Information processing program, information processing method, and information processing device

Abstract

An information processing device (100) comprises a model (110) that, in response to input data representing the structure of a molecule of interest, generates output data representing energy relating to the molecule of interest and force relating to the molecule of interest. The information processing device (100) acquires, for individual pieces of structure data (101), a plurality of first feature amounts relating to the structure of the molecule of interest in accordance with the result of inputting the structure data (101) to the model (110). The information processing device (100) calculates, for each piece of structure data (101), a second feature amount relating to the distance between atoms in the structure of the molecule of interest represented by the structure data (101). The information processing device (100) selects, from among the plurality of pieces of structure data (101), structure data (101) to be used when updating the model (110), the structure data (101) being selected on the basis of the distribution of coordinates in a multidimensional space (120) representing combinations of the plurality of acquired first feature amounts and the calculated second feature amounts.

Description

Information processing program, information processing method, and information processing device

[0001] The present invention relates to an information processing program, an information processing method, and an information processing device.

[0002] Conventionally, there is a molecular dynamics simulation technique that uses an equation of motion to simulate the time evolution of the motion of each atom in an atomic system including N atoms that form a molecule of interest, based on the potential of the molecule of interest. Here, there is a technique for training a model that represents the potential used in the molecular dynamics simulation. The model is a neural network. For example, the model estimates the energy and force of the molecule of interest from the structure of the molecule of interest.

[0003] Prior art techniques include, for example, a technique that uses structural description information corresponding to a selected crystal structure candidate as input data, performs machine learning using energy information related to the selected crystal structure candidate as training data, and realizes a crystal form prediction device.

[0004] International Publication No. 2020 / 203922

[0005] However, in conventional techniques, it can be difficult to efficiently train a model that represents potential. For example, if appropriate training data is not prepared, the accuracy of the trained model in estimating the energy and force of a target molecule from the structure of the target molecule can be reduced.

[0006] In one aspect, the present invention aims to make models easier to train.

[0007] According to one embodiment, an information processing program, an information processing method, and an information processing device are proposed that, for each of a plurality of structure data each representing the structure of a target molecule, acquire a plurality of first feature quantities related to the structure of the target molecule in accordance with the result of inputting the structure data into a model that generates output data representing energy related to the target molecule in accordance with the input data representing the structure of the target molecule, calculate a second feature quantity related to the distance between atoms in the structure of the target molecule represented by the structure data for each of the plurality of structure data, and select at least one of the plurality of structure data to be used when updating the model based on the distribution of the coordinates, by referring to coordinates in a multidimensional space that represent a combination of the acquired plurality of first feature quantities and the calculated second feature quantities corresponding to each of the plurality of structure data.

[0008] According to one aspect, the model can be made easier to train.

[0009] FIG. 1 is an explanatory diagram showing an example of an information processing method according to an embodiment. FIG. 2 is an explanatory diagram showing an example of an information processing system 200. FIG. 3 is a block diagram showing an example of the hardware configuration of an information processing device 100. FIG. 4 is a block diagram showing an example of the functional configuration of the information processing device 100. FIG. 5 is an explanatory diagram showing an example of an operation example of the information processing device 100. FIG. 6 is an explanatory diagram showing an example of MD calculation. FIG. 7 is an explanatory diagram showing an example of selecting a structure data group 502. FIG. 8 is an explanatory diagram showing an example of an effect. FIG. 9 is a flowchart showing an example of an overall processing procedure.

[0010] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, an information processing program, an information processing method, and an information processing device according to embodiments of the present invention will be described in detail with reference to the accompanying drawings.

[0011] (Example of Information Processing Method According to Embodiment) FIG. 1 is an explanatory diagram showing an example of an information processing method according to an embodiment. The information processing device 100 is a computer that facilitates training a model that represents potential. The information processing device 100 is, for example, a server or a PC (Personal Computer). The model is, for example, a neural network. In the following description, the neural network may be referred to as "NN (Neural Network)."

[0012] Conventionally, molecular dynamics simulations are sometimes performed that use equations of motion to simulate the time evolution of the motion of each atom in an atomic system containing N atoms that form a molecule of interest, based on the potential of the molecule of interest. In the following description, molecular dynamics simulations may be referred to as "MD (Molecular Dynamics) calculations."

[0013] The potential is a parameter that forms an equation of motion. The potential represents, for example, interactions between atoms. Specifically, the potential realizes a function of estimating the energy and force related to a target molecule from the structure of the target molecule. The potential is set, for example, by a user. A potential set by a user is also called an empirical potential. On the other hand, there are cases where the potential is represented using a model. The model is, for example, a neural network. A model that represents the potential is also called, for example, a machine learning potential.

[0014] Here, a model representing the potential may be trained. For example, the model may be trained based on training data that has the structure of the target molecule as an input sample and the energy and force related to the target molecule as labels indicating the correct output. Training is performed, for example, by backpropagation.

[0015] Specifically, when training a model, density functional theory calculations are performed on structural data representing the structure of a target molecule, and training data is prepared by identifying the energy and force associated with the target molecule and assigning labels. The labels indicate the identified energy and force associated with the target molecule. In the following description, density functional theory may be referred to as "DFT (Density Functional Theory)."

[0016] By preparing training data by DFT calculation, it is expected that the accuracy of the model will approach the accuracy of DFT calculation. The accuracy of the model indicates how accurately the energy and force related to the target molecule can be estimated from the structure of the target molecule in the model.

[0017] However, if appropriate training data is not prepared, the accuracy of the trained model will decrease. Furthermore, the more training data is prepared in order to improve the accuracy of the model, the longer the time required to train the model will increase. Therefore, it is desirable to selectively prepare appropriate training data to reduce the time required to train the model while improving the accuracy of the model.

[0018] In contrast to this, a first method is conceivable, in which one or more pieces of structure data to be labeled are selected from the plurality of pieces of structure data based on two-dimensional feature quantities related to each piece of structure data, making it easier to prepare appropriate training data. The label indicates the correct output and indicates the energy and force related to the target molecule. Specifically, it is conceivable to arrange the plurality of pieces of structure data in a two-dimensional space using the two-dimensional feature quantities as coordinates, and uniformly select one or more pieces of structure data in the two-dimensional space. The two-dimensional feature quantities can be obtained, for example, by performing principal component analysis on the plurality of feature quantities in the intermediate layer of a neural network when the structure data is input to the neural network. This first method is also referred to as active learning, for example. For more information on active learning, see, for example, Reference 1 below.

[0019] Reference 1: Shimizu, Koji, et al. “Phase stability of Au-Li binary systems studied using neural network potential.” Physical Review B 103.9 (2021): 094112.

