WO2022163629A1

WO2022163629A1 - Estimation device, training device, estimation method, generation method and program

Info

Publication number: WO2022163629A1
Application number: PCT/JP2022/002604
Authority: WO
Inventors: 幾品川; 大器山口
Original assignee: 株式会社 Preferred Networks
Priority date: 2021-01-28
Filing date: 2022-01-25
Publication date: 2022-08-04

Abstract

[Problem] To estimate HOMO energy with high accuracy and at low calculation cost. [Solution] The estimation device comprises one or a plurality of memories and one or a plurality of processors. The one or plurality of processors acquires, from a graph including a first feature amount related to a molecule, a second feature amount acquired on the basis of a first principle calculation and relating to energy allocated to atoms constituting the molecule, and acquires, on the basis of the graph and the second feature amount, a third feature amount related to one or a plurality of types of energy of the molecule.

Description

Estimation device, training device, estimation method, generation method and program

The present disclosure relates to an estimation device, a training device, an estimation method, a generation method, and a program.

HOMO (Highest Occupied Molecular Orbital) energy and LUMO (Lowest Unoccupied Molecular Orbital) energy are physical property values used for predicting the reactivity of a compound. These values can be calculated using quantum scientific calculations. However, this calculation time requires a calculation amount of O(n ³ ) with respect to the size n of the operation matrix of the target molecule, resulting in a high calculation cost. In recent years, attempts have been made to predict use using neural networks, especially GNNs (Graph Neural Networks). Accuracy may be lower than the prediction result of the physical property value of

This disclosure proposes a technique for reducing the cost and increasing the accuracy in the calculation of predetermined types of energy of molecules such as HOMO energy and LUMO energy.

According to one embodiment, the estimating device comprises one or more memories and one or more processors. The one or more processors acquire, from a graph including the first feature amount related to the molecule, a second feature amount related to the energy assigned to the atoms constituting the molecule obtained based on first-principles calculation, and Based on the graph and the second feature, a third feature for one or more predetermined types of energy of the molecule is obtained.

The figure which shows roughly the process of the estimation apparatus which concerns on one Embodiment. The figure which shows roughly the process of the estimation apparatus which concerns on one Embodiment. The figure which shows roughly the process of the estimation apparatus which concerns on one Embodiment. FIG. 4 is a diagram schematically showing processing of a training device according to one embodiment; FIG. 4 is a diagram for explaining a specific example of analysis according to one embodiment; 1 is a block diagram illustrating an example hardware implementation of an estimator and/or a training device, according to one embodiment. FIG.

Embodiments of the present invention will be described below with reference to the drawings. The drawings and description of the embodiments are given by way of example and are not intended to limit the invention.

(Estimation device)
FIG. 1 is a diagram for schematically explaining the operation of the estimation device 1 according to one embodiment. In the estimation device 1, for example, a graph representing the structure of a molecule (hereinafter also referred to as a structure graph) is processed to obtain HOMO energy.

A structure graph is a graphical representation of the structure of a compound. This structure graph may be, for example, a graph in which atoms are represented as nodes and connection states are represented as edges. This graph may be, for example, an adjacency matrix or the like in which the state of connection between atoms constituting a compound is appropriately described. For example, there may be a plurality of adjacency matrices representing multiple connections. This graph may represent, for example, a matrix representing a geometrical state such as a distance matrix between atoms constituting a compound. In addition to these, any data that can appropriately represent the three-dimensional state of atoms in a compound may be used. A node may have a feature vector that uniquely identifies an atom, such as a one-hot vector, or a feature vector that describes an atom and its surroundings.

The estimation device 1 estimates the HOMO energy from the structure graph using the trained model 10, the first readout function 12, and the second readout function 14. The estimating device 1 includes, for example, a processor that executes processing and a memory (which may include a storage or the like) that stores data, as will be described later in detail.

The trained model 10 has an input layer corresponding to the structure graph described above, and outputs a graph in which information on the structure of molecules is added with information on feature amounts of atoms in the entire molecule or molecular structure (first feature amount). . That is, the trained model 10 converts information (which may be a graph) about the structure of molecules into a graph including the first feature amount and outputs the graph. The trained model 10 may be, for example, a model generated by a graph neural network, graph convolution network, or the like.

The first readout function 12 executes readout to acquire a feature value (second feature value) related to the HOAO (Highest Occupied Atomic Orbital) energy of the atoms constituting the compound from the graph including the second feature value. is a function. This first readout function 12 may be expressed by a combination of linear predetermined operations, or this readout function itself may be configured as a neural network. Hereinafter, in the present disclosure, the HOAO energy refers to the energy corresponding to the HOMO energy of the molecule in the atoms constituting the compound.

This first readout function 12 may be, for example, a function that acquires the HOAO energy of an atom as a second feature quantity when a graph is input. When a compound is composed of atom 1, atom 2, . As shown in the figure, E _{HOAO, atom 1} is output as the HOAO energy of the atom 1 atom, E _{HOAO, atom 2} is output as the HOAO energy of the atom 2 atom, ..., HOAO of the atom n atom Output E _{HOAO, atom n} as energy.

