device such as a server, for example, and used by one or more information processing devices 10.

[0027] (Example of a compound) Here, an example of a compound is a protein. A protein is formed by, for example, several dozen types of amino acids linked in a line in various orders. The properties (functions) of this protein are exerted not only by the sequence of amino acids (primary structure) but also by the formation of a specific three-dimensional structure.

[0028] An amino acid sequence generally consists of several tens to several hundreds of amino acid residues. Representing these amino acid residues using a rational formula or the like would be extremely lengthy. Therefore, in order to concisely represent an amino acid sequence, a method is used in which the type of amino acid residue is represented by a single letter of the alphabet. For example, a serine residue is represented by "S" and a glutamine residue by "Q." In addition, a total of 20 types of amino acid residues are each represented by a single letter of the alphabet. Such a string of letters of the alphabet becomes, for example, sequence information. Of course, the sequence information may also include any other information related to the amino acid sequence.

[0029] The three-dimensional structure information is information including, for example, the specific three-dimensional structure and function of a protein, and includes at least one of the protein structure and the protein function.

[0030] A protein structure is information about the structure of a protein. For example, the protein structure includes information including a sequence of three-dimensional coordinates for each atom, molecule, bond, functional group, etc. that constitutes the protein. The sequence of three-dimensional coordinates is sometimes called volume data. Of course, the protein structure may include any information about the protein structure.

[0031] The function of a protein includes, for example, at least one of the hydrophilicity or rigidity of the protein. Some proteins have localized hydrophilicity in part of their structure. Other proteins have localized rigidity (difficulty in bending). Functional labels representing such hydrophilicity or rigidity are included, for example, in the function of a protein. A functional label is, for example, a numerical value representing the range of three-dimensional coordinates having hydrophilicity or rigidity and the level of hydrophilicity or rigidity. Conversely, a numerical value representing the range of three-dimensional coordinates having hydrophobicity or non-rigidity may also be included in the functional label. Furthermore, when a protein has a localized Y-shaped structure, the arm of the Y may exhibit the function of capturing viruses. A functional label representing such an immune function may also be included in the function of a protein. Of course, the function of a protein may include any information related to the function of the protein.

[0032] <1-2. Configuration example of information processing device> A configuration example of the information processing device 10 according to the first embodiment will be described with reference to Fig. 2 and Fig. 3. Fig. 2 is a diagram showing a configuration example of the information processing device 10 according to the first embodiment. Fig. 3 is a diagram showing a presentation example of three-dimensional structure information and characteristic value information of a compound according to the first embodiment.

[0033] As shown in FIG. 2, the information processing device 10 according to the first embodiment includes a structure prediction unit 11, a characteristic value prediction unit 12, a selection unit 13, and a sequence prediction unit 14.

[0034] The structure prediction unit 11 receives sequence information (compound sequence information A1) of a compound that is a candidate for design (candidate compound), predicts the three-dimensional structure of the compound from a sequence based on the sequence information, obtains three-dimensional structure information regarding the predicted three-dimensional structure, and sends it to the characteristic value prediction unit 12. For example, the structure prediction unit 11 receives an input sequence in which the molecules of the compound are expressed as a string of characters, and predicts the three-dimensional structure using a neural network (e.g., a machine learning model). Then, the structure prediction unit 11 sends three-dimensional structure information that represents the predicted three-dimensional structure of the compound as a three-dimensional coordinate sequence of atoms to the characteristic value prediction unit 12, for example.

[0035] The neural network of the structure prediction unit 11 can be a method that has already been published in a paper or the like. For example, the neural network can be configured using AlphaFold, which is published in Non-Patent Document 1, or RoseTTAFold, which is published in Non-Patent Document 2. However, the neural network configuration is not limited to these.

[0036] When the candidate compound to be designed is a protein, the sequence information is an amino acid sequence, but is not limited to this. That is, the sequence information is not limited to amino acids, and may be a character string representing the molecular formula of a nucleic acid, polymer, or the like. Furthermore, the sequence information may use the sequence of a known compound as a template, or may be a character string newly prepared by the designer.

[0037] The property value prediction unit 12 predicts various property values ​​of a compound from the sequence information and three-dimensional structure information of the compound, acquires property value information relating to the various predicted property values, and sends it to the selection unit 13. For example, the property value prediction unit 12 receives the sequence information and three-dimensional structure information of the compound as input, and predicts the chemical or physical property values ​​of the compound using a neural network (e.g., a machine learning model).

[0038] The neural network of the characteristic value prediction unit 12 is, for example, a multi-head neural network that has multiple output layers and predicts multiple characteristic values. However, the configuration of the neural network is not limited to this. For example, denaturation temperature, solubility, pH characteristics, biosynthesis cost (molecular synthesis cost), torsional rigidity, etc. can be used as characteristic values. Furthermore, for example, an active site, which indicates the location of the active center within a compound, can also be used as a characteristic value.

[0039] The selection unit 13 has a presentation unit 13 a, an input unit 13 b, and a processing unit 13 c. The selection unit 13 presents predicted three-dimensional structure information and characteristic value information of the compound to the designer using the presentation unit 13 a, accepts an input operation from the designer using the input unit 13 b, and selects the presented compound as a desired compound or a next candidate compound using the processing unit 13 c in accordance with the accepted input operation (request) from the designer.

[0040] The presentation unit 13a visualizes the three-dimensional structure information and characteristic value information of the predicted compound and presents it to the designer. This presentation unit 13a may be, for example, a display unit or a printer. The display unit may be, for example, a liquid crystal display, an organic electroluminescence (EL) display, a head-mounted display, or the like. A plurality of display units may be provided depending on the application.

[0041] The input unit 13b accepts input operations from a user such as a designer. This input unit 13b may be, for example, a keyboard, a mouse, operation keys, or a touch panel. When a touch panel is used, the user performs various operations by touching the touch panel with a finger or a stylus. The input unit 13b may also be a voice input device (for example, a microphone) that accepts input operations by voice from the user.

[0042] The processing unit 13c determines a compound (candidate compound) being presented by the presentation unit 13a as a desired compound in response to a user's input operation on the input unit 13b. The processing unit 13c also changes the three-dimensional structure information of the compound (candidate compound) being presented by the presentation unit 13a in response to a user's input operation on the input unit 13b, and determines the presented compound with the changed three-dimensional structure as a next candidate compound.

[0043] 3, the presentation unit 13a is, for example, a display unit, which visualizes the three-dimensional structure information predicted by the structure prediction unit 11 as a three-dimensional structure image A2 and the characteristic value information predicted by the characteristic value prediction unit 12 as a table A3, and presents them to a user (designer). The three-dimensional structure image A2 shows the three-dimensional structure, and the table A3 shows various characteristic values. This allows the designer to visually evaluate candidate compounds on the information processing device 10 based on the predicted three-dimensional structure information and various characteristic values.

[0044] For example, if the designer looks at the three-dimensional structure image A2 and table A3 presented by the presentation unit 13a and determines that the displayed candidate compound is a desired compound, the designer operates the input unit 13b, such as the keyboard 131 or the mouse 132, to select the displayed candidate compound as the desired compound, thereby enabling the designer to obtain information such as three-dimensional structure information and sequence information of the desired compound.

[0045] Furthermore, if the designer determines that the displayed candidate compound is not the desired compound, the designer operates the input unit 13b to edit the three-dimensional structure, for example, on the display screen of the presentation unit 13a. While viewing the predicted three-dimensional structure, the designer can operate the input unit 13b to deform, for example, by bending a partial region of the three-dimensional structure. At this time, the processing unit 13c updates the three-dimensional coordinate sequence of the atoms according to the edited three-dimensional structure. When the designer finishes editing the three-dimensional structure, the processing unit 13c sends three-dimensional structure information of the next candidate, representing the updated three-dimensional coordinate sequence of the atoms, to the sequence prediction unit 14.

[0046] 2 , the sequence prediction unit 14 predicts the sequence of a compound based on the three-dimensional structure information of the compound (next candidate compound) selected by the selection unit 13, obtains sequence information on the predicted sequence, and sends it to the structure prediction unit 11. For example, the sequence prediction unit 14 translates the three-dimensional structure information of the compound selected by the selection unit 13 into sequence information of the compound. The sequence prediction unit 14 receives, for example, a three-dimensional atomic coordinate sequence as input, and translates (converts) the three-dimensional atomic coordinate sequence into the molecular sequence of the compound using a neural network (e.g., a machine learning model).

