Data processing method, device, model training method and free energy prediction method

Abstract

The present application relates to a data processing method, device, model training method and free energy prediction method. The data processing method comprises: obtaining data to be processed, the data to be processed comprises attribute information of each of multiple atoms in a target molecule; generating a node set and a node position set for the target molecule in response to the attribute information of each of the multiple atoms; generating a node scalar feature N for the node set; s and node vector feature N v , and based on the coordinate information of each node in the node position set, an edge scalar feature E for the node set is generated s and edge vector feature E v ; Based on the node scalar feature N for the node set s 、Node vector feature N v , edge scalar feature E s and edge vector feature E v A virtual molecular graph is constructed to determine the molecular feature X of the target molecule based on the virtual molecular graph, and the solvation free energy is determined based on the molecular feature X of the target molecule. The present application can improve the accuracy of the determined solvation free energy.

Description

Technical Field

[0001] The present application relates to the field of computer simulation technology, and in particular to a data processing method, device, model training method and free energy prediction method. Background Art

[0002] With the rapid development of computer technology and artificial intelligence technology, computer simulation technology is being applied to more and more scenarios, such as material design, drug design, etc.

[0003] However, the applicant has discovered that the molecular solvation free energy obtained by the related art has low accuracy. Summary of the invention

[0004] In order to solve or partially solve the problems existing in the related art, the present application provides a data processing method, device, model training method and free energy prediction method, which can effectively improve the accuracy of the obtained molecular solvation free energy.

[0005] The first aspect of the present application provides a data processing method, comprising: obtaining data to be processed, the data to be processed comprising attribute information of each of a plurality of atoms in a target molecule; generating a node set and a node position set for the target molecule in response to the attribute information of each of the plurality of atoms, wherein the plurality of nodes in the node set respectively represent atoms of a specific atomic type, and the node position set comprises coordinate information of each node in the node set in a specific coordinate system; generating a node scalar feature N for the node set s and node vector feature N v , and based on the coordinate information of each node in the node position set, an edge scalar feature E