As another example, the first readout function 12 may output a feature quantity that can be easily converted to HOAO energy as the second feature quantity. In this case, if necessary, a linear or nonlinear function for obtaining the HOAO energy from the second feature quantity may be separately provided. The second feature amount may be in a form appropriate as additional information to the graph output by the trained model 10. FIG. That is, the second feature amount may be the HOAO energy of the atom itself, or may be in a form that can be added to the graph as a feature amount related to the HOAO energy even if it is not the HOAO energy itself.

The second readout function 14 implements readout processing for the combined quantity of the graph representing the feature quantity output by the trained model 10 and the second feature quantity obtained by the first readout function 12, and the HOMO energy of the molecule. This is a function that acquires a related feature amount (third feature amount). The second readout function 14, like the first readout function, may also be formed as a linear or non-linear function, for example a trained neural network model.

The second readout function 14 can obtain the HOMO energy of the molecule by using, for example, the concatenated graph output by the trained model 10 and the second feature output by the first readout function 12 as input. to obtain a third feature. As with the second feature amount, the third feature amount may be the HOMO energy itself, or may be a feature amount that can be easily converted into the HOMO energy.

The estimation device 1 outputs the HOMO energy output by the second readout function 14 or the HOMO energy converted from the third feature quantity.

Using this diagram, the processing of the estimation device 1 will be explained.

The processor of the estimation device 1 first acquires information about the structure of the compound (molecule), and if necessary converts this information into a structure graph (S100). Acquisition of information may be realized through an input/output interface. If the acquired information is a structure graph, the processor executes the following process using the acquired structure graph.

If the acquired information is not a structure graph, the processor converts the information about the structure of the compound into a structure graph. This conversion technique may be performed by a commonly used technique. For example, when a compositional formula or a molecular formula is obtained, the processor may obtain a structure graph by collating a database or the like. When obtained as a structural formula, a tool capable of converting from this structural formula may be used. When a structural formula is acquired as tree information, the tree information is appropriately converted into a structure graph.

Next, the processor inputs the acquired structure graph to the trained model 10 to acquire a graph containing the first feature (S102). As described above, this graph is a graph having data obtained by adding the first feature amount that can be derived from the geometric structure of the molecule to the structure graph showing the geometric structure of the molecule.

Next, the processor inputs the obtained graph to the first reading function 12 (S104).

With this first readout function 12, the processor extracts a second feature quantity related to HOAO energy (S106). Alternatively, the HOAO energy may be obtained directly in the first readout function 12 . In other words, the first readout function may be in a form capable of obtaining the HOAO energy itself as the second feature quantity.

Next, the processor concatenates the graph output from the trained model 10 and the second feature amount information output from the first readout function 12 (S108). It should be noted that the second feature amount may be appropriately combined as graph information without being limited to connecting. The second readout function 14 may be configured to appropriately receive the inputs of the second feature amount and the graph separately.

The processor inputs this connected connection graph to the second readout function 14 and acquires the third feature amount regarding the HOMO energy of the molecule (S110). Alternatively, the HOMO energy may be obtained directly in the second readout function 14 . In this case, the next processing of S112 can be omitted. In other words, the second readout function may be in the form of obtaining the HOMO energy itself as the third feature quantity.

Next, the processor obtains the HOMO energy of the molecule based on the third feature quantity output from the second readout function 14 (S112).

In this way, the estimation device 1 obtains the HOMO energy of a compound from the molecular structure and the HOAO energy estimated from this structure.

As described above, according to the present embodiment, it is possible to obtain highly accurate HOMO energy by using information on HOAO energy. The estimating device 1 inputs, as the HOAO energy, information on the maximum value of the HOAO energy that is highly correlated with the HOMO energy, or information using the second feature value related to the information on the maximum value of the HOAO energy, together with the graph, to the second reading function 14. It may be in the form of By using the HOAO energy that is highly correlated with the HOMO energy in this way, highly accurate estimation can be realized with a reduced amount of information to be used.

In the above example, the estimating device 1 that uses HOAO energy to estimate HOMO energy was described, but the same can be achieved with LUMO energy.

Fig. 2 shows an outline of the estimation device 1 for estimating the LUMO energy. As shown in FIG. 2, the estimating device 1 obtains a graph having the first feature by inputting the structure graph to the trained model 10 by the processor, and uses the first readout function 12 from this graph to obtain the LUAO A second feature for the (Lowest Unoccupied Atomic Orbital) energy can be obtained, and a third feature for the LUMO energy can be obtained from the graph and the second feature using the second readout function 14 . Similar to obtaining the HOMO energy described above, the second feature quantity may be the LUAO energy or its lowest value, and the third feature quantity may be the LUMO energy itself. Hereinafter, in the present disclosure, LUAO energy refers to the energy corresponding to the LUMO energy of a molecule in atoms constituting a compound.