[0047] The neural network of the sequence prediction unit 14 may be, for example, a graph transformer or a graph autoencoder that represents a three-dimensional atomic coordinate sequence as an interatomic graph and translates it into a sequence, but the configuration of the neural network is not limited to these.

[0048] The sequence information predicted by the sequence prediction unit 14 is again input into the structure prediction unit 11 and the characteristic value prediction unit 12, and the selection unit 13 indicates to the designer whether the results (e.g., three-dimensional structure information, characteristic value information, etc.) edited by the designer are expected. As described above, for example, if a candidate compound is a desired compound, the designer operates the input unit 13b to select the candidate compound as the desired compound. This allows the designer to obtain information such as three-dimensional structure information and sequence information of the desired compound.

[0049] (Three-dimensional structure image) Here, the three-dimensional structure image A2 may be, for example, a point cloud image, a polygon image, a mesh image, a surface image, a slice image, a three-view drawing, etc. The specific display format of the three-dimensional structure image A2 is not limited.

[0050] A point cloud image is an image in which data is represented by a set of points. For example, each atom contained in a protein is represented by a point and displayed as a point cloud image. Specifically, the positions of points in the point cloud image are calculated based on the three-dimensional coordinates of the atoms contained in the three-dimensional structure, and the point cloud image is generated. Note that, in addition to atoms, molecules, functional groups, functional labels, or the main chains and side chains of proteins may also be represented by points and displayed as a point cloud image. Alternatively, points may be displayed in different colors depending on the type of atom or functional label. Note that a point cloud is sometimes called a point cloud.

[0051] A polygon image is an image in which data is represented by polygons. For example, the local shape of a protein is represented by triangles and quadrangles. A mesh image is an image in which data is represented by multiple polygons. For example, the shape of a protein is represented by a shape made up of connected triangles and quadrangles. A mesh image can also be considered a collection of polygon images. A surface image is an image in which data is represented by smooth curved surfaces. For example, the shape of a protein is represented by smooth curved surfaces.

[0052] A slice image is an image that represents a cross section of a protein. For example, a cross section at a predetermined position of a point cloud image is displayed as a slice image. Alternatively, a cross section of a polygon image, mesh image, or surface image may be displayed. A three-view image is an image that represents the shape of a protein when viewed from three directions. For example, a three-view image may include views from any direction, such as a front view, top view, bottom view, right side view, left side view, and back view, with a predetermined surface of the protein being the front.

[0053] Displaying the three-dimensional structure image A2 in these display formats allows the user to intuitively grasp the structure of the protein, etc. Furthermore, the slice image allows the user to easily grasp the internal structure of the protein (a structure that cannot be seen from the outside). Note that the display format, the position of the cross section in the slice image, the orientation in the three-view view, etc. can be appropriately changed by the user via the input unit 13b, for example.

[0054] (Editing of 3D Structure) The input operation related to editing the 3D structure includes, for example, at least one of an editing operation of the protein structure and an editing operation of the protein function. Note that, in order to realize the editing operation, any GUI (Graphical User Interface) such as various windows, buttons, check boxes, tabs, input fields, etc. may be arranged.

[0055] For example, changing the arrangement of atoms is editing the protein structure, and an operation corresponding to this editing is realized by, for example, dragging a point representing an atom in the three-dimensional structure image A2. Furthermore, for example, not only changing the arrangement of atoms but also editing such as new arrangement, deletion, selection, and changing the type (alpha carbon, beta carbon, oxygen, nitrogen, etc.) of atoms are possible. These editing operations are realized by clicking or dragging a point representing an atom in the three-dimensional structure image A2. At this time, input information corresponding to the editing, such as "delete atom" or "the new type of atom is carbon," is input to the processing unit 13c. Alternatively, similar editing may be possible for molecules, functional groups, protein main chains, side chains, etc. In this case, editing such as deformation of molecules, etc. may also be possible.

[0056] It may also be possible to arrange atoms and the like collectively in a desired region. That is, in addition to the method of pinpoint-locating atoms and the like at a single point, a method may be adopted in which a desired region is designated by a drag operation and atoms and the like are arranged collectively in the entire region. Similarly, it may be possible to select, move, delete, etc. all atoms and the like in a region collectively.

[0057] Furthermore, the bond relationship between atoms may be editable. For example, two atoms may be specified by clicking, a bond type selection screen may be displayed by right-clicking, and a desired type (such as a hydrogen bond) may be selected using a checkbox or the like. Alternatively, only the skeleton (rough shape) of the protein may be specified by the user, and the detailed arrangement of atoms and the like may be automatically determined according to the specified skeleton. Alternatively, only the skeleton (rough shape) of the protein may be specified by the user, and the detailed arrangement of atoms and the like may be automatically determined according to the specified skeleton.

[0058] The assignment of functional labels is, for example, editing of protein functions, and it is possible to locally assign functional labels representing functions such as hydrophilic, hydrophobic, rigid, and non-rigid. The operation corresponding to this editing is realized, for example, by selecting a desired area in the three-dimensional structure image A2 by dragging, and then selecting the functional label to be assigned using a checkbox or the like. At this time, input information corresponding to the editing, such as information such as "the new function of the functional label is hydrophilic, and the coordinate range is X = 10 to 20, Y = 10 to 30, Z = 20 to 40," is input to the processing unit 13c.

[0059] When a functional label is assigned, the arrangement of atoms and other elements is automatically determined based on the assigned functional label. For example, if a functional label of "hydrophilic" is assigned to a certain region, the arrangement of atoms and other elements within the region is automatically determined so that the protein in that region has a "hydrophilic" function. This makes it possible to assign a function even when a user wants to give a protein a desired function but does not know how to arrange atoms and other elements.

[0060] <1-3. Example of Compound Design Processing> An example of compound design processing according to the first embodiment will be described with reference to Fig. 4. Fig. 4 is a flowchart showing the flow of the example of compound design processing according to the first embodiment.

[0061] 4 , in step S11, the structure prediction unit 11 predicts the three-dimensional structure of the compound from the input sequence information of the compound and acquires the three-dimensional structure information of the compound. In step S12, the property value prediction unit 12 predicts the property values ​​of the compound from the input sequence information of the compound and the predicted three-dimensional structure information and acquires the property value information of the compound. In step S13, the presentation unit 13a presents the three-dimensional structure and property values ​​of the compound to the designer based on the acquired three-dimensional structure information and property value information. In step S14, the processing unit 13c determines whether the displayed compound (candidate compound) is a desired compound in response to the designer's input operation on the input unit 13b.

[0062] For example, if the displayed compound is not the desired compound, the designer operates the input unit 13b to change the sequence information input to the structure prediction unit 11 to other sequence information or edit the three-dimensional structure of the displayed compound. Also, if the displayed compound is the desired compound, the designer operates the input unit 13b to inform the processing unit 13c that the displayed compound is the desired compound.

[0063] In step S14, if the processing unit 13c determines that the displayed compound is not the desired compound in response to the designer's input operation on the input unit 13b (No in step S14), in step S15, the processing unit 13c changes the sequence information of the compound to be input to the structure prediction unit 11 to other sequence information or updates the three-dimensional structure of the displayed compound in response to the designer's input operation on the input unit 13b, thereby determining a next candidate compound.

[0064] On the other hand, in step S14, if the processing unit 13c determines that the displayed compound is a desired compound in response to the designer's input operation on the input unit 13b (Yes in step S14), then in step S16, the processing unit 13c determines that the displayed compound is the desired compound.

[0065] <1-4. Neural Network Training Method> A neural network training method according to the first embodiment will be described with reference to Fig. 5 to Fig. 9. Fig. 5 and Fig. 6 are diagrams for explaining the structure / sequence prediction AI (artificial intelligence) according to the first embodiment. Fig. 7 to Fig. 9 are diagrams for explaining the characteristic value prediction AI according to the first embodiment.

[0066] The structure / sequence prediction AI includes, for example, a structure prediction unit 11 and a sequence prediction unit 14, and the characteristic value prediction AI includes, for example, a characteristic value prediction unit 12 (see FIG. 2). The learning flow for each will be explained in order. Note that neural network learning refers to the process of adjusting parameters (for example, weights and biases) so that the output layer produces the result desired by humans (the correct answer).