[0020] Even with this first method, it may be difficult to efficiently train a model representing the potential. For example, for a target molecular structure not represented by the training data, the model tends to have difficulty accurately estimating the energy and force associated with the target molecule. Here, in the first method, one or more structural data are uniformly selected in two-dimensional space without considering the distance between atoms, so the accuracy of the model may be low for a target molecular structure corresponding to a case where the distance between atoms is close. Furthermore, when performing MD calculations, the low accuracy of the model may cause the distance between atoms to become too close, causing atoms to fly out of the atomic system due to the repulsive force between atoms, resulting in the failure of the MD calculations.

[0021] Furthermore, in order to take into account cases where the distance between atoms is short, a second method can be considered, which is an extension of the first method, in which a plurality of pieces of structure data in which the size of the atomic system is reduced are prepared, and one or more pieces of structure data to which a label is to be assigned are selected from the plurality of pieces of structure data. This second method is also called, for example, a volume scan. For more information on volume scan, see, for example, Reference 2 below.

[0022] Reference 2: Magdau, Ioan-Bogdan, et al. “Machine learning force fields for molecular liquids: Ethylene Carbonate / Ethyl Methyl Carbonate binary solvent.” npj Computational Materials 9.1 (2023): 146.

[0023] Even with this second method, it may be difficult to efficiently train a model representing the potential. As with the first method, the second method uniformly selects one or more structural data in two-dimensional space without considering the distance between atoms, which may result in low model accuracy for the target molecular structure when the distance between atoms is short. Furthermore, when performing MD calculations, the low model accuracy may cause the distance between atoms to become too close, resulting in atoms jumping out of the atomic system due to the repulsive force between atoms, potentially resulting in the failure of the MD calculations.

[0024] Therefore, in this embodiment, an information processing method that can make it easier to train a model that represents potential will be described.

[0025] 1, an information processing device 100 has a model 110. The model 110 generates output data representing energy and forces related to a target molecule in response to input data representing the structure of the target molecule. The input data is, for example, structural data 101. The model 110 is, for example, a neural network. The model 110 is a target to be trained. Training corresponds to, for example, updating the model 110.

[0026] The information processing device 100 stores a plurality of pieces of structure data 101. Preferably, each of the plurality of pieces of structure data 101 represents, for example, the structure of a target molecule in a reduced system. For example, each of the plurality of pieces of structure data 101 may represent the structure of a target molecule in a non-reduced system. For example, the structure data 101 may further represent the size of a simulation box, the presence or absence of a periodic boundary condition, or the presence or absence of stress. The structure data 101 is generated, for example, by using a volume scan. The structure data 101 may also be generated by using a method other than a volume scan. The structure data 101 may also be generated by using a pressure setting for the system. The information processing device 100 may generate and store the plurality of pieces of structure data 101 by, for example, using a volume scan.

[0027] (1-1) The information processing device 100 acquires, for each piece of structure data 101, a plurality of first feature quantities related to the structure of a target molecule in accordance with the result of inputting the structure data 101 into the model 110. For example, the information processing device 100 extracts, for each piece of structure data 101, a plurality of feature quantities in the intermediate layer of the model 110 when the structure data 101 is input into the model 110. For example, the information processing device 100 acquires the plurality of first feature quantities by performing principal component analysis on the plurality of extracted feature quantities for each piece of structure data 101 and calculating the plurality of first feature quantities. The number of first feature quantities is, for example, two. This allows the information processing device 100 to acquire a guideline for selecting at least one piece of structure data 101 to be used when updating the model 110 from among the plurality of pieces of structure data 101.

[0028] (1-2) The information processing device 100 calculates, for each of the plurality of structure data 101, a second feature amount related to the interatomic distance in the structure of the target molecule represented by the structure data 101. The second feature amount is, for example, a statistical value of the interatomic distance in the structure of the target molecule represented by the structure data 101. The statistical value is, for example, a minimum value, a maximum value, an average value, a median value, or a mode value. This allows the information processing device 100 to obtain a guideline for selecting at least one of the plurality of structure data 101 to be used when updating the model 110. Specifically, the information processing device 100 can make it possible to select, as the structure data 101 to be used when updating the model 110, the structure data 101 corresponding to a case where the interatomic distance is relatively short, which may cause the MD calculation to fail.

[0029] (1-3) The information processing device 100 identifies coordinates in the multidimensional space 120 that represent a combination of the acquired multiple first feature amounts and the calculated second feature amounts corresponding to each of the multiple structure data 101. The multidimensional space 120 is, for example, a three-dimensional space. The information processing device 100 refers to the identified coordinates and selects at least one of the multiple structure data 101 to be used when updating the model 110 based on the distribution of the coordinates in the multidimensional space 120. The information processing device 100 selects one or more structure data 101 in the multidimensional space 120, for example, uniformly from the perspective of the second feature amount related to the distance between atoms in addition to the first feature amount.

[0030] This allows the information processing device 100 to select appropriate structure data 101 to be used when updating the model 110 so as to improve the accuracy of the model 110. Here, for example, in a conventional method, the structure data 101 indicated by a thick circle is selected based on multiple first feature quantities without taking the second feature quantity into consideration, and it may be impossible to select structure data 101 corresponding to a case where the distance between atoms is relatively short. In contrast, the information processing device 100 can select structure data 101 included in region 121 of the multidimensional space 120 and corresponding to a case where the distance between atoms is relatively short. This allows the information processing device 100 to easily and efficiently update the model 110. For example, the information processing device 100 can reduce the time required to update the model 110. For example, the information processing device 100 can improve the accuracy of the updated model 110.

[0031] (1-4) The information processing device 100 updates the model 110 based on at least one of the selected structural data 101. For example, the information processing device 100 performs a DFT calculation on at least one of the selected structural data 101, and generates training data in which the structural data 101 is associated with a label indicating the correct output. The label indicates, for example, the energy and force related to the target molecule.

[0032] The information processing device 100 updates the model 110 based on, for example, the generated training data. Specifically, the information processing device 100 updates the parameters of the model 110 so that the output of the model 110 when the structure data 101 of the generated training data is input to the model 110 approaches the correct output indicated by the label of the generated training data. This allows the information processing device 100 to efficiently update the model 110 so as to improve the accuracy of the model 110.

[0033] Furthermore, the information processing device 100 can prepare multiple pieces of structure data 101 so that each piece of structure data 101 represents the structure of the target molecule in a reduced system. Therefore, the information processing device 100 can update the model 110 by using structure data 101 that corresponds to an irregular case in which the distance between atoms is relatively close, which can cause the MD calculation to fail. Therefore, the information processing device 100 can easily improve the accuracy of the model 110.

[0034] Here, the case where the functions of the information processing device 100 are realized by a single computer has been described, but this is not limiting. For example, the functions of the information processing device 100 may be realized by the cooperation of multiple computers. Furthermore, for example, the functions of the information processing device 100 may be realized on the cloud.