FIG. 3 is a diagram showing still another example of the estimation device 1. FIG. As shown in FIG. 3, the estimating device 1 uses the processor to input the structure graph to the trained model 10 to obtain a graph having the first feature amount, and from this graph, the first readout function 12 is used to obtain the HOAO It is also possible to adopt a form in which a second feature amount regarding energy and LOAO energy is obtained, and a third feature amount regarding HOMO energy and LUMO energy is obtained in parallel from the graph and the second feature amount using the second readout function 14 . As in the acquisition of the HOMO energy and LUMO energy described above, the second feature quantity may be the HOAO energy and LUAO energy themselves, and the third feature quantity may be the HOMO energy and LUMO energy themselves.

(training device)
Next, the training device 2 that generates the trained model 10 used in the estimation device 1 will be described.

FIG. 4 is a diagram schematically showing the outline of the processing of the training device 2 according to one embodiment. As shown in FIG. 4, training device 2 performs optimization of model 20 using model 20, first readout function 22, and second readout function . The training device 2 may perform optimization of the first readout function 22 and the second readout function 24 together, if desired. As with the estimation device 1, the training device 2 includes, for example, a processor that executes processing and a memory (which may include a storage or the like) that stores data, as will be described later in detail.

The model 20 trained by the training device 2 is used as the trained model 10 in the estimation device 1 described above. For example, the estimation device 1 may generate the trained model 10 and perform estimation based on information such as the hyperparameters of the model 20 optimized by the training device 2 and the optimized parameters. Thus, model 20 in FIG. 4 may also be, for example, a model based on training a graph neural network.

The estimating device 1 and the training device 2 may operate on the same processor or may operate on different processors. Even if at least one of the first readout function 22 and the second readout function 24 is optimized, the same processing as in the model 20 and the trained model 10 is performed.

Further, as an example, the case of training the trained model 10 corresponding to the estimation device 1 of FIG. 1 will be described, but the case of training the trained model 10 corresponding to the estimation device 1 of FIG. A similar implementation is possible.

The processing of the training device 2 will be explained using FIG.

First, the processor of the training device 2 estimates the HOAO energy and the HOMO energy from the information on the molecular structure by the process (forward propagation process) corresponding to FIG. 1 (S200).

Next, the processor calculates the loss between the HOAO energy and HOMO energy given as teacher data and the HOAO energy and HOMO energy estimated in the processing of S200 (S202). For the HOAO energy given as teacher data, for example, NBO analysis (Natural Bond Orbital Analysis) may be used. Moreover, it is not limited to NBO analysis, and may be obtained from wave functions such as Hamiltonian and Schrödinger's equation, or may be obtained by other appropriate methods.

For calculation of the loss, for example, the mean square error (MSE) between the teacher data and the estimated values of the HOAO energy of the atoms that make up the molecule and the HOMO energy value of the molecule is used. The error used need not be the MSE, e.g. Root Mean Square Error (RMSE), some other norm, or any other suitable error estimator in common machine learning methods A loss function may be used.

Next, the processor executes error backpropagation processing based on the loss calculated in the processing of S202, and updates the parameters of the model 20 (S204). The processor may also update parameters for at least one of the first readout function 22 or the second readout function 24 as needed.

The error backpropagation processing, activation function, loss function, etc. regarding the model 20 may be executed by a general machine learning method.

As described above, according to the training device 2 of the present embodiment, the HOAO energy estimation is performed as an auxiliary method to train the model 20 having appropriate feature values for HOMO energy estimation. becomes possible. By using the HOAO energy supplementally for training, we can increase the number of variables to backpropagate when training the model 20 . Therefore, it is possible to efficiently update the model 20 and the like. In addition, since the HOAO energy is a physical quantity that is strongly related to the HOMO energy, incorporating the HOAO energy into the estimation can improve the accuracy of the HOMO energy estimation. As described for the estimator 1, the maximum of the HOAO energies may be used in backpropagation.

As described in the estimation device 1, the model 20, the first readout function 22, and the second readout function 24 may estimate LUAO energy and LUMO energy. In this case, the training device 2 optimizes the model 20, the first readout function 22, and the second readout function 24 by backpropagating the loss of LUAO energy and LUMO energy. Similarly, HOAO and LUAO energies may be estimated along with HOMO and LUMO energies. In this case as well, the training device 2 performs appropriate optimization according to the output.

　HOMO energy and LUMO energy are indicators of the likelihood of chemical reactions occurring. Therefore, the estimation of the HOMO energy and LUMO energy by the estimation device 1 using the model 20 generated in this way as the trained model 10 can be applied to searches such as material searches and reaction searches, for example. .

As described above, the HOMO energy is estimated based on the HOAO energy, and similarly the LUMO energy is estimated based on the LUAO energy. At least one of the HOAO energy and the LUAO energy may be other energy assigned to atoms calculated based on first-principles calculations. HOAO energy and LUAO energy may be calculated by first-principles calculation.

That is, the estimating device 1 obtains, from the graph including the first feature amount, the second feature amount related to the energy assigned to the atoms forming the molecule obtained based on the first-principles calculation, and calculates the graph and the second feature amount. In other words, obtaining a third feature for at least one of the HOMO energy and the LUMO energy of the molecule based on the feature.