[0067] (Structure / Sequence Prediction AI) Learning of the structure / sequence prediction AI is divided into two phases (Phase 1 and Phase 2). Phase 1 will be described with reference to FIG. 5, and Phase 2 will be described with reference to FIG. 6. Each prediction model in FIGS. 5 and 6 is a machine learning model (AI model). These machine learning models are realized by machine learning such as a neural network.

[0068] First, in phase 1 shown in FIG. 5, the structure prediction model 11a and the sequence prediction model 14a collect training data, which are pairs of compound sequence information B1 and compound structure information (three-dimensional structure information of the compound) B2, from the sequence / structure information DB 20, and perform parallel asynchronous learning.

[0069] 5, the structure prediction model 11a predicts the three-dimensional structure of a compound from compound sequence information B1 and outputs predicted structure information B3, which is three-dimensional structure information related to the predicted three-dimensional structure. Also, the sequence prediction model 14a predicts a sequence from compound structure information B2 and outputs predicted sequence information B4 related to the predicted sequence.

[0070] For example, the structure prediction model 11a performs training to minimize the error between the compound structure information B2 in the sequence / structure information DB 20 and the predicted structure information B3 (error minimization 11b). Furthermore, for example, the sequence prediction model 14a performs training to minimize the error between the compound sequence information B1 in the sequence / structure information DB 20 and the predicted sequence information B4 (error minimization 14b). After each training convergence, the next, latter phase 2 of training is performed. The convergence of the training is determined, for example, by the designer and is determined in response to the designer's input operation on the input unit 13b.

[0071] Next, in Phase 2 shown in Figure 6, only compound sequence information B1 is used to train the learning model. First, compound sequence information B1 in the sequence / structure information DB 20 is input to the structure prediction model 11a, which predicts a three-dimensional structure from the compound sequence information B1. Predicted structure information B3 relating to the predicted three-dimensional structure is input to the sequence prediction model 14a, which predicts a sequence from the predicted structure information B3. An error is evaluated between predicted sequence information B4 relating to the predicted sequence and the compound sequence information B1 in the sequence / structure information DB 20 (error minimization 14c). The error value is used to update the parameters of the structure prediction model 11a and the sequence prediction model 14a using the error backpropagation method.

[0072] By repeating this cycle of error minimization 14c, the vast amount of sequence information with unknown structures that exists in the world can be used for learning. Furthermore, by updating the parameters of sequence prediction model 14a and structure prediction model 11a in an end-to-end manner simultaneously, consistent learning can be achieved in which the structure prediction results are translated into the original sequence.

[0073] End-to-end learning is, for example, in deep learning, where weights in all layers of a neural network, from the input layer (end) to the output layer (end), are learned at once for one problem or task. In other words, the structure prediction model 11a and the sequence prediction model 14a are generated by end-to-end machine learning.

[0074] Here, for example, the structure prediction unit 11 shown in Fig. 2 includes the trained structure prediction model 11a described above, and is capable of predicting a three-dimensional structure from the sequence information of a compound (compound sequence information A1) using the structure prediction model 11a. Also, for example, the sequence prediction unit 14 shown in Fig. 2 includes the trained sequence prediction model 14a described above, and is capable of predicting a sequence from the three-dimensional structure information of a compound using the sequence prediction model 14a.

[0075] (Characteristic Value Prediction AI) Learning of the characteristic value prediction AI is divided into three phases (phase 1, phase 2, and phase 3). Phase 1 will be described with reference to Fig. 7, phase 2 with reference to Fig. 8, and phase 3 with reference to Fig. 9. Each of the prediction models in Fig. 7, Fig. 8, and Fig. 9 is a machine learning model.

[0076] 7 , training data for learning the characteristic values ​​of compounds is selected by machine learning. The information processing device 10 has an experimental reliability prediction model 15. For example, the characteristic value prediction unit 12 of the information processing device 10 may include the experimental reliability prediction model 15.

[0077] The experimental reliability prediction model 15 collects, from a characteristic information DB 30 related to the characteristics of the compounds, characteristic value information C1 related to the characteristic values ​​of the compounds and experimental condition information C2, which is information on the experimental conditions under which the characteristic values ​​were obtained, and performs learning. In the example of Fig. 7, the experimental reliability prediction model 15 predicts experimental reliability C3 from the characteristic value information C1 and the experimental condition information C2, and outputs the predicted experimental reliability C3.

[0078] For example, the experiment reliability prediction model 15 performs learning by performing error minimization 15a to regress a reliability score C4 from the characteristic value information C1 and the experimental condition information C2 of the compound, and outputs the experiment reliability C3. After the learning has converged, the experiment reliability prediction model 15 can evaluate whether the experiment is valid as the experiment reliability C3 from the characteristic value information C1 and the experimental condition information C2 in the characteristic information DB 30.

[0079] Examples of information about the experimental conditions include the temperature and humidity during the experiment, the experimental procedure, the amount of dripping, and the possibility of contamination. Furthermore, the property values ​​accumulated through experimental analysis of compounds can vary greatly due to subtle changes in the experimental conditions. Therefore, an evaluator who is an expert in chemical synthesis evaluates the appropriateness of the property values ​​while referring to the experimental conditions, and the degree of evaluation is defined as a reliability score C4.

[0080] The information processing device 10 includes an evaluation unit 16. The evaluation unit 16 includes a presentation unit 16a, an input unit 16b, and a processing unit 16c. The evaluation unit 16 presents characteristic value information C1 and experimental condition information C2 to the designer using the presentation unit 16a, accepts input operations from the designer using the input unit 16b, and determines a reliability score C4 using the processing unit 16c in response to the accepted input operation (request) from the designer. For example, the designer determines a higher reliability score C4 as the reliability of the experimental conditions increases.

[0081] The presentation unit 16a corresponds to the presentation unit 13a, the input unit 16b corresponds to the input unit 13b, and the processing unit 16c corresponds to the processing unit 13c. In other words, the evaluation unit 16 corresponds to the selection unit 13, but the processing unit 16c performs a process of determining the reliability score C4 in addition to the process related to the processing unit 13c.

[0082] The processing unit 16c determines the reliability score C4 in response to an input operation by the designer via the input unit 16b. For example, the designer operates the input unit 13b to evaluate whether the characteristic values ​​are appropriate while referring to the experimental conditions, and determines the reliability score C4.

[0083] 8, the property value prediction model 12a collects and learns compound sequence information C5, compound structure information (three-dimensional structure information of the compound) C6, and various property values ​​C7 from the property information DB 30. In the example of Fig. 8, the property value prediction model 12a predicts property values ​​from the compound sequence information C5 and compound structure information C6, and outputs property value information C8 related to the various predicted property values.

[0084] The characteristic value prediction model 12a is realized, for example, by a multi-head neural network having multiple output layers for predicting various characteristic values, such as an active site in the sequence C8a, a denaturation temperature C8b, a solubility C8c, a pH characteristic C8d, and a biosynthetic cost C8e.

[0085] For example, the property value prediction model 12a predicts various property values ​​of a compound from compound sequence information C5 and compound structure information C6, and performs learning while comparing property value information C8 relating to the various predicted property values ​​with various property values ​​C7 of training data collected from the property information DB 30. At this time, the experimental reliability prediction model 15 trained in the first half of Phase 1 is used in parallel.

[0086] For example, the experimental reliability prediction model 15 evaluates the validity of various characteristic values ​​C7 used as training data as experimental reliability C3, and the characteristic value prediction model 12a performs weighted error minimization 12b in learning, placing more importance on data with higher experimental reliability C3. This makes it possible to achieve more valid learning from characteristic value information C8 that has a large variance in experimental analysis.

[0087] In phase 3 shown in FIG. 9, the experimental condition information C2 and various characteristic values ​​C7 are known, but data on a compound whose three-dimensional structure is unknown is added, and the learning parameters are further tuned (fine tuning).

[0088] In this tuning, predicted structure information B3 regarding the three-dimensional structure predicted by the structure prediction model 11a is input to the property value prediction model 12a. Then, the error between the property value information C8 regarding the predicted property value and the correct property values ​​C7 is fed back to the structure prediction model 11a and the property value prediction model 12a by the error backpropagation method, thereby updating the learning parameters of the structure prediction model 11a and the property value prediction model 12a. This makes it possible to further improve the accuracy of the structure prediction model 11a and the property value prediction model 12a using information on compounds whose three-dimensional structures are unknown.