[0035] Furthermore, for example, the information processing device 100 may cooperate with other computers. Specifically, the information processing device 100 may realize a function of selecting structured data 101 and cooperate with other computers that prepare the structured data 101, other computers that prepare training data, and other computers that update the model 110. The case where the information processing device 100 cooperates with other computers will be described later with reference to, for example, FIG. 2 .

[0036] (Example of Information Processing System 200) Next, an example of an information processing system 200 to which the information processing device 100 shown in FIG. 1 is applied will be described with reference to FIG.

[0037] 2 is an explanatory diagram showing an example of an information processing system 200. In FIG. 2, the information processing system 200 includes an information processing device 100, a training processing device 201, and a client device 202.

[0038] In the information processing system 200, the information processing device 100 and the training processing device 201 are connected via a wired or wireless network 210. The network 210 is, for example, a local area network (LAN), a wide area network (WAN), the Internet, etc. In the information processing system 200, the information processing device 100 and the client device 202 are connected via the wired or wireless network 210.

[0039] The information processing device 100 is a computer that facilitates updating a model that represents a potential. The information processing device 100 acquires an update request that requests updating of the model. In response to the update request, the information processing device 100 transmits a generation request that requests generation of multiple pieces of structure data to the training processing device 201. The information processing device 100 receives the multiple pieces of structure data corresponding to the generation request from the training processing device 201.

[0040] The information processing device 100 acquires, for each of a plurality of pieces of structure data, a plurality of first feature amounts related to the structure of a target molecule when the structure data is input to a model. The information processing device 100 transmits, for example, a calculation request requesting calculation of the first feature amounts to the training processing device 201. The information processing device 100 receives, from the training processing device 201, the plurality of first feature amounts corresponding to the calculation request.

[0041] The information processing device 100 calculates, for each of the plurality of structure data, a second feature value related to the distance between atoms in the structure of the target molecule represented by the structure data. The information processing device 100 selects, from the plurality of structure data, one or more structure data to be used when updating the model, based on the acquired plurality of first feature values ​​corresponding to each of the plurality of structure data and the calculated second feature value. The information processing device 100 transmits an update request to the training processing device 201, which includes the selected one or more structure data and requests that the model be updated.

[0042] The information processing device 100 receives an implementation request requesting that an MD calculation be performed from the client device 202. The information processing device 100 transmits the received implementation request to the training processing device 201. The information processing device 100 receives the results of the MD calculation from the training processing device 201. The information processing device 100 transmits the results of the MD calculation to the client device 202. The information processing device 100 is, for example, a server or a PC.

[0043] The training processing device 201 is a computer having a model. When the training processing device 201 receives a generation request, it uses the model to perform MD calculations on the target molecule to generate multiple pieces of structural data, and transmits the generated data to the information processing device 100. When the training processing device 201 receives a calculation request, it calculates multiple first feature amounts for each piece of structural data, and transmits the calculated data to the information processing device 100.

[0044] Upon receiving the update request, the training processing device 201 performs a DFT calculation on each of the one or more pieces of structure data included in the update request and assigns a label to generate training data. The training processing device 201 updates the model based on the generated training data. Upon receiving the execution request, the training processing device 201 transmits the results of the MD calculation performed using the model to the information processing device 100. The training processing device 201 is, for example, a server or a PC.

[0045] The client device 202 is a computer owned by a system user. The client device 202 generates an implementation request requesting that an MD calculation be performed and transmits it to the information processing device 100. The client device 202 receives the results of the MD calculation from the information processing device 100. The client device 202 outputs the results of the MD calculation so that the system user can refer to them. The client device 202 is, for example, a PC, a tablet terminal, or a smartphone.

[0046] Here, the case where the information processing device 100 is a device different from a training device has been described, but this is not limited to this. For example, the information processing device 100 may have a function as a training device and operate as a training device. In this case, the information processing system 200 does not need to include a training device.

[0047] Here, the case where the information processing device 100 is a device different from the client device 202 has been described, but this is not limited to this. For example, the information processing device 100 may have the function of the client device 202 and may also operate as the client device 202. In this case, the information processing system 200 does not need to include the client device 202.

[0048] (Example of Hardware Configuration of Information Processing Apparatus 100) Next, an example of the hardware configuration of the information processing apparatus 100 will be described with reference to FIG.

[0049] Fig. 3 is a block diagram showing an example of the hardware configuration of the information processing device 100. In Fig. 3, the information processing device 100 has a CPU (Central Processing Unit) 301, a memory 302, and a network I / F (Interface) 303. The information processing device 100 also has a recording medium I / F 304, a recording medium 305, a display 306, and an input device 307. The components are connected to each other via a bus 300.

[0050] Here, the CPU 301 is responsible for overall control of the information processing device 100. The memory 302 includes, for example, a read-only memory (ROM), a random access memory (RAM), and a flash ROM. Specifically, for example, the flash ROM and ROM store various programs, and the RAM is used as a work area for the CPU 301. The programs stored in the memory 302 are loaded into the CPU 301, causing the CPU 301 to execute the coded processes.

[0051] The network I / F 303 is connected to the network 210 via a communication line, and is connected to other computers via the network 210. The network I / F 303 manages the internal interface with the network 210 and controls the input and output of data from other computers. The network I / F 303 is, for example, a modem or a LAN adapter.

[0052] The recording medium I / F 304 controls reading and writing of data from and to the recording medium 305 under the control of the CPU 301. The recording medium I / F 304 is, for example, a disk drive, a solid state drive (SSD), or a universal serial bus (USB) port. The recording medium 305 is a non-volatile memory that stores data written under the control of the recording medium I / F 304. The recording medium 305 is, for example, a disk, a semiconductor memory, or a USB memory. The recording medium 305 may be detachable from the information processing device 100.

[0053] The display 306 displays data such as a cursor, an icon, a toolbox, a document, an image, or function information. The display 306 is, for example, a cathode ray tube (CRT), a liquid crystal display, or an organic electroluminescence (EL) display. The input device 307 has keys for inputting letters, numbers, various instructions, etc., and inputs data. The input device 307 is, for example, a keyboard or a mouse. The input device 307 may also be, for example, a touch panel input pad or a numeric keypad.

[0054] The information processing device 100 may include, in addition to the above-described components, a camera, for example. The information processing device 100 may include, in addition to the above-described components, a printer, a scanner, a microphone, or a speaker, for example. The information processing device 100 may include, for example, a plurality of recording medium I / Fs 304 and recording media 305. The information processing device 100 may not include, for example, a display 306 or an input device 307. The information processing device 100 may not include, for example, a recording medium I / F 304 or a recording medium 305.