Similarly, the training device 2 inputs a structure graph showing the structure of a molecule to the model, acquires a graph containing the first feature amount related to the molecule, and extracts the molecule acquired based on the first-principles calculation from this graph. Acquiring a second feature value related to the energy assigned to the constituent atoms, acquiring the HOMO energy (or at least one of the LUMO energy) of the molecule based on the graph and the second feature value, and obtaining the energy assigned to the acquired atom and In other words, the HOMO energy (or at least one of the LUMO energies) is compared with teacher data to update the parameters of the model.

Also, at least one of this HOAO energy or LUAO energy is, for example, the energy assigned to each atom calculated based on the NBO analysis. As another example, the energy may be calculated based on wavefunctions rather than NBO analysis. That is, NBO analysis or wave function can be used as first-principles calculation, but it is not limited to these.

Below, we will explain in detail how to obtain HOAO energy and LUAO energy while giving specific examples.

First, the definition of HOAO energy will be explained. By NBO analysis, the energy E _i of the atomic orbital i and the occupancy σ _i can be obtained. Orbital i is assigned to an atom or a bond of two atoms. The HOAO energy for an atom X can be obtained by the following processes.

(1) Extract only the orbitals _i assigned to the atom X, and sort them in descending order of energy Ei. Let the order after rearrangement be trajectory j.

(2) The occupancy σ _i of bonding orbitals or antibonding orbitals, which are orbitals for bonding, is multiplied by 1/2.

(3) Until Σ σ _i = 1,

Calculate

(4) In (3), if Σ σ _i < 1 for all orbits j, instead of equation (1),

Calculate as

For the LUAO energy E _LUAO , sort E _i in descending order in (1) above, and use (1 - σ _j ) instead of σ _j .

For example, suppose that the molecular structure is A - B - C for atoms A, B, and C, and the energy E _i of orbital i can be obtained as follows.
Trajectory 0: (A - B), E = x [eV], σ = 0.05
Trajectory 1: (B - C), E = y [eV], σ = 0.50
Trajectory 2: B, E = z [eV], σ = 0.98
Trajectory 3: A, E = a [eV], σ = 0.99
Now suppose that x > y > z > a.

In this case, the HOAO energy of each atom can be calculated as follows.
A: 0.05/2 x x + min(0.99, 1 - 0.05/2) x a
B: 0.05 / 2 * x + 0.50 / 2 * y + min(0.98, 1 - (0.05 / 2 + 0.50 / 2)) * z
C: (y × 0.50 / 2) / (0.50 / 2)

In atom A, orbital 0 and orbital 3 are extracted in descending order of energy. Orbital 0 has a occupancy of 0.05 and orbital 3 has a occupancy of 0.99, so the occupancy exceeds 1 at orbital 3. Therefore, the HOAO energy is 0.05 / 2 × x + (1 - 0.05 / 2) × a. The term min(·) x a in the above is a term for calculating the energy of orbital 3 in a range where the occupancy does not exceed 1.

More specifically, the occupancy rate exceeds 1 in orbital 3, and based on formula (1), the range where the occupancy rate is 1 in orbital 3 (1 - 0.05 / 2) is the occupancy rate in orbital 3, Calculate HOAO energy. The reason for dividing by 2 is that the orbital 0 is the orbital for the bond between atoms A and B, and corresponds to (2) above.

In this way, formula (1) means calculating the sum of the energies in each orbital until the occupancy becomes 1. That is, when formula (1) is used to calculate the HOAO energy, the orbital having the minimum energy in formula (1) is calculated by regarding the difference between the occupancy rate in the previous orbital and 1 as the occupancy rate. Run.

The same is true for atom B, and orbital 0, orbital 1, and orbital 2 are extracted in descending order of energy. The occupancy of orbital 0 (0.05 / 2) + the occupancy of orbital 1 (0.50 / 2) is less than 1, and the occupancy up to orbital 2 is greater than 1. Therefore, the occupancy rate in orbital 2 is assumed to be (1 - (0.05 / 2 + 0.50 / 2)).

This formula for B is 0.05 / 2 × x + min(0.50 / 2, 1 - 0.05 / 2) × y + min(0.98, 1 - (0.05 / 2 + 0.50 / 2)) may be transformed as x z.

For atom C, the sum of all σ does not exceed 1. Therefore, by applying the formula (2), it can be calculated by the above formula.

A specific example will be described below using the NBO analysis results of water molecules (H ₂ O).

Fig. 5 is a table that extracts only the necessary parts of the NBO analysis results of water molecules.

　In the column of atoms, elements with atomic indexes are shown. A water molecule is composed of three atoms, and the elements of each atom are defined as H, O, and H in index order. Hence, the respective atoms are denoted as H1, O2 and H3.

In principle, up to two electrons can enter one orbital. Therefore, the occupancy is described as a value converted so that the occupancy is 1 when two electrons are in one orbital.

There are no specific regulations for orbital numbers, but orbital numbers are assigned in order of increasing energy in order to facilitate compatibility with (1) to (4) above.

I will explain the calculation of the HOAO energy of atomic O2.