[0089] Here, for example, the characteristic value prediction unit 12 shown in FIG. 2 includes the trained characteristic value prediction model 12 a described above, and it is possible to predict various characteristic values ​​(including active sites) of a compound from the sequence information (compound sequence information A1) and three-dimensional structure information of the compound using the characteristic value prediction model 12 a.

[0090] By using the information processing system 1 according to the first embodiment, i.e., the compound design system, a user (a designer) can visually and interactively design a desired compound (e.g., a new compound that achieves a desired structure and desired properties), and can also improve the efficiency of designing the desired compound (e.g., the efficiency of developing a new compound).

[0091] <2. Second embodiment> <2-1. Configuration example of information processing device> An example of the configuration of an information processing device 10A according to the second embodiment will be described with reference to Fig. 10. Fig. 10 is a diagram showing an example of the configuration of the information processing device 10A according to the second embodiment.

[0092] As shown in FIG. 10 , the information processing device 10A according to the second embodiment includes a structure prediction unit (second structure prediction unit) 21, an active site prediction unit 22, a conformation prediction unit 23, and a stability prediction unit 24, in addition to the structure prediction unit (first structure prediction unit) 11, characteristic value prediction unit 12, selection unit 13, and sequence prediction unit 14 according to the first embodiment.

[0093] The structure prediction unit 21 predicts the three-dimensional structure of the second compound based on the sequence information of the second compound (second compound sequence information A1b), and sends the three-dimensional structure information regarding the predicted three-dimensional structure to the active site prediction unit 22. For example, the structure prediction unit 21 may use the same prediction method as the structure prediction unit 11 according to the first embodiment. Note that the structure prediction unit 11 according to the first embodiment receives input of the sequence information of the first compound (first compound sequence information A1a). The first compound and the second compound may be, for example, a protein, a glycoprotein, or a nucleic acid.

[0094] The active site prediction unit 22 predicts the active site of the second compound based on the sequence information and three-dimensional structure information of the second compound, and sends active site information regarding the predicted active site to the conformation prediction unit 23. The active site is one of various characteristic values. Therefore, the active site prediction unit 22 may use the same prediction method as the characteristic value prediction unit 12 according to the first embodiment.

[0095] The conformation prediction unit 23 predicts the conformation of a complex of the first compound and the second compound (e.g., a docked conformation) based on the conformation information and active site information of the first compound and the conformation information and active site information of the second compound, and sends conformation information on the predicted conformation of the complex to the stability prediction unit 24 and the selection unit 13. For example, the conformation prediction unit 23 predicts the conformation of a complex in which the first compound (e.g., a candidate compound) and the second compound (e.g., a protein) are bound using a neural network (e.g., a machine learning model).

[0096] The neural network of the conformation prediction unit 23 may be a method that has already been published in a paper or the like. For example, EquiBind, which is published in Non-Patent Document 3, may be used as the neural network configuration. However, the configuration of the neural network is not limited to this.

[0097] The stability prediction unit 24 predicts the stability of the complex based on the conformational information of the complex, and sends stability information (e.g., docking score) regarding the stability of the complex to the selection unit 13. For example, the stability prediction unit 24 predicts the stability of the complex of the first compound and the second compound as a docking score using a neural network (e.g., a machine learning model).

[0098] The neural network of the stability prediction unit 24 may be, for example, a graph neural network that represents the first compound and the second compound using a graphical model to predict the stability of the complex, although the configuration of the neural network is not limited to this.

[0099] With this configuration, the presentation unit 13a visualizes and presents to the user the conformation information of the complex predicted by the conformation prediction unit 23 and the stability information of the complex predicted by the stability prediction unit 24, in addition to the three-dimensional structure information of the first compound predicted by the structure prediction unit 11 and the characteristic value information predicted by the characteristic value prediction unit 12. This allows the designer to visually evaluate candidate compounds on the information processing device 10 based on the predicted three-dimensional structure information and characteristic value information of the first compound, as well as the conformation information and stability information of the complex. Thereafter, the flow of compound design by the designer is the same as in the first embodiment.

[0100] (Graphical Model) Here, the aforementioned graphical model is a graph that represents a probability dependency. Specifically, a graphical model is composed of multiple nodes and multiple edges. Nodes are connected to each other by edges, and typically, nodes are represented as circles and edges are represented as lines connecting the nodes. For example, the length of the edge connecting two nodes is determined according to the magnitude of some probability associated with the two nodes. When the probability is relatively large, the distance between the edges is relatively short, and when the probability is relatively small, the distance between the edges is relatively long.

[0101] For example, a graphical model is created by regarding atoms as nodes and the probability of atoms bonding to each other as edges. For example, if the probability of a first atom bonding to a second atom is high, the node representing the first atom and the node representing the second atom are connected by a short edge. On the other hand, if the probability of bonding is low, the nodes are connected by a long edge. It is known that the probability of atoms bonding to each other depends on the distance between the atoms.

[0102] For example, when the distance between atoms is short, the probability that the atoms are bonded to each other is high. On the other hand, when the distance is long, the probability that the atoms are bonded to each other is low. In other words, a graphical model may be created by regarding the distance between atoms as an edge. In this case, when the distance between atoms is long, the nodes are connected by a long edge. This also means that the probability that the atoms are bonded to each other is low. Conversely, when the distance between atoms is short, the nodes are connected by a short edge. This also means that the probability that the atoms are bonded to each other is high.

[0103] Alternatively, atoms may be connected by an edge only if the distance between them is shorter than a predetermined threshold (e.g., 10 angstroms). Such atom pairs whose distance is shorter than the threshold (considered to be in contact) are sometimes called contact atom pairs. Furthermore, functional labels may be embedded in nodes and edges. That is, node features and edge features may be generated based on the functional labels. Other than that, the specific method for generating a graphical model is not limited.

[0104] <2-2. Example of Compound Design Processing> An example of compound design processing according to the second embodiment will be described with reference to Fig. 11. Fig. 11 is a flowchart showing the flow of the example of compound design processing according to the second embodiment.

[0105] As shown in FIG. 11 , steps S11 to S16 according to the second embodiment are basically the same as steps S11 to S16 according to the first embodiment, but the amount of information presented in step S13a according to the second embodiment (e.g., three-dimensional structure, characteristic value, three-dimensional conformation, docking score) is greater than the amount of information presented in step S11 according to the first embodiment (e.g., three-dimensional structure, characteristic value).

[0106] In step S21, the structure prediction unit 21 predicts the three-dimensional structure of the second compound from the input sequence information of the second compound, and obtains three-dimensional structure information of the second compound relating to the predicted three-dimensional structure.

[0107] In step S22, the active site prediction unit 22 predicts the active site of the second compound from the input sequence information of the second compound and the predicted three-dimensional structure information of the second compound, and acquires active site information (an example of characteristic value information) of the second compound.

[0108] In step S23, the conformation prediction unit 23 predicts the conformation of a complex of the first compound and the second compound from the three-dimensional structure information and active site information of the first compound and the obtained three-dimensional structure information and active site information of the second compound, and obtains conformation information of the predicted complex.

[0109] In step S24, the stability prediction unit 24 predicts the stability (docking score) of the complex from the acquired conformational information of the complex, acquires stability information regarding the predicted stability of the complex, and proceeds to step S13a.

[0110] In step S13a, the presenting unit 13a presents to the designer the conformation and stability (docking score) of the complex in addition to the conformation and characteristic value of the first compound based on the conformation information and characteristic value information of the first compound and the conformation information and stability information of the complex. This allows the designer to understand the predicted conformation and characteristic value of the first compound as well as the conformation and stability of the complex.

[0111] 2-3. Neural Network Training Method The neural network training method according to the second embodiment will be described with reference to Fig. 12 and Fig. 13. Fig. 12 and Fig. 13 are diagrams for explaining the interaction prediction AI according to the second embodiment. The interaction prediction AI includes, for example, a conformation prediction unit 23 and a stability prediction unit 24 (see Fig. 10). The learning flow will be described in order.

[0112] (Interaction Prediction AI) Learning of the interaction prediction AI is divided into two phases (Phase 1 and Phase 2). Phase 1 will be described with reference to Fig. 12, and Phase 2 will be described with reference to Fig. 13. Each prediction model in Fig. 12 and Fig. 13 is a machine learning model.