[0055] (Example of Hardware Configuration of Training Processing Device 201) An example of the hardware configuration of the training processing device 201 is similar to the example of the hardware configuration of the information processing device 100 shown in FIG. 3, and therefore a description thereof will be omitted.

[0056] (Example of Hardware Configuration of Client Device 202) The example of the hardware configuration of the client device 202 is similar to the example of the hardware configuration of the information processing device 100 shown in FIG. 3, and therefore a description thereof will be omitted.

[0057] (Example of Functional Configuration of Information Processing Apparatus 100) Next, an example of the functional configuration of the information processing apparatus 100 will be described with reference to FIG.

[0058] 4 is a block diagram showing an example of the functional configuration of the information processing device 100. The information processing device 100 includes a storage unit 400, an acquisition unit 401, a generation unit 402, a first calculation unit 403, a second calculation unit 404, a selection unit 405, a training unit 406, and an output unit 407.

[0059] The storage unit 400 is realized by, for example, a storage area such as the memory 302 or the recording medium 305 shown in Fig. 3. In the following, a case where the storage unit 400 is included in the information processing device 100 will be described, but this is not limiting. For example, the storage unit 400 may be included in a device different from the information processing device 100, and the stored contents of the storage unit 400 may be accessible from the information processing device 100.

[0060] The acquiring unit 401 to the output unit 407 function as an example of a control unit. Specifically, the acquiring unit 401 to the output unit 407 realize their functions by, for example, causing the CPU 301 to execute a program stored in a storage area such as the memory 302 or the recording medium 305 shown in Fig. 3, or by using the network I / F 303. The processing results of each functional unit are stored in a storage area such as the memory 302 or the recording medium 305 shown in Fig. 3.

[0061] The storage unit 400 stores various information that is referenced or updated during processing by each functional unit. The storage unit 400 stores, for example, a model. The model has a function of generating output data that represents energy and forces related to a target molecule in response to input data that represents the structure of the target molecule. The structure includes, for example, the positions of each of the multiple atoms that form the target molecule. The structure includes, for example, the velocities of each of the multiple atoms that form the target molecule.

[0062] The model is, for example, a neural network. Specifically, the model is a neural network generated based on one or more pieces of training data. The training data, for example, represents structure data representing an input sample and a label that is the correct output answer, in association with each other. The structure data represents the structure of the target molecule. The label indicates the energy and force related to the target molecule. The model is, for example, set in advance by a user. The model may be, for example, acquired by the acquisition unit 401. The model is, for example, updated by the training unit 406.

[0063] The storage unit 400 stores, for example, a plurality of pieces of structure data. Preferably, each of the plurality of pieces of structure data represents the structure of the target molecule in a reduced system. The structure includes, for example, the position of each of the plurality of atoms forming the target molecule. The structure includes, for example, the velocity of each of the plurality of atoms forming the target molecule. Each of the plurality of pieces of structure data may represent the structure of the target molecule in a non-reduced system. The structure data is generated, for example, by the generation unit 402. The structure data may be acquired, for example, by the acquisition unit 401.

[0064] The storage unit 400 stores, for example, one or more pieces of training data. The training data is used when updating the model. Each of the one or more pieces of training data represents structure data representing an input sample and a label representing a correct output answer, in association with each other. The structure data represents the structure of a target molecule. Specifically, the label indicates the energy and force related to the target molecule, which correspond to the structure of the target molecule represented by the structure data. The training data is generated, for example, by the training unit 406.

[0065] The acquisition unit 401 acquires various types of information used in processing by each functional unit. The acquisition unit 401 stores the acquired various types of information in the storage unit 400 or outputs it to each functional unit. The acquisition unit 401 may also output the various types of information stored in the storage unit 400 to each functional unit. The acquisition unit 401 acquires various types of information based on, for example, a user's operation input. The acquisition unit 401 may also receive various types of information from, for example, a device other than the information processing device 100.

[0066] The acquisition unit 401 acquires, for example, an update request requesting that a model be updated. The update request may, for example, specify a target molecule. The update request may, for example, include a model. The update request may, for example, include multiple pieces of structure data. Specifically, the acquisition unit 401 acquires the update request by accepting an input of the update request. Specifically, the acquisition unit 401 may acquire the update request by receiving the update request from another computer. The other computer is, for example, the client device 202.

[0067] The acquisition unit 401 acquires, for example, a model. Specifically, the acquisition unit 401 acquires the model by accepting input of the model. Specifically, the acquisition unit 401 may acquire the model by receiving the model from another computer. The other computer is, for example, the client device 202. Specifically, the acquisition unit 401 may acquire the model by extracting the model from an update request.

[0068] The acquiring unit 401 acquires, for example, a plurality of pieces of structure data. Specifically, the acquiring unit 401 acquires the plurality of pieces of structure data by accepting input of the plurality of pieces of structure data. Specifically, the acquiring unit 401 may acquire the plurality of pieces of structure data by receiving the plurality of pieces of structure data from another computer. The other computer is, for example, the client device 202. Specifically, the acquiring unit 401 may acquire the plurality of pieces of structure data by extracting the plurality of pieces of structure data from an update request.

[0069] The acquisition unit 401 may receive a start trigger for starting processing of any of the functional units. The start trigger may be, for example, a predetermined operation input by a user. The start trigger may be, for example, reception of predetermined information from another computer. The start trigger may be, for example, output of predetermined information by any of the functional units. The acquisition unit 401 receives, for example, acquisition of an update request as a start trigger for starting processing of the generation unit 402, the first calculation unit 403, the second calculation unit 404, the selection unit 405, and the training unit 406.

[0070] The generation unit 402 acquires the plurality of structure data by generating the plurality of structure data. The generation unit 402 generates the plurality of structure data, for example, by reducing the system and performing MD calculations using a model. The MD calculations may be coarse-grained. Specifically, the generation unit 402 randomly sets a plurality of initial structures of the target molecule. Specifically, the generation unit 402 numerically integrates the equation of motion for each of the set initial structures and simulates the time evolution for the initial structure, thereby generating a plurality of structure data each representing the structure of the target molecule. The generation unit 402 may also generate the plurality of structure data by, for example, performing MD calculations using a model without reducing the system. This allows the generation unit 402 to prepare a plurality of structure data that serve as the basis for training data.

[0071] The first calculation unit 403 calculates, for each of the plurality of structural data, a plurality of first feature quantities related to the structure of the target molecule in accordance with the result of inputting the structural data into the model. For example, the first calculation unit 403 calculates, for each of the plurality of structural data, a plurality of first feature quantities related to the structure of the target molecule by performing principal component analysis on the output of the intermediate layer of the model when the structural data is input into the model. The number of first feature quantities is, for example, two. This allows the first calculation unit 403 to obtain a guideline for selecting at least one of the plurality of structural data to be used when updating the model.