The orbitals containing the atom O2 are 1 to 10 and 13 to 19 in descending order of energy. The occupancy σ is added based on the method of calculating the HOAO energy. Calculating the sum of the occupancy rates in the above order, Σσ _j = 0.000020 < 1 for orbitals 1 to 10, 13 and 14, and Σσ _j = 1.000020 > 1 for orbitals 1 to 10 and 13 to 15. Become. In the

trajectories

13 and 14, the values in the table divided by 2 based on (2) are added.

Calculate the HOAO energy based on equation (1). In orbit 15, the occupancy rate is assumed to be 1.000000 - (1.000020 - 1) = 0.999980 instead of 1.000000. As a result, the HOAO energy of atom O2 is calculated as -0.37941...[hartree]. More specifically, it is represented by Equation (3).

The HOAO energy of atom H1 is similarly calculated. Orbitals containing atom H1 are

orbitals

11, 13 and 17. Even if all

orbitals

11, 13 and 17 are used, the occupancy is 0.5005075<1. Therefore, based on (4) and equation (2) above, the HOQO energy of atom H1 is -0.805638...[hartree]. More specifically, it is represented by Equation (4).

Atomic H2 can be obtained in the same way.

The LUAO energy can be calculated by rearranging the orbitals (or tracing them in reverse order) and taking the occupancy as 1 - σ. For example, consider the LUAO energy of atom H1. Orbitals containing atom H1 are

orbitals

17, 13, and 11 in descending order of energy. The energy is calculated considering the occupancy rate (1 - σ). The occupancy is Σ(1 - σ _j ) = 0.5001175 < 1 for

orbitals

17 and 13, and Σ(1 - σ _j ) = 1.4994925 > 1 for orbital 11. By calculating the occupancy rate of orbital 17 in the same manner as above, the LUAO of atom H1 can be obtained from the following equation.

As a result, the LUAO energy of atom H1 can be obtained as 0.61109...[hartree].

In the above, the estimation of HOMO energy and LUMO energy is referred to as the predetermined type of energy of the molecule, but the embodiments in the present disclosure can also be applied to the estimation of feature values related to energy of molecules in other fields. . For example, it can be applied to intensive variables other than HOMO energy and LUMO energy. As a non-limiting example, atomic charge (corresponding to HOAO energy) can be used to predict molecular polarizability (corresponding to HOMO energy). This is shown as an example, and the embodiment of the present disclosure can be applied to a method of training using the values of the tonic variables localized to atoms in estimating the tonic variables of molecules and compounds. is possible.

Note that the trained model 10, model 20, first readout functions 12, 22, and second readout functions 14, 24 in the estimation device 1 or training device 2 in the present disclosure can define energy etc. as continuous values instead of discrete values. can be anything. When defined as continuous values, see, for example, C. Chen, et. al., “Graph Networks as a Universal Machine Learning Framework for Molecules and Crystals,” https://arxiv.org/abs/ The technique of 1812.05055v2 can be adapted and used, or it can be implemented by other appropriate methods.

All of the above trained models may be concepts that include, for example, models that have been trained as described and further distilled by a general method.

A part or all of each device (estimation device 1 or training device 2) in the above-described embodiments may be configured by hardware, or a CPU (Central Processing Unit), GPU (Graphics Processing Unit), etc. It may be configured by information processing of software (program) to be executed. In the case of software information processing, software that realizes at least a part of the functions of each device in the above-described embodiments can be transferred to a flexible disk, CD-ROM (Compact Disc-Read Only Memory), or USB (Universal Serial Bus) memory or other non-temporary storage medium (non-temporary computer-readable medium) and read into a computer to execute software information processing. Alternatively, the software may be downloaded via a communication network. Furthermore, information processing may be performed by hardware by implementing software in a circuit such as an ASIC (Application Specific Integrated Circuit) or FPGA (Field Programmable Gate Array).

The type of storage medium that stores the software is not limited. The storage medium is not limited to a detachable one such as a magnetic disk or an optical disk, and may be a fixed storage medium such as a hard disk or memory. Also, the storage medium may be provided inside the computer, or may be provided outside the computer.

FIG. 6 is a block diagram showing an example of the hardware configuration of each device (estimating device 1 or training device 2) in the above-described embodiment. Each device includes, for example, a processor 71, a main storage device 72 (memory), an auxiliary storage device 73 (memory), a network interface 74, and a device interface 75, which are connected via a bus 76. may be implemented as a computer 7 integrated with the

The computer 7 in FIG. 6 has one of each component, but may have a plurality of the same components. In addition, although one computer 7 is shown in FIG. 6, the software may be installed in multiple computers, and each of the multiple computers may execute the same or different processing of the software. good too. In this case, it may be in the form of distributed computing in which each computer communicates via the network interface 74 or the like to execute processing. In other words, each device (estimation device 1 or training device 2) in the above-described embodiment is a system that realizes functions by seven or more computers executing instructions stored in seven or more storage devices. may be configured. Alternatively, the information transmitted from the terminal may be processed by one or more computers provided on the cloud, and the processing result may be transmitted to the terminal.