[0113] 12 , the conformation prediction model 23a first collects, from a complex information DB 50 related to a compound complex, complex structure information D1, which is three-dimensional structure information of the complex, and sequence information (first compound sequence information D2a and second compound sequence information D2b) of each of the first and second compounds constituting the complex, three-dimensional structure information (first compound structure information D3a and second compound structure information D3b), and active site information (first compound active site information D4a and second compound active site information D4b), and performs learning. The complex information DB 50 is configured to be able to communicate information with the information processing device 10A via the network 40.

[0114] 12 , the conformation prediction model 23 a predicts the conformation of the complex based on the first compound sequence information D2 a, the second compound sequence information D2 b, the first compound structure information D3 a, the second compound structure information D3 b, the first compound active site information D4 a, and the second compound active site information D4 b, and outputs conformation information D5 regarding the predicted conformation. This conformation prediction model 23 a can also predict the conformation of the complex from at least the first compound structure information D3 a, the second compound structure information D3 b, the first compound active site information D4 a, and the second compound active site information D4 b.

[0115] For example, the conformation prediction model 23a performs learning so as to minimize the error between the complex structure information D1 and the conformation information D5 in the complex information DB 50 (error minimization 23b). After the learning converges, the next learning phase, Phase 2, is performed. The convergence of the learning is determined, for example, by the designer and is determined in response to the input operation of the designer via the input unit 13b.

[0116] Next, in Phase 2 shown in Figure 13, the stability prediction model 24a learns to predict the stability of the predicted complex conformation. The stability prediction model 24a learns to predict a docking score D7 from the conformation information D5 input from the conformation prediction model 23a, the first compound structure information D3a, the second compound structure information D3b, and the binding affinity information D6. In the example of Figure 13, the stability prediction model 24a predicts a docking score D7 as the stability of the complex from the conformation information D5, the first compound structure information D3a, and the second compound structure information D3b, and outputs the predicted docking score D7.

[0117] For example, the stability prediction model 24a evaluates the error between the binding affinity information D6 and the docking score D7 output from the stability prediction model 24a (error minimization 23c), and updates the parameters of the stability prediction model 24a and the conformation prediction model 23a in an end-to-end manner using difference backpropagation. In other words, the conformation prediction model 23a and the stability prediction model 24a are generated by end-to-end machine learning. This allows the stability prediction model 24a to be adjusted to further improve the prediction accuracy of the final docking score D7. Examples of the binding affinity information D6 that can be used include binding affinity such as IC50 (50% inhibitory concentration), dissociation constant, and binding energy.

[0118] Here, for example, the conformation prediction unit 23 shown in FIG. 10 includes the trained conformation prediction model 23 a described above, and using this conformation prediction model 23 a, it is possible to predict the conformation of a complex of a first compound and a second compound from the three-dimensional structure information and active site information of the first compound and the three-dimensional structure information and active site of the second compound.

[0119] Furthermore, for example, the stability prediction unit 24 shown in FIG. 10 includes the trained stability prediction model 24a described above, and is capable of predicting the stability of a complex (docking score D7) from the conformational information D5 of the complex using the stability prediction model 24a.

[0120] By using the information processing system 1 according to the second embodiment, i.e., the compound design system, it is possible to obtain the same effects as those of the first embodiment. For example, a user (designer) can visually and interactively design a desired compound while taking into consideration the conformation information D5 and stability information of a complex between a desired compound and a compound for conjugation.

[0121] <3. Third embodiment> <3-1. Configuration example of information processing device> An example of the configuration of an information processing device 10B according to the third embodiment will be described with reference to Fig. 14. Fig. 14 is a diagram showing an example of the configuration of the information processing device 10B according to the third embodiment.

[0122] As shown in FIG. 14, the information processing device 10B according to the third embodiment is similar to the information processing device 10A according to the second embodiment, but updates the neural networks (e.g., structure prediction model 11a, characteristic value prediction model 12a, and sequence prediction model 14a) of each block (e.g., structure prediction unit 11, characteristic value prediction unit 12, sequence prediction unit 14, and structure prediction unit 21).

[0123] In the third embodiment, the learning parameters of the neural network of each block are updated using data held by the designer. For example, when the designer obtains new information on the sequence and three-dimensional structure of a compound, the designer uses the pair of information to additionally learn one or both of the blocks of the structure prediction unit 11 and the structure prediction unit 21 (e.g., the structure prediction model 11a). In this case, the block to be updated may be selected depending on the additional information available.

[0124] Furthermore, in order to notify the designer whether each block requires additional learning, the presenting unit 13a may present the prediction reliability of each block to the designer. In the example of Fig. 14, the prediction reliability is shown as "Reliability: XX". The calculation of this prediction reliability will be described later.

[0125] 3-2. Neural Network Training Method> A neural network training method according to the third embodiment will be described with reference to Fig. 15. Fig. 15 is a diagram for explaining calculation of prediction reliability according to the third embodiment.

[0126] 15 , the prediction unit 100 according to the third embodiment also predicts the prediction reliability, which is the reliability of the prediction itself. The prediction unit 100 corresponds to, for example, any of the structure prediction unit 11, the characteristic value prediction unit 12, the sequence prediction unit 14, and the structure prediction unit 21. Note that the prediction unit 100 may also correspond to, for example, any of the active site prediction unit 22, the conformation prediction unit 23, and the stability prediction unit 24.

[0127] 15, it is assumed that the neural network (e.g., machine learning model) of the prediction unit 100 has multiple outputs. The neural network corresponds to, for example, one of the structure prediction model 11a, the property value prediction model 12a, and the sequence prediction model 14a, or one of the conformation prediction model 23a and the stability prediction model 24a.

[0128] When information is input to the neural network of the prediction unit 100, a prediction result is output. That is, output information E2 is generated in response to input information E1. The learning parameters of the neural network are updated so as to minimize the error between the output information E2 of the prediction result and the correct answer information E3. At this time, the error itself is regressed by another output (the output of prediction error D4).

[0129] After that, the neural network, whose learning has converged, is able to make predictions for unknown input information and simultaneously output the expected prediction error E4. This prediction error E4 is used as a prediction reliability, which is an index of the reliability of the prediction.

[0130] According to this configuration, by presenting the prediction reliability to the designer using the presentation unit 13 a, etc., it is possible to encourage the designer to add additional data. Also, in the first and second embodiments, the designer can grasp the validity of the output of the prediction unit 100.

[0131] By using the information processing system 1 according to the third embodiment, i.e., the compound design system, it is possible to obtain the same effects as those of the first and second embodiments. Furthermore, it is possible to realize additional learning for any or all of the blocks. Furthermore, by presenting the prediction reliability of each block, it is possible to inform the designer whether or not each block requires additional learning.

[0132] 4. Actions and Effects As described above, any of the information processing devices 10, 10A, and 10B according to the above-described embodiments includes a structure prediction unit 11 that predicts the three-dimensional structure of a candidate compound based on sequence information of the candidate compound and obtains three-dimensional structure information of the candidate compound, a characteristic value prediction unit 12 that predicts characteristic values ​​of the candidate compound based on the sequence information and three-dimensional structure information of the candidate compound and obtains characteristic value information of the candidate compound, and a presentation unit 13 a that presents the characteristic value information and three-dimensional structure information of the candidate compound to a user. This allows users such as designers to easily grasp the characteristic value information and three-dimensional structure information of the candidate compound, thereby realizing high throughput in the cycle of designing and evaluating desired compounds and enabling efficient production of desired compounds.

[0133] Furthermore, any of the information processing devices 10, 10A, and 10B according to the above-described embodiments may further include an input unit 13 b that accepts a user's input operation, and a processing unit 13 c that determines a candidate compound as a next candidate compound in response to the user's input operation on the input unit 13 b. This allows the user to appropriately specify the next candidate, thereby realizing a high-throughput cycle of designing and evaluating desired compounds (design and evaluation cycle).

[0134] Furthermore, the processing unit 13c may change the three-dimensional structure information of the candidate compound in response to an input operation by the user via the input unit 13b, thereby enabling the user to appropriately change the three-dimensional structure information of the candidate compound, thereby achieving high throughput in the design evaluation cycle.