[0072] The second calculation unit 404 calculates, for each of the plurality of structure data, a second feature value related to the distance between atoms in the structure of the target molecule represented by the structure data. The second feature value is, for example, a statistical value of the distance between atoms in the structure of the target molecule represented by the structure data. The statistical value is, for example, a minimum value, a maximum value, an average value, a median value, or a mode value. This allows the second calculation unit 404 to obtain a guideline for selecting at least one of the plurality of structure data to be used when updating the model.

[0073] The selection unit 405 sets coordinates in a multidimensional space that represent combinations of the acquired first feature amounts and the calculated second feature amounts corresponding to each of the plurality of structure data. Based on the distribution of the set coordinates, the selection unit 405 selects at least one of the plurality of structure data to be used when updating the model.

[0074] For example, based on the distribution of the set coordinates, the selection unit 405 calculates the distance between the coordinates corresponding to each of the plurality of structure data and the coordinates corresponding to other structure data. The distance is, for example, Euclidean distance. For example, the selection unit 405 selects a predetermined number of structure data from the plurality of structure data in descending order of the structure data having the largest calculated distance from the other structure data.

[0075] Specifically, the selection unit 405 selects a predetermined number of structured data from among the plurality of structured data in descending order of the statistical value of the calculated distance from other structured data. The statistical value may be, for example, a minimum value, a maximum value, an average value, a median value, or a mode value. This allows the selection unit 405 to select appropriate structured data to be used when updating the model so as to improve the accuracy of the model.

[0076] The training unit 406 updates the model. For example, the training unit 406 calculates the energy and force of the target molecule by performing a DFT calculation on each piece of structure data selected by the selection unit 405, and sets a label corresponding to the structure data. For example, the training unit 406 generates training data that represents each piece of concept data selected by the selection unit 405 in association with the structure data and the set label. The training unit 406 updates the model based on the generated training data, for example.

[0077] Specifically, the training unit 406 updates the parameters of the model so that when structural data from the generated training data is input to the model, the output of the model approaches the labels from the generated training data, thereby enabling the training unit 406 to efficiently update the model so as to improve its accuracy.

[0078] Furthermore, the information processing device 100 may repeatedly perform a series of processes for updating the model using the generation unit 402, the first calculation unit 403, the second calculation unit 404, the selection unit 405, and the training unit 406. The information processing device 100, for example, repeatedly performs the series of processes until a predetermined termination condition is met. The predetermined termination condition may be, for example, performing the series of processes a predetermined number of times. The predetermined termination condition may be, for example, that the amount of change in the model parameters before and after the update is equal to or less than a threshold. In this case, the generation unit 402 generates, for example, new structural data, each representing the structure of the target molecule in the reduced system.

[0079] The first calculation unit 403 acquires, for example, for each of the plurality of new structure data, a plurality of first feature quantities related to the structure of the target molecule in accordance with a result of inputting the structure data into the previously updated model. The second calculation unit 404 calculates, for each of the plurality of new structure data, a second feature quantity related to the distance between atoms in the structure of the target molecule represented by the structure data.

[0080] The selection unit 405 sets coordinates in a multidimensional space that represent combinations of the acquired first feature amounts and the calculated second feature amounts corresponding to each of the new structure data pieces. The selection unit 405 selects at least one of the new structure data pieces to be used when updating the model, based on the distribution of the set coordinates, for example.

[0081] The training unit 406, for example, performs a DFT calculation on each piece of structure data selected by the selection unit 405, thereby setting a label corresponding to the structure data. The training unit 406, for example, generates training data that associates each piece of structure data selected by the selection unit 405 with the set label. The training unit 406 updates the model, for example, based on the generated training data. This allows the information processing device 100 to repeatedly update the model so as to improve the accuracy of the model.

[0082] The output unit 407 outputs the processing results of at least one of the functional units. The output format may be, for example, display on a display, printout to a printer, transmission to an external device via the network I / F 303, or storage in a storage area such as the memory 302 or the recording medium 305. This allows the output unit 407 to notify the user of the processing results of at least one of the functional units, thereby improving the convenience of the information processing device 100.

[0083] The output unit 407 outputs, for example, the structure data selected by the selection unit 405. Specifically, the output unit 407 outputs the structure data selected by the selection unit 405 so that the user can refer to it. Specifically, the output unit 407 transmits the structure data selected by the selection unit 405 to another computer. The other computer is, for example, the client device 202. In this way, the output unit 407 can provide appropriate structure data to be used when updating the model to the outside, and can enable the process of updating the model to be performed externally.

[0084] The output unit 407 outputs, for example, the training data generated by the training unit 406. Specifically, the output unit 407 outputs the training data generated by the training unit 406 so that the user can refer to it. Specifically, the output unit 407 transmits the training data generated by the training unit 406 to another computer. The other computer is, for example, the client device 202. In this way, the output unit 407 can provide appropriate training data to be used when updating the model to an external device, and can enable the model updating process to be performed externally.

[0085] The output unit 407 outputs, for example, the model updated by the training unit 406. Specifically, the output unit 407 outputs the model updated by the training unit 406 so that it can be referenced by a user. Specifically, the output unit 407 transmits the model updated by the training unit 406 to another computer. The other computer is, for example, the client device 202. In this way, the output unit 407 can make the model with improved accuracy for estimating the energy and the like related to the target molecule available externally.

[0086] Here, the case where the information processing device 100 includes the generating unit 402, the first calculating unit 403, the second calculating unit 404, the selecting unit 405, and the training unit 406 has been described, but this is not limiting. For example, the information processing device 100 may not include any of the functional units. Specifically, the information processing device 100 may not include the generating unit 402. In this case, the information processing device 100 may specifically cooperate with another computer including the generating unit 402.

[0087] (Example of Operation of Information Processing Device 100) Next, an example of operation of the information processing device 100 will be described with reference to FIGS.

[0088] 5 is an explanatory diagram showing an example of an operation of the information processing device 100. In FIG. 5, the information processing device 100 stores a machine learning potential 500. The machine learning potential 500 is a neural network. The machine learning potential 500 has a function of generating output data representing the energy and force related to a target molecule, etc., in accordance with input data representing the structure of the target molecule.

[0089] (5-1) The information processing device 100 generates a structure data group 501 including a plurality of structure data for which the energy and force related to the target molecule are unknown by performing an MD calculation using the machine learning potential 500. The structure data group 501 may include, for example, a plurality of structure data for which the energy and force related to the target molecule are known. Now, moving on to the description of FIG. 6 , an example of the MD calculation will be described.