Various operations of each device (estimating device 1 or training device 2) in the above-described embodiments are executed in parallel using one or more processors or using multiple computers via a network. good too. Also, various operations may be distributed to a plurality of operation cores in the processor and executed in parallel. Also, part or all of the processing, means, etc. of the present disclosure may be executed by at least one of a processor and a storage device provided on a cloud capable of communicating with the computer 7 via a network. Thus, each device in the above-described embodiments may be in the form of parallel computing by one or more computers.

The processor 71 may be an electronic circuit (processing circuit, processing circuitry, CPU, GPU, FPGA, ASIC, etc.) including a computer control device and arithmetic device. Also, the processor 71 may be a semiconductor device or the like including a dedicated processing circuit. The processor 71 is not limited to an electronic circuit using electronic logic elements, and may be realized by an optical circuit using optical logic elements. Also, the processor 71 may include arithmetic functions based on quantum computing.

The processor 71 can perform arithmetic processing based on the data and software (programs) input from each device, etc. of the internal configuration of the computer 7, and output the arithmetic result and control signal to each device, etc. The processor 71 may control each component of the computer 7 by executing the OS (Operating System) of the computer 7, applications, and the like.

Each device (estimating device 1 and/or training device 2) in the above-described embodiments may be realized by one or more processors 71. Here, the processor 71 may refer to one or more electronic circuits arranged on one chip, or may refer to one or more electronic circuits arranged on two or more chips or two or more devices. You can point When multiple electronic circuits are used, each electronic circuit may communicate by wire or wirelessly.

The main storage device 72 is a storage device that stores instructions and various data to be executed by the processor 71 , and the information stored in the main storage device 72 is read by the processor 71 . Auxiliary storage device 73 is a storage device other than main storage device 72 . These storage devices mean any electronic components capable of storing electronic information, and may be semiconductor memories. The semiconductor memory may be either volatile memory or non-volatile memory. A storage device for storing various data in each device (estimation device 1 or training device 2) in the above-described embodiments may be realized by the main storage device 73 or the auxiliary storage device 73, and is built into the processor 71. It may be realized by an internal memory. For example, the storage unit 102 in the above-described embodiment may be realized by the main storage device 72 or the auxiliary storage device 73.

Multiple processors may be connected (coupled) to one storage device (memory), or a single processor may be connected. A plurality of storage devices (memories) may be connected (coupled) to one processor. Each device (estimating device 1 or training device 2) in the above-described embodiments is composed of at least one storage device (memory) and a plurality of processors connected (coupled) to this at least one storage device (memory). In this case, at least one of the plurality of processors may include a configuration connected (coupled) to at least one storage device (memory). Also, this configuration may be realized by storage devices (memory) and processors included in a plurality of computers. Furthermore, a configuration in which a storage device (memory) is integrated with a processor (for example, a cache memory including an L1 cache and an L2 cache) may be included.

The network interface 74 is an interface for connecting to the communication network 8 wirelessly or by wire. As for the network interface 74, an appropriate interface such as one conforming to existing communication standards may be used. The network interface 74 may exchange information with the external device 9A connected via the communication network 8. FIG. The communication network 8 may be any one of WAN (Wide Area Network), LAN (Local Area Network), PAN (Personal Area Network), etc., or a combination thereof. It is sufficient if information can be exchanged between them. Examples of WAN include the Internet, examples of LAN include IEEE802.11 and Ethernet (registered trademark), and examples of PAN include Bluetooth (registered trademark) and NFC (Near Field Communication).

The device interface 75 is an interface such as USB that directly connects with the external device 9B.

The external device 9A is a device connected to the computer 7 via a network. External device 9B is a device that is directly connected to computer 7 .

For example, the external device 9A or the external device 9B may be an input device. The input device is, for example, a device such as a camera, microphone, motion capture, various sensors, a keyboard, a mouse, or a touch panel, and provides the computer 7 with acquired information. Alternatively, a device such as a personal computer, a tablet terminal, or a smartphone including an input unit, a memory, and a processor may be used.

Also, the external device 9A or the external device 9B may be, for example, an output device. The output device may be, for example, a display device such as LCD (Liquid Crystal Display), CRT (Cathode Ray Tube), PDP (Plasma Display Panel), or organic EL (Electro Luminescence) panel. A speaker or the like for output may be used. Alternatively, a device such as a personal computer, a tablet terminal, or a smartphone including an output unit, a memory, and a processor may be used.

Also, the external device 9A or the external device 9B may be a storage device (memory). For example, the external device 9A may be a network storage or the like, and the external device 9B may be a storage such as an HDD.

Also, the external device 9A or the external device 9B may be a device having the functions of some of the components of each device (the estimation device 1 or the training device 2) in the above-described embodiments. That is, the computer 7 may transmit or receive part or all of the processing results of the external device 9A or the external device 9B.

In the present specification (including claims), the expression "at least one (one) of a, b and c" or "at least one (one) of a, b or c" (including similar expressions) Where used, includes any of a, b, c, a-b, ac, b-c, or a-b-c. Also, multiple instances of any element may be included, such as a-a, a-b-b, a-a-b-b-c-c, and so on. It also includes the addition of other elements than the listed elements (a, b and c), such as having d such as a-b-c-d.