[0135] Furthermore, any of the information processing devices 10, 10A, and 10B according to the above-described embodiments may further include a sequence prediction unit 14 that predicts the sequence of a next candidate compound based on three-dimensional structural information of the next candidate compound and obtains the sequence information of the next candidate compound, the structure prediction unit 11 that predicts the three-dimensional structure of the next candidate compound based on the sequence information of the next candidate compound and obtains the three-dimensional structural information of the next candidate compound, the property value prediction unit 12 that predicts the property values ​​of the next candidate compound based on the sequence information and three-dimensional structural information of the next candidate compound and obtains the property value information of the next candidate compound, and the presentation unit 13 a that presents the property value information and three-dimensional structural information of the next candidate compound to the user. This reliably achieves high throughput in the design evaluation cycle and enables efficient generation of desired compounds.

[0136] Alternatively, the structure prediction unit 11 may predict the three-dimensional structure information of the candidate compound using a structure prediction model 11a generated by machine learning based on the sequence information and three-dimensional structure information of each type of compound, and the sequence prediction unit 14 may predict the sequence information of the next candidate compound using a sequence prediction model 14a generated by machine learning based on the sequence information and three-dimensional structure information of each type of compound. This makes it possible to obtain the three-dimensional structure information of the candidate compound and the sequence information of the next candidate compound with high accuracy.

[0137] The structure prediction model 11a and the sequence prediction model 14a may be generated by end-to-end machine learning, which allows the consistency between the edited three-dimensional structure information and the sequence information of the next candidate compound to be maintained even when the three-dimensional structure information of the candidate compound is edited by the user.

[0138] Furthermore, the sequence prediction unit 14 may calculate the prediction reliability of the predicted sequence information of the next candidate compound, and the presentation unit 13a may present the prediction reliability to the user. This allows the user to understand the prediction reliability of the sequence prediction unit 14. For example, if the prediction reliability is lower than a desired value, the user can operate the input unit 13b to instruct the sequence prediction unit 14 to re-learn the sequence prediction model 14a.

[0139] The structure prediction unit 11 may also predict the three-dimensional structure of a candidate compound using a structure prediction model 11 a generated by machine learning based on sequence information and three-dimensional structure information for each type of compound, thereby enabling the three-dimensional structure information of the candidate compound to be obtained with high accuracy.

[0140] Furthermore, the structure prediction unit 11 may calculate the prediction reliability of the predicted three-dimensional structure information of the candidate compound, and the presentation unit 13a may present the prediction reliability to the user. This allows the user to understand the prediction reliability of the structure prediction unit 11. For example, if the prediction reliability is lower than a desired value, the user can operate the input unit 13b to instruct the structure prediction unit 11 to re-learn the structure prediction model 11a.

[0141] Furthermore, the property value prediction unit 12 may predict the property values ​​of the candidate compounds using a property value prediction model 12a generated by machine learning based on the sequence information, three-dimensional structure information, and property value information (e.g., various property values ​​C7) for each type of compound, thereby enabling the property value information of the candidate compounds to be obtained with high accuracy.

[0142] Furthermore, the characteristic value prediction unit 12 predicts experimental reliability C3 for an experiment to find a candidate compound using an experimental reliability prediction model 15 generated by machine learning based on characteristic value information C1 and experimental condition information C2 for each type of compound, and the characteristic value prediction model 12a may be generated by machine learning based on the experimental reliability C3 in addition to sequence information, three-dimensional structure information, and characteristic value information for each type of compound. This makes it possible to obtain characteristic value information of the candidate compound with high accuracy.

[0143] The experimental reliability prediction model 15 may be generated by machine learning based on the characteristic value information and experimental condition information C2 for each compound, as well as the reliability score C4 regarding the experimental reliability for each compound evaluated by a user, thereby enabling the experimental reliability C3 to be obtained with high accuracy.

[0144] Furthermore, the three-dimensional structure information for each type of compound may include three-dimensional structure information of candidate compounds predicted by the structure prediction unit 11. This makes it possible to improve the prediction accuracy of the characteristic value prediction model 12a by using information on compounds whose three-dimensional structures are unknown.

[0145] Furthermore, the property value prediction unit 12 may calculate the prediction reliability of the predicted property value of the candidate compound, and the presentation unit 13a may present the prediction reliability to the user. This allows the user to understand the prediction reliability of the property value prediction unit 12. For example, if the prediction reliability is lower than a desired value, the user can operate the input unit 13b to instruct the property value prediction unit 12 to re-learn the property value prediction model 12a.

[0146] Furthermore, either of the information processing devices 10A and 10B according to the above-described embodiments may further include a structure prediction unit 21 that predicts the three-dimensional structure of the compound for conjugation based on sequence information of the compound for conjugation with respect to the candidate compound, and obtains three-dimensional structure information of the compound for conjugation; an active site prediction unit 22 that predicts the active site of the compound for conjugation based on the sequence information and three-dimensional structure information of the compound for conjugation, and obtains active site information of the compound for conjugation; a conformation prediction unit 23 that predicts the conformation of a complex between the candidate compound and the compound for conjugation based on the active site information and three-dimensional structure information of the compound for conjugation and the three-dimensional structure information and characteristic value information of the candidate compound, and obtains conformation information D5 of the complex; and a stability prediction unit 24 that predicts the stability of the complex based on the conformation information D5 of the complex, and obtains stability information of the complex, and the presentation unit 13a may present the stability information of the complex to the user, in addition to the three-dimensional structure information and characteristic value information of the candidate compound. This allows users such as designers to easily grasp the property value information and three-dimensional structure information of candidate compounds, as well as the stability information of the complex, thereby realizing a high-throughput design and evaluation cycle and enabling the efficient generation of desired compounds.

[0147] Furthermore, the conformation prediction unit 23 may predict the conformation of the complex using a conformation prediction model 23a generated by machine learning based on the sequence information, three-dimensional structure information, and active site information of each of the multiple compounds for each type of complex, and the three-dimensional structure information of each of the multiple complexes, and the stability prediction unit 24 may predict the stability of the complex using a stability prediction model 24a generated by machine learning based on the three-dimensional structure information and three-dimensional structure information D5 of each of the multiple compounds for each type of complex. This makes it possible to accurately obtain the conformation information D5 and stability information of the complex.

[0148] The conformation prediction model 23a and the stability prediction model 24a may be generated by end-to-end machine learning, thereby improving the prediction accuracy of the stability prediction model 24a.

[0149] Furthermore, the conformation prediction unit 23 or the stability prediction unit 24 may obtain a prediction reliability for predicting the conformation or stability of the complex, and the presentation unit 13a may present the prediction reliability to the user. This allows the user to understand the prediction reliability of the conformation prediction unit 23 or the stability prediction unit 24. For example, if the prediction reliability is lower than a desired value, the user can operate the input unit 13b to instruct re-learning of the conformation prediction model 23a of the conformation prediction unit 23 or the stability prediction model 24a of the stability prediction unit 24.

[0150] 5. Other Embodiments The processes according to the above-described embodiments (or modifications) may be implemented in various different forms (modifications) other than the above-described embodiments. For example, among the processes described in the above embodiments, all or part of the processes described as being performed automatically can be performed manually, or all or part of the processes described as being performed manually can be performed automatically using a known method. In addition, the process procedures, specific names, and information including various data and parameters shown in the above documents and drawings can be changed as desired unless otherwise specified. For example, the various information shown in each drawing is not limited to the information shown in the drawings.

[0151] Furthermore, the components of each device shown in the figure are conceptual functional components and do not necessarily have to be physically configured as shown in the figure. In other words, the specific form of distribution and integration of each device is not limited to that shown in the figure, and all or part of them can be functionally or physically distributed and integrated in any unit depending on various loads, usage conditions, etc.

[0152] Furthermore, the above-described embodiments (or variations) can be combined as appropriate within the scope of not causing any contradiction in the processing content. Furthermore, the effects described in this specification are merely examples and are not intended to be limiting, and other effects may also be obtained.

[0153] Furthermore, in the above-described embodiment (or modified example), a system refers to a collection of multiple components (devices, modules (components), etc.), regardless of whether all of the components are contained in the same housing. Therefore, multiple devices housed in separate housings and connected via a network 40, and a single device housed in a single housing and containing multiple modules, are both systems.

[0154] Furthermore, in the above-described embodiment (or modified example), a cloud computing configuration can be adopted in which one function is shared and processed jointly by multiple devices via the network 40. Furthermore, each step described in the above-described process flow (e.g., flowchart) can be executed by one device or shared and executed by multiple devices. Furthermore, when one step includes multiple processes, the multiple processes included in that one step can be executed by one device or shared and executed by multiple devices.