[0090] FIG. 6 is an explanatory diagram showing an example of MD calculation. In FIG. 6, the information processing device 100 stores an equation of motion that uses a machine learning potential 500. (6-1) The information processing device 100 sets an initial structure for a target molecule in which the size of the system has been reduced using volume scanning. The size of the system may be reduced, for example, by setting a pressure on the system. The initial structure includes, for example, the coordinates r of each of the multiple atoms that form the target molecule and the velocity v of each atom. The information processing device 100, for example, randomly sets the initial structure for the target molecule.

[0091] The information processing device 100 performs a series of processes shown in (6-2) and (6-3) below on the set initial structure of the target molecule. (6-2) The information processing device 100 calculates the force f acting on the atom based on the structure of the target molecule and a constant related to the force f acting on the atom. (6-3) The information processing device 100 numerically integrates the equation of motion to update the coordinates r of each atom and the velocity v of each atom, thereby updating the structure of the target molecule.

[0092] The information processing device 100 simulates the time evolution of the target molecule by repeatedly performing the series of processes described in (6-2) and (6-3) above on the structure of the target molecule that was most recently updated a predetermined number of times. The information processing device 100 generates structure data representing the structure of the target molecule that was last updated by repeatedly performing the series of processes a predetermined number of times. The information processing device 100 compiles the generated structure data in the case where each of a plurality of different structures that were randomly set is adopted as an initial structure, and sets the generated structure data as a structure data group 501.

[0093] 5, (5-2) the information processing device 100 selects, from the structured data group 501, a structured data group 502 to be used when updating the machine learning potential 500. Now, moving on to the explanation of FIG. 7, an example will be described in which the information processing device 100 selects the structured data group 502 from the structured data group 501.

[0094] 7 is an explanatory diagram showing an example of selecting a structure data group 502. In the example of Fig. 7, the structure data group 501 includes a plurality of unlabeled structure data in which, for example, the energy and force related to the target molecule are unknown. The labels indicate the energy and force related to the target molecule.

[0095] Furthermore, the structure data group 501 includes a plurality of labeled structure data for which, for example, the energy and force related to the target molecule are known. In Fig. 7, unlabeled structure data corresponds to pg10 (unlabeled). Labeled structure data corresponds to pg10 (labeled).

[0096] (7-1) When the information processing device 100 inputs each piece of structural data in the structural data group 501 into the machine learning potential 500, it acquires a plurality of feature quantities output by the intermediate layer of the machine learning potential 500. The information processing device 100 calculates feature quantity 1 and feature quantity 2 related to the structure of the target molecule by performing principal component analysis on each piece of structural data in the structural data group 501 based on the acquired plurality of feature quantities.

[0097] 7 shows a two-dimensional space having axes corresponding to feature 1 and feature 2 related to the structure of a target molecule. Graph 700 shows the distribution of each structural data in the two-dimensional space when feature 1 and feature 2 corresponding to each structural data are set as coordinate values ​​on each axis.

[0098] (7-2) For each piece of structure data in the structure data group 501, the information processing device 100 calculates, as feature quantity 3, the minimum interatomic distance in the structure of the target molecule represented by the structure data. Feature quantity 3 may preferably represent the minimum distance between atom O and atom H. Graph 710 in FIG. 7 shows a three-dimensional space having axes corresponding to feature quantity 1, feature quantity 2, and feature quantity 3, respectively. Graph 710 shows the distribution of each piece of structure data in the three-dimensional space when feature quantity 1, feature quantity 2, and feature quantity 3 corresponding to each piece of structure data are used as coordinate values ​​for each axis.

[0099] (7-3) For each piece of structured data in the structured data group 501, the information processing device 100 calculates a statistical value of the Euclidean distance between that piece of structured data and other piece of structured data in three-dimensional space. The statistical value is, for example, a minimum value, a maximum value, an average value, a median value, or a mode value. In the example of Fig. 7, the statistical value is specifically the maximum value. The information processing device 100 selects a predetermined number of pieces of structured data from the structured data group 501 in descending order of the calculated statistical value, thereby selecting a structured data group 502 including a predetermined number of pieces of structured data.

[0100] This allows the information processing device 100 to select a structure data group 502 that includes appropriate structure data for improving the accuracy of the machine learning potential 500. The information processing device 100 can select the structure data group 502 so that it includes a balanced amount of structure data that corresponds to cases where the distance between atoms is relatively short, which can be a factor in the failure of the MD calculation, for example. This allows the information processing device 100 to easily and efficiently update the machine learning potential 500 so as to improve the accuracy of the machine learning potential 500.

[0101] Returning to the explanation of FIG. 5 , (5-3) the information processing device 100 performs a DFT calculation on each piece of structure data in the selected structure data group 502, calculates the energy and force related to the target molecule, and sets a label. The label indicates the energy and force related to the target molecule. The information processing device 100 may calculate the energy and force related to the target molecule and set a label using a method other than the DFT calculation.

[0102] (5-4) For each piece of structured data in the selected structured data group 502, the information processing device 100 generates training data that represents the structured data as an input sample and represents the set label as an output correct answer. As a result, the information processing device 100 compiles the generated training data to prepare a training data group 503. The information processing device 100 registers the training data group 503 in a dataset 510. The dataset 510 is realized by the memory 302.

[0103] (5-5) The information processing device 100 refers to the dataset 510 and updates the machine learning potential 500 based on the training data group 503. The update is achieved, for example, by backpropagation. The information processing device 100 updates the parameters of the machine learning potential 500 so that the output of the machine learning potential 500 in response to the input of structural data included in each training data approaches the label included in the training data. This allows the information processing device 100 to improve the accuracy of the machine learning potential 500.

[0104] The information processing device 100 may repeatedly perform the series of processes shown in (5-1), (5-2), (5-3), (5-4), and (5-5) above. This allows the information processing device 100 to improve the accuracy of the machine learning potential 500. Next, we will move on to the description of FIG. 8 and describe an example of the effect achieved by the information processing device 100.

[0105] FIG. 8 is an explanatory diagram showing an example of the effect. In FIG. 8, table 800 shows the results of comparing the method using the information processing device 100 with the first conventional method, the second conventional method, and the third conventional method. In the following description, the method using the information processing device 100 may be referred to as the "proposed method." The first conventional method selects 11,597 pieces of structured data to be labeled from among the plurality of pieces of structured data that do not reduce the size of the system, based on two-dimensional features related to each piece of structured data. This prepares 11,597 pieces of training data.

[0106] The second conventional method, like the first conventional method, selects 11,597 pieces of structured data and prepares 11,597 pieces of training data. Then, the second conventional method selects an additional 1,000 pieces of structured data to be labeled from the multiple pieces of structured data based on two-dimensional features related to each piece of structured data in the multiple pieces of structured data with the reduced system size, and prepares an additional 1,000 pieces of training data. Therefore, the second conventional method prepares a total of 12,597 pieces of training data, including the 11,597 pieces of training data prepared in the same way as the first conventional method and the new 1,000 pieces of training data.