In this specification (including claims), when expressions such as "data as input / based on data / according to / according to" (including similar expressions) are used, unless otherwise specified, It includes the case where various data itself is used as an input, and the case where various data subjected to some processing (for example, noise added, normalized, intermediate representation of various data, etc.) is used as an input. In addition, if it is stated that some result can be obtained "based on/according to/depending on the data", this includes cases where the result is obtained based only on the data, other data other than the data, It may also include cases where the result is obtained under the influence of factors, conditions, and/or states. In addition, if it is stated that "data will be output", unless otherwise specified, if the various data themselves are used as output, or if the various data have undergone some processing (for example, noise addition, normalization, etc.) This also includes the case where the output is a converted version, an intermediate representation of various data, etc.).

In this specification (including the claims), when the terms "connected" and "coupled" are used, they refer to direct connection/coupling, indirect connection/coupling , electrically connected/coupled, communicatively connected/coupled, operatively connected/coupled, physically connected/coupled, etc. intended as a term. The term should be interpreted appropriately according to the context in which the term is used, but any form of connection/bonding that is not intentionally or naturally excluded is not included in the term. should be interpreted restrictively.

In this specification (including claims), when the phrase "A configured to B" is used, the physical structure of element A is such that it is capable of performing operation B has a configuration, including that a permanent or temporary setting/configuration of element A is configured/set to actually perform action B good. For example, if element A is a general-purpose processor, the processor has a hardware configuration that can execute operation B, and operation B can be performed by setting a permanent or temporary program (instruction). It just needs to be configured to actually run. In addition, when the element A is a dedicated processor or a dedicated arithmetic circuit, etc., regardless of whether or not control instructions and data are actually attached, the circuit structure of the processor actually executes the operation B. It just needs to be implemented.

In this specification (including the claims), when terms denoting containing or possessing (e.g., "comprising/including" and "having, etc.") are used, by the object of the terms It is intended as an open-ended term, including the case of containing or possessing things other than the indicated object. When the object of these terms of inclusion or possession is an expression which does not specify a quantity or implies a singular number (expressions in which the article is a or an), the expression shall be construed as not being limited to a specific number. It should be.

In the specification (including the claims), expressions such as "one or more" or "at least one" are used in some places, and quantities are specified in other places. Where no or suggestive of the singular (a or an articles) are used, the latter is not intended to mean "one." In general, expressions that do not specify a quantity or imply a singular number (indicative of the articles a or an) should be construed as not necessarily being limited to a particular number.

In this specification, when it is stated that a particular configuration of an embodiment has a particular effect (advantage/result), unless there is a specific reason otherwise, other one or more having that configuration It should be understood that this effect can be obtained also for the embodiment of However, it should be understood that the presence or absence of the effect generally depends on various factors, conditions, and/or states, and that the configuration does not always provide the effect. The effect is only obtained by the configuration described in the embodiment when various factors, conditions, and/or states are satisfied, and in the claimed invention defining the configuration or a similar configuration , the effect is not necessarily obtained.

In this specification (including claims), when terms such as "optimize" are used, finding a global optimum, finding an approximation of a global optimum, finding a local optimum and approximating a local optimum, should be interpreted appropriately depending on the context in which the term is used. It also includes stochastically or heuristically approximating these optimum values.

In this specification (including claims), when a plurality of pieces of hardware perform predetermined processing, each piece of hardware may work together to perform the predetermined processing, or a part of the hardware may perform the predetermined processing. You may do all of Also, some hardware may perform a part of the predetermined processing, and another hardware may perform the rest of the predetermined processing. In the present specification (including claims), when expressions such as "one or more pieces of hardware perform the first process and the one or more pieces of hardware perform the second process" are used , the hardware that performs the first process and the hardware that performs the second process may be the same or different. In other words, the hardware that performs the first process and the hardware that performs the second process may be included in the one or more pieces of hardware. Note that hardware may include an electronic circuit or a device including an electronic circuit.

Although the embodiments of the present disclosure have been described in detail above, the present disclosure is not limited to the individual embodiments described above. Various additions, changes, replacements, partial deletions, etc. are possible without departing from the conceptual idea and spirit of the present invention derived from the content defined in the claims and equivalents thereof. For example, in all the embodiments described above, when numerical values or formulas are used for explanation, they are shown as an example and are not limited to these. Also, the order of each operation in the embodiment is shown as an example, and is not limited to these.

Claims

comprising one or more memories and one or more processors,
The one or more processors are
Get the structure graph of the first molecule,
calculating a predetermined feature quantity of the first molecule from the structure graph of the first molecule using a trained model;
The trained model is a model that has been trained using at least a quantity related to the energy of each atom that is calculated in advance for each of a plurality of atoms that constitute molecules of teacher data,
estimation device.

The calculation using the trained model includes calculating, for each atom, an energy-related quantity corresponding to a plurality of atoms forming the first molecule, and using at least the energy-related quantity calculated for each atom, the first calculating the predetermined feature of the molecule;
The estimating device according to claim 1.