[0155] 6. Example of Hardware Configuration The information processing devices 10, 10A, and 10B according to the first to third embodiments described above may be realized by, for example, a computer 500 having a configuration as shown in Fig. 16. Fig. 16 is a diagram showing an example of the hardware configuration of the computer 500.

[0156] As shown in FIG. 16, a computer 500 includes a CPU (Central Processing Unit) 510, a ROM (Read Only Memory) 520, and a RAM (Random Access Memory) 530.

[0157] The CPU 510, ROM 520, and RAM 530 are interconnected by a bus 540. An input / output interface 550 is further connected to this bus 540. An input unit 560, an output unit 570, a recording unit 580, a communication unit 590, and a drive 600 are connected to this input / output interface 550.

[0158] The input unit 560 is composed of, for example, a keyboard, a mouse, a microphone, an image sensor, etc. The output unit 570 is composed of, for example, a display, a speaker, etc. The recording unit 580 is composed of, for example, a hard disk, a non-volatile memory, etc. The communication unit 590 is composed of, for example, a network interface, etc. The drive 600 drives a removable recording medium 610 such as, for example, a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory.

[0159] In the computer 500 configured as described above, the CPU 510 performs the above-described series of processes by, for example, loading a program recorded in the recording unit 580 into the RAM 530 via the input / output interface 550 and the bus 540 and executing the program.

[0160] In other words, when the computer 500 functions as any one of the information processing devices 10, 10A, and 10B according to the first to third embodiments, the CPU 510 of the computer 500 executes a program loaded into the RAM 530 to realize the functions of each part of the information processing device 10, 10A, and 10B according to the first to third embodiments.

[0161] The program executed by CPU 510 may be provided, for example, by being recorded on a removable recording medium 610 such as a package medium, or may be provided via a wired or wireless transmission medium such as a local area network, the Internet, or digital satellite broadcasting.

[0162] That is, the program may be installed in recording unit 580 via input / output interface 550 by attaching removable recording medium 610 to drive 600, or may be received by communication unit 590 via a wired or wireless transmission medium and installed in recording unit 580. Alternatively, the program may be installed in ROM 520 or recording unit 580 in advance.

[0163] The program may be a program that processes in chronological order according to the order described in this specification, or may be a program that processes in parallel or at the required timing, such as when called.

[0164] <7. Supplementary Notes> The present technology can also be configured as follows. (1) An information processing device comprising: a structure prediction unit that predicts a three-dimensional structure of a candidate compound based on sequence information of the candidate compound and obtains three-dimensional structure information of the candidate compound; a characteristic value prediction unit that predicts a characteristic value of the candidate compound based on the sequence information and three-dimensional structure information of the candidate compound and obtains characteristic value information of the candidate compound; and a presentation unit that presents the characteristic value information and three-dimensional structure information of the candidate compound to a user. (2) The information processing device described in (1), further comprising: an input unit that accepts an input operation from the user; and a processing unit that determines the candidate compound as a next candidate compound in accordance with the user's input operation on the input unit. (3) The information processing device described in (2), wherein the processing unit changes the three-dimensional structure information of the candidate compound in accordance with the user's input operation on the input unit. (4) The information processing device according to (2) or (3), further comprising a sequence prediction unit that predicts a sequence of the next candidate compound based on three-dimensional structure information of the next candidate compound and obtains the sequence information of the next candidate compound, wherein the structure prediction unit predicts a three-dimensional structure of the next candidate compound based on the sequence information of the next candidate compound and obtains the three-dimensional structure information of the next candidate compound, the characteristic value prediction unit predicts a characteristic value of the next candidate compound based on the sequence information and three-dimensional structure information of the next candidate compound and obtains the characteristic value information of the next candidate compound, and the presentation unit presents the characteristic value information and three-dimensional structure information of the next candidate compound to the user. (5) The information processing device according to (4), wherein the structure prediction unit predicts three-dimensional structure information of the candidate compound using a structure prediction model generated by machine learning based on the sequence information and three-dimensional structure information of each of the various compounds. (6) The information processing device according to (5), wherein the structure prediction model and the sequence prediction model are generated by end-to-end machine learning.(7) The information processing device according to any one of (4) to (6), wherein the sequence prediction unit calculates a prediction reliability of predicted sequence information of the next candidate compound, and the presentation unit presents the prediction reliability to the user. (8) The information processing device according to any one of (1) to (7), wherein the structure prediction unit predicts a three-dimensional structure of the candidate compound using a structure prediction model generated by machine learning based on sequence information and three-dimensional structure information for each type of compound. (9) The information processing device according to any one of (1) to (8), wherein the structure prediction unit calculates a prediction reliability of predicted three-dimensional structure information of the candidate compound, and the presentation unit presents the prediction reliability to the user. (10) The information processing device according to any one of (1) to (9), wherein the property value prediction unit predicts a property value of the candidate compound using a property value prediction model generated by machine learning based on sequence information, three-dimensional structure information, and property value information for each type of compound. (11) The information processing device according to (10), wherein the property value prediction unit predicts experimental reliability for an experiment to find the candidate compound using an experimental reliability prediction model generated by machine learning based on property value information and experimental condition information for each type of compound, and the property value prediction model is generated by machine learning based on the experimental reliability in addition to sequence information, three-dimensional structure information, and property value information for each type of compound. (12) The information processing device according to (11), wherein the experimental reliability prediction model is generated by machine learning based on property value information and experimental condition information for each type of compound, and a reliability score for the experimental reliability evaluated by a user for each type of compound. (13) The information processing device according to any one of (10) to (12), wherein the three-dimensional structure information for each type of compound includes three-dimensional structure information of the candidate compound predicted by the structure prediction unit. (14) The information processing device according to any one of (1) to (13), wherein the characteristic value prediction unit calculates a prediction reliability of the predicted characteristic value of the candidate compound, and the presentation unit presents the prediction reliability to the user.(15) The information processing device according to any one of (1) to (14), further comprising: a structure prediction unit that predicts a three-dimensional structure of the compound for combination based on sequence information of the compound for combination with the candidate compound, and obtains three-dimensional structure information of the compound for combination; an active site prediction unit that predicts an active site of the compound for combination based on the sequence information and three-dimensional structure information of the compound for combination, and obtains active site information of the compound for combination; a conformation prediction unit that predicts a conformation of a complex of the candidate compound and the compound for combination based on the active site information and three-dimensional structure information of the compound for combination and three-dimensional structure information and characteristic value information of the candidate compound, and obtains conformation information of the complex; and a stability prediction unit that predicts stability of the complex based on the conformation information of the complex, and obtains stability information of the complex, wherein the presentation unit presents to the user the stability information of the complex in addition to the three-dimensional structure information and characteristic value information of the candidate compound. (16) The information processing device according to any one of (15) to (17), wherein the conformation prediction unit predicts the conformation of the complex using a conformation prediction model generated by machine learning based on sequence information, three-dimensional structure information, and active site information of each of a plurality of compounds for each type of complex and the three-dimensional structure information of each of the various complexes, and the stability prediction unit predicts the stability of the complex using a stability prediction model generated by machine learning based on the three-dimensional structure information and three-dimensional conformation information of each of the multiple compounds for each type of complex. (17) The information processing device according to (16), wherein the conformation prediction model and the stability prediction model are generated by end-to-end machine learning. (18) The information processing device according to any one of (15) to (17), wherein the conformation prediction unit or the stability prediction unit calculates a prediction reliability for predicting the conformation or stability of the complex, and the presentation unit presents the prediction reliability to the user.(19) An information processing system comprising: a structure prediction unit that predicts a three-dimensional structure of a candidate compound based on sequence information of the candidate compound and obtains three-dimensional structure information of the candidate compound; a characteristic value prediction unit that predicts a characteristic value of the candidate compound based on the sequence information and three-dimensional structure information of the candidate compound and obtains characteristic value information of the candidate compound; and a presentation unit that presents the characteristic value information and three-dimensional structure information of the candidate compound to a user. (20) An information processing method in which an information processing device performs the following: predicts a three-dimensional structure of the candidate compound based on sequence information of the candidate compound and obtains three-dimensional structure information of the candidate compound; predicts a characteristic value of the candidate compound based on the sequence information and three-dimensional structure information of the candidate compound and obtains characteristic value information of the candidate compound; and presents the characteristic value information and three-dimensional structure information of the candidate compound to a user. (21) An information processing device comprising any or all of the components included in the information processing device described in any one of (1) to (18). (22) An information processing method using any or all of the components included in the information processing device according to any one of (1) to (18).