[0107] The third conventional method, like the first conventional method, selects 11,597 pieces of structured data and prepares 11,597 pieces of training data. Then, the third conventional method selects 500 pieces of structured data to be labeled from the plurality of pieces of structured data based on two-dimensional features related to each piece of structured data obtained by reducing the size of the system, and prepares an additional 500 pieces of training data. Therefore, the third conventional method prepares a total of 12,097 pieces of training data, including the 11,597 pieces of training data prepared like the first conventional method and the new 500 pieces of training data.

[0108] The information processing device 100 selects 11,597 pieces of structural data and prepares 11,597 pieces of training data, as in the first conventional method. Then, the information processing device 100 selects 500 pieces of structural data to be labeled from the plurality of pieces of structural data based on three-dimensional feature quantities related to each piece of structural data of the plurality of pieces of structural data with reduced system size, and prepares an additional 500 pieces of training data. As described above, the three-dimensional feature quantities include feature quantities related to interatomic distances. Specifically, the interatomic distance is the O-H distance. O is oxygen. H is hydrogen. Therefore, the information processing device 100 prepares a total of 12,097 pieces of training data, including the 11,597 pieces of training data prepared in the same manner as the first conventional method and the new 500 pieces of training data.

[0109] Table 800 shows the number of structural data (number of additional data), the histogram of O-H distances [distance: Å], and the MD calculation time [ns] for each of the proposed method, the first conventional method, the second conventional method, and the third conventional method. The number of structural data is the total number of selected structural data. The number of structural data corresponds to the total number of prepared training data. The number of additional data is the total number of additionally selected structural data. The number of additional data corresponds to the total number of additionally prepared training data. The histogram of O-H distances [distance: Å] is a frequency distribution table of O-H distances for multiple structural data.

[0110] Specifically, the MD calculation time [ns] indicates the cumulative MD calculation time when the MD calculation is performed four times until it is completed within 20 ns or collapses. Specifically, if any of the MD calculations collapses before completing 20 ns, the MD calculation time [ns] will be less than 80 ns. The smaller the MD calculation time [ns], the more likely the MD calculation is to collapse, indicating lower accuracy of the model based on the prepared training data.

[0111] As shown in Table 800, the proposed method can select structural data with smaller O-H distances than the first conventional method, the second conventional method, and the third conventional method. Therefore, the proposed method can improve the accuracy of models for target molecular structures with small O-H distances based on the selected structural data.

[0112] Therefore, as shown in Table 800, the proposed method, unlike the first and third conventional methods, can perform MD calculations without collapsing by improving the accuracy of models for target molecular structures with small O-H distances. Also, as shown in Table 800, the proposed method, unlike the second conventional method, can reduce the number of additional structural data items to be selected, thereby reducing the time required to update the model.

[0113] (Overall Processing Procedure) Next, an example of an overall processing procedure executed by the information processing device 100 will be described with reference to Fig. 9. The overall processing is realized by, for example, the CPU 301, storage areas such as the memory 302 and the recording medium 305, and the network I / F 303 shown in Fig. 3.

[0114] 9 is a flowchart showing an example of the overall processing procedure, in which the information processing apparatus 100 generates a plurality of structure data by contracting the system using volume scanning (step S901).

[0115] Next, the information processing device 100 inputs each of the structural data into the neural network and extracts two first feature quantities related to each of the structural data from the neural network (step S902).The information processing device 100 then calculates a second feature quantity representing the minimum interatomic distance for each of the structural data (step S903).

[0116] Next, the information processing device 100 calculates the distances between the coordinate points of each structure data in three-dimensional space, based on the coordinate points of the structure data represented by a combination of two first feature amounts and one second feature amount for the structure data (step S904).The information processing device 100 then selects a predetermined number of coordinates in descending order of the distances from other coordinates based on the distances between the coordinate points (step S905).

[0117] Next, the information processing device 100 performs DFT calculations on the structure data corresponding to each of the selected coordinates and assigns labels (step S906).The information processing device 100 then updates the neural network based on the labeled structure data (step S907).

[0118] The information processing device 100 performs MD calculations using the updated NN (step S908). The information processing device 100 ends the entire process. This allows the information processing device 100 to improve the accuracy of the NN. Here, the information processing device 100 may change the order of the processes of some of the steps in the flowchart of FIG. 9 and execute them. For example, the order of the processes of steps S902 and S903 can be changed.

[0119] Furthermore, the information processing device 100 may omit some of the steps in the flowchart of Fig. 9. For example, the process of step S908 can be omitted. Specifically, the information processing device 100 may further improve the accuracy of the NN by performing the overall process multiple times without performing the process of step S908. Specifically, the information processing device 100 may perform the process of step S908 after performing the overall process multiple times.

[0120] (Application Examples of the Information Processing Device 100) Next, application examples of the information processing device 100 will be described. The information processing device 100 can be applied, for example, to the fields of drug discovery and material development. Specifically, the information processing device 100 can be applied when it is desired to perform MD calculations to investigate structural changes in proteins.

[0121] As described above, the information processing device 100 can store a model that generates output data representing the energy of a target molecule in response to input data representing the structure of the target molecule. The information processing device 100 can acquire multiple first feature values ​​related to the structure of the target molecule for each of multiple pieces of structure data representing the structure of the target molecule in response to the results of inputting the structure data into the model. The information processing device 100 can calculate, for each piece of structure data, a second feature value related to the interatomic distance in the structure of the target molecule represented by the structure data. The information processing device 100 can select at least one piece of structure data to be used when updating the model based on the distribution of coordinates in multidimensional space that represent combinations of multiple first feature values ​​and second feature values. This allows the information processing device 100 to select appropriate structure data to be used when updating the model to improve the accuracy of the model.

[0122] The information processing device 100 can utilize multiple pieces of structural data each representing the structure of a target molecule in a reduced system. This allows the information processing device 100 to prepare structural data corresponding to cases where the distance between atoms is relatively short, thereby improving the accuracy of the model.

[0123] According to the information processing device 100, it is possible to update a model based on training data that associates any of the selected structural data with the energy of the target molecule obtained by performing a DFT calculation on the structural data, thereby enabling the information processing device 100 to prepare appropriate training data including the energy accurately determined by the DFT calculation.

[0124] The information processing device 100 can acquire, for each of a plurality of new structure data, a plurality of first feature quantities related to the structure of a target molecule in accordance with the result of inputting the structure data into an updated model. The information processing device 100 can calculate, for each of the plurality of new structure data, a second feature quantity related to the distance between atoms in the structure of the target molecule represented by the structure data. The information processing device 100 can select at least one of the plurality of new structure data to be used when updating the model based on the distribution of coordinates in multidimensional space representing combinations of a plurality of first feature quantities and a plurality of second feature quantities. This allows the information processing device 100 to further update the model to improve its accuracy.