3. The estimation device according to claim 1 or 2, wherein the precalculated energy-related quantity is an energy-related quantity calculated based on first-principles calculation.

comprising one or more memories and one or more processors,
The one or more processors are
Get the structure graph of the first molecule,
obtaining a first graph by processing the structure graph of the first molecule with a trained first neural network model;
calculating a predetermined feature amount of the first molecule by performing a second type of arithmetic processing based on the first graph;
The first neural network model is
a predetermined energy calculated by performing a first type of arithmetic processing based on a graph obtained by processing a structural graph of molecules of teacher data with the first neural network model;
a correct value of the calculated predetermined energy;
An estimator that has been trained at least based on

The estimation device according to claim 4, wherein the first type of arithmetic processing and the second type of arithmetic processing are different types of arithmetic processing.

The estimation device according to claim 4 or claim 5, wherein the first type of arithmetic processing includes arithmetic processing of a readout function for acquiring graph feature values.

The estimation device according to any one of claims 4 to 6, wherein the calculated correct value of the predetermined energy is calculated based on first-principles calculation.

The estimation device according to any one of claims 4 to 6, wherein the second type of arithmetic processing includes arithmetic processing using a neural network model.

The estimation device according to any one of claims 1 to 8, wherein the predetermined feature amount relates to HOMO energy or LUMO energy of a molecule.

comprising one or more memories and one or more processors,
The one or more processors are
Obtaining a second feature value related to the energy assigned to the atoms constituting the molecule from the graph containing the first feature value related to the molecule,
obtaining a third feature for one or more predetermined types of energy of the molecule based on the graph and the second feature;
estimation device.

the allocated energy is obtained based on first-principles calculations;
11. An estimating device according to claim 10.

The one or more processors are
inputting a structure graph showing the structure of the molecule into a trained model to obtain the graph;
12. An estimating device according to claim 10 or claim 11.

wherein the trained model is a model based on a graph neural network;
13. An estimating device according to claim 12.

The one or more processors are
Obtaining a feature value related to the maximum value of HOAO (Highest Occupied Atomic Orbital) energy as the second feature value;
14. The estimating device according to any one of claims 10 to 13.

The energy obtained based on the first-principles calculation includes the HOAO (Highest Occupied Atomic Orbital) energy of the atoms constituting the molecule,
The estimating device according to any one of claims 10 to 14.

The one or more processors are
obtaining the HOMO energy of the molecule as the predetermined type of energy based on the third feature;
16. The estimating device according to any one of claims 10 to 15.

The one or more processors are
further obtaining information on the LUMO (Lowest Unoccupied Molecular Orbital) energy of the molecule as the third feature quantity;
17. The estimating device according to any one of claims 10 to 16.

The energy obtained based on the first-principles calculation includes the LUAO (Lowest Unoccupied Atomic Orbital) energy of the atoms that make up the molecule,
18. An estimating device according to claim 17.

The one or more processors are
Obtaining a feature value related to the minimum value of LOAO energy as the second feature value;
19. An estimating device according to claim 18.

The energy obtained based on the first-principles calculation is the energy obtained by NBO (Natural Bond Orbital) analysis,
20. The estimating device according to any one of claims 15-19.

The energy obtained based on the first-principles calculation is the energy obtained based on the wave function,
20. The estimating device according to any one of claims 15-19.

The one or more processors are
obtaining the LUMO energy of the molecule as the predetermined type of energy based on the third feature;
11. An estimating device according to claim 10.

comprising one or more memories and one or more processors,
The one or more processors are
inputting a structure graph showing the structure of a molecule into a model to obtain a graph containing a first feature value related to the molecule;
obtaining a second feature value related to the energy assigned to the atoms constituting the molecule obtained from the graph based on first-principles calculation;
obtaining a predetermined type of energy of the molecule based on the graph and the second feature;
updating parameters of the model by comparing the obtained energy assigned to the atom and the predetermined type of energy with teacher data;
training equipment.

The one or more processors are
Updating the parameters of the model by back propagating an error between the obtained energy assigned to the atom and the predetermined type of energy and teacher data;
24. A training device according to claim 23.

the predetermined type of energy is the HOMO energy of the molecule;
The training data includes HOAO energy data,
25. A training device according to either claim 23 or claim 24.

The teacher data is data obtained by NBO analysis or data obtained based on a wave function,
26. A training device according to claim 25.

The one or more processors are
further obtaining the LUMO energy of the molecule based on the graph and the second feature;
updating parameters of the model based on the energy assigned to the atom and the HOMO energy as well as the LUMO energy;
26. A training device according to claim 25.

A trained model generated by the training device according to any one of claims 23 to 27.

by one or more processors,
inputting a structure graph showing the structure of a molecule into a model to obtain a graph containing a first feature value related to the molecule;
obtaining a second feature value related to the energy assigned to the atoms constituting the molecule obtained from the graph based on first-principles calculation;
obtaining one or more predetermined types of energy of the molecule based on the graph and the second feature;
updating parameters of the model by comparing the obtained energy assigned to the atom and the one or more predetermined types of energy with teacher data;
A method for generating the model.