[0165] 1 Information processing system 10 Information processing device 10A Information processing device 10B Information processing device 11 Structure prediction unit 11a Structure prediction model 11b Error minimization 12 Property value prediction unit 12a Property value prediction model 12b Error minimization 13 Selection unit 13a Presentation unit 13b Input unit 131 Keyboard 132 Mouse 13c Processing unit 14 Sequence prediction unit 14a Sequence prediction model 14b Error minimization 14c Error minimization 15 Experimental reliability prediction model 15a Error minimization 16 Evaluation unit 16a Presentation unit 16b Input unit 16c Processing unit 21 Structure prediction unit 22 Active site prediction unit 23 Conformation prediction unit 23a Conformation prediction model 23b Error minimization 23c Error minimization 24 Stability prediction unit 24a Stability prediction model 20 Sequence / structure information DB 30 Property information DB 40 Network 50 Complex information DB 100 Prediction section A1a First compound sequence information A1b Second compound sequence information A2 Three-dimensional structure image A3 Table B1 Compound sequence information B2 Compound structure information B3 Predicted structure information B4 Predicted sequence information C1 Property value information C2 Experimental condition information C3 Experimental reliability C4 Reliability score C5 Compound sequence information C6 Compound structure information C7 Various property values ​​C8 Property value information C8a Active site C8b Denaturation temperature C8c Solubility C8d pH characteristics C8e Biosynthesis cost D1 Complex structure information D2a First compound sequence information D2b Second compound sequence information D3a First compound structure information D3b Second compound structure information D4a D4b Information on the active site of the first compound D5 Conformational information D6 Binding affinity information D7 Docking score E1 Input information E2 Output information E3 Correct answer information E4 Prediction error

Claims

1. An information processing device comprising: a structure prediction unit that predicts a three-dimensional structure of a candidate compound based on sequence information of the candidate compound and obtains three-dimensional structure information of the candidate compound; a characteristic value prediction unit that predicts characteristic values ​​of the candidate compound based on the sequence information and three-dimensional structure information of the candidate compound and obtains characteristic value information of the candidate compound; and a presentation unit that presents the characteristic value information and three-dimensional structure information of the candidate compound to a user.

2. The information processing device according to claim 1, further comprising: an input unit that accepts an input operation by the user; and a processing unit that determines the candidate compound as a next candidate compound in response to the user's input operation on the input unit.

3. The information processing device according to claim 2, wherein the processing section changes the three-dimensional structure information of the candidate compound in response to an input operation by the user on the input section.

4. An information processing device as described in claim 2, further comprising a sequence prediction unit that predicts the sequence of the next candidate compound based on three-dimensional structure information of the next candidate compound and obtains sequence information of the next candidate compound, wherein the structure prediction unit predicts the three-dimensional structure of the next candidate compound based on the sequence information of the next candidate compound and obtains three-dimensional structure information of the next candidate compound, wherein the characteristic value prediction unit predicts the characteristic values ​​of the next candidate compound based on the sequence information and three-dimensional structure information of the next candidate compound and obtains characteristic value information of the next candidate compound, and wherein the presentation unit presents the characteristic value information and three-dimensional structure information of the next candidate compound to the user.

5. The information processing device according to claim 4, wherein the structure prediction unit predicts three-dimensional structure information of the candidate compound using a structure prediction model generated by machine learning based on sequence information and three-dimensional structure information for each type of compound, and the sequence prediction unit predicts sequence information of the next candidate compound using a sequence prediction model generated by machine learning based on the sequence information and three-dimensional structure information for each type of compound.

6. The information processing device according to claim 5, wherein the structure prediction model and the sequence prediction model are generated by end-to-end machine learning.

7. The information processing device according to claim 4, wherein the sequence prediction unit determines a prediction reliability of the predicted sequence information of the next candidate compound, and the presentation unit presents the prediction reliability to the user.

8. The information processing device according to claim 1, wherein the structure prediction unit predicts the three-dimensional structure of the candidate compound using a structure prediction model generated by machine learning based on sequence information and three-dimensional structure information for each type of compound.

9. The information processing device according to claim 1, wherein the structure prediction unit determines a prediction reliability of the predicted three-dimensional structure information of the candidate compound, and the presentation unit presents the prediction reliability to the user.

10. The information processing device according to claim 1, wherein the characteristic value prediction unit predicts the characteristic values ​​of the candidate compounds using a characteristic value prediction model generated by machine learning based on sequence information, three-dimensional structure information and characteristic value information for each type of compound.

11. The information processing device described in claim 10, wherein the characteristic value prediction unit predicts the experimental reliability for an experiment to find the candidate compound using an experimental reliability prediction model generated by machine learning based on characteristic value information and experimental condition information for each type of compound, and the characteristic value prediction model is generated by machine learning based on the experimental reliability in addition to sequence information, three-dimensional structure information, and characteristic value information for each type of compound.

12. The information processing device according to claim 11, wherein the experimental reliability prediction model is generated by machine learning based on the characteristic value information and experimental condition information for each of the various compounds, as well as a reliability score regarding the experimental reliability for each of the various compounds evaluated by a user.

13. The information processing device according to claim 10, wherein the three-dimensional structure information for each of the various compounds includes three-dimensional structure information of the candidate compounds predicted by the structure prediction unit.

14. The information processing device according to claim 1, wherein the characteristic value prediction unit determines a prediction reliability for predicting the characteristic value of the candidate compound, and the presentation unit presents the prediction reliability to the user.

15. The information processing device according to claim 1, further comprising: a structure prediction unit that predicts a three-dimensional structure of the compound for combination based on sequence information of the compound for combination for the candidate compound, and obtains three-dimensional structure information of the compound for combination; an active site prediction unit that predicts an active site of the compound for combination based on the sequence information and three-dimensional structure information of the compound for combination, and obtains active site information of the compound for combination; a conformation prediction unit that predicts a conformation of a complex of the candidate compound and the compound for combination based on the active site information and three-dimensional structure information of the compound for combination, and three-dimensional structure information and characteristic value information of the candidate compound, and obtains conformation information of the complex; and a stability prediction unit that predicts the stability of the complex based on the conformation information of the complex, and obtains stability information of the complex, wherein the presentation unit presents the stability information of the complex to the user in addition to the three-dimensional structure information and characteristic value information of the candidate compound.

16. The information processing device described in claim 15, wherein the conformation prediction unit predicts the conformation of the complex using a conformation prediction model generated by machine learning based on sequence information, three-dimensional structure information and active site information of each of a plurality of compounds for each type of complex and the three-dimensional structure information of each of the various complexes, and the stability prediction unit predicts the stability of the complex using a stability prediction model generated by machine learning based on the three-dimensional structure information and three-dimensional conformation information of each of a plurality of compounds for each type of complex.

17. The information processing device according to claim 16, wherein the conformation prediction model and the stability prediction model are generated by end-to-end machine learning.

18. The information processing device according to claim 15, wherein the conformation prediction unit or the stability prediction unit determines a prediction reliability for predicting the conformation or stability of the complex, and the presentation unit presents the prediction reliability to the user.

19. An information processing system comprising: a structure prediction unit that predicts a three-dimensional structure of a candidate compound based on sequence information of the candidate compound and obtains three-dimensional structure information of the candidate compound; a characteristic value prediction unit that predicts characteristic values ​​of the candidate compound based on the sequence information and three-dimensional structure information of the candidate compound and obtains characteristic value information of the candidate compound; and a presentation unit that presents the characteristic value information and three-dimensional structure information of the candidate compound to a user.

20. An information processing method in which an information processing device performs the following: predicting a three-dimensional structure of a candidate compound based on sequence information of the candidate compound, and obtaining three-dimensional structure information of the candidate compound; predicting characteristic values ​​of the candidate compound based on the sequence information and three-dimensional structure information of the candidate compound, and obtaining characteristic value information of the candidate compound; and presenting the characteristic value information and three-dimensional structure information of the candidate compound to a user.

Images