[0125] According to the information processing device 100, a predetermined number of structure data can be selected from among a plurality of structure data in descending order of the Euclidean distance from other structure data based on the distribution of coordinates. As a result, the information processing device 100 can uniformly select a predetermined number of structure data from the viewpoint of the minimum interatomic distance, so as to select structure data corresponding to a case where the interatomic distance is relatively short, which may cause the MD calculation to fail.

[0126] According to the information processing device 100, it is possible to generate a plurality of pieces of structure data by performing MD calculations using a model, thereby enabling the information processing device 100 to prepare a plurality of pieces of structure data by itself.

[0127] The information processing device 100 can employ a neural network (NN) for the model. The information processing device 100 can acquire a plurality of first feature quantities related to the structure of a target molecule by performing principal component analysis on the output of an intermediate layer of the model when the structural data is input to the model for each of a plurality of structural data. This allows the information processing device 100 to efficiently acquire a plurality of first feature quantities related to the structure of a target molecule, making it easier to uniformly select structural data from the perspective of the structure of the target molecule.

[0128] The information processing device 100 can employ a neural network (NN) generated based on a plurality of training data sets that respectively correspond to input data samples representing the structure of a target molecule and output data solutions representing the energy of the target molecule, thereby enabling the information processing device 100 to utilize a model in a format that can accurately represent potential.

[0129] The information processing device 100 can calculate a second feature amount indicating the minimum interatomic distance in the structure of a target molecule. This makes it easier for the information processing device 100 to select structure data corresponding to a case where the interatomic distance is relatively short, which may cause the MD calculation to fail, from the perspective of the minimum interatomic distance.

[0130] The information processing method described in this embodiment can be realized by executing a prepared program on a computer such as a PC or a workstation. The information processing program described in this embodiment is recorded on a computer-readable recording medium and executed by being read from the recording medium by the computer. The recording medium may be a hard disk, a flexible disk, a CD (Compact Disc)-ROM, an MO (Magneto Optical Disc), a DVD (Digital Versatile Disc), or the like. The information processing program described in this embodiment may also be distributed via a network such as the Internet.

[0131] 100 Information processing device 101 Structured data 110 Model 120 Multidimensional space 121 Region 200 Information processing system 201 Training processing device 202 Client device 210 Network 300 Bus 301 CPU 302 Memory 303 Network I / F 304 Recording medium I / F 305 Recording medium 306 Display 307 Input device 400 Storage unit 401 Acquisition unit 402 Generation unit 403 First calculation unit 404 Second calculation unit 405 Selection unit 406 Training unit 407 Output unit 500 Machine learning potential 501, 502 Structured data group 503 Training data group 510 Data set 700, 710 Graph 800 Table

Claims

1. An information processing program that causes a computer to execute the following processes: acquire a plurality of first feature amounts related to the structure of the target molecule in accordance with the results of inputting the structure data into a model that generates output data representing the energy of the target molecule in accordance with the input data representing the structure of the target molecule for each of the plurality of structure data; calculate a second feature amount related to the distance between atoms in the structure of the target molecule represented by the structure data for each of the plurality of structure data; and refer to coordinates in a multidimensional space that represent a combination of the acquired plurality of first feature amounts and the calculated second feature amount corresponding to each of the plurality of structure data, and select at least one of the plurality of structure data to be used when updating the model based on the distribution of the coordinates.

2. The information processing program according to claim 1, wherein the plurality of structural data each represent a structure of a target molecule in a reduced system.

3. The information processing program according to claim 2, characterized in that the computer is caused to execute a process of updating the model based on training data that associates any of the selected structural data with the energy related to the target molecule obtained by performing density functional theory calculations on any of the structural data.

4. The information processing program according to claim 3, characterized in that it causes the computer to execute the following processes: for each of a plurality of new structural data representing the structure of the target molecule, acquire a plurality of first feature amounts related to the structure of the target molecule in accordance with the result of inputting the structural data into the updated model; for each of the plurality of new structural data, calculate a second feature amount related to the distance between atoms in the structure of the target molecule represented by the structural data; and, by referring to coordinates in multidimensional space representing a combination of the acquired first feature amounts and the calculated second feature amounts corresponding to each of the plurality of new structural data, select at least one of the plurality of new structural data to be used when updating the model based on the distribution of the coordinates.

5. The information processing program according to claim 1, characterized in that the selection process selects a predetermined number of structured data from the plurality of structured data in order of the structured data having a relatively large Euclidean distance from other structured data based on the distribution of the coordinates.

6. The information processing program according to any one of claims 1 to 5, characterized in that the computer executes a process of generating the plurality of structural data by carrying out a dynamics simulation using the model.

7. An information processing program according to any one of claims 1 to 5, characterized in that the model is a neural network, and the acquisition process acquires a plurality of first feature quantities related to the structure of the target molecule by performing principal component analysis on the output of an intermediate layer of the model when the structural data of each of the plurality of structural data is input into the model.

8. An information processing program according to any one of claims 1 to 5, characterized in that the model is a neural network generated based on a plurality of training data that respectively correspond to samples of input data representing the structure of the target molecule and correct answers of output data representing the energy related to the target molecule.

9. An information processing program according to any one of claims 1 to 5, characterized in that the second feature quantity indicates the minimum value of the distance between atoms in the structure of the target molecule.

10. An information processing method characterized by a computer executing the following processes: acquiring a plurality of first feature quantities related to the structure of the target molecule in accordance with the results of inputting the structure data into a model that generates output data representing the energy related to the target molecule in accordance with the input data representing the structure of the target molecule for each of the plurality of structure data; calculating a second feature quantity related to the distance between atoms in the structure of the target molecule represented by the structure data for each of the plurality of structure data; and referencing coordinates in a multidimensional space that represent a combination of the acquired plurality of first feature quantities and the calculated second feature quantities corresponding to each of the plurality of structure data, and selecting at least one of the plurality of structure data to be used when updating the model based on the distribution of the coordinates.

11. An information processing device characterized by having a control unit that acquires a plurality of first feature amounts related to the structure of a target molecule in accordance with the results of inputting the structure data into a model that generates output data representing the energy related to the target molecule in accordance with the input data representing the structure of the target molecule for each of the plurality of structure data; calculates a second feature amount related to the distance between atoms in the structure of the target molecule represented by the structure data for each of the plurality of structure data; and refers to coordinates in a multidimensional space that represent a combination of the acquired plurality of first feature amounts and the calculated second feature amount corresponding to each of the plurality of structure data, and selects at least one of the plurality of structure data to be used when updating the model based on the distribution of the coordinates.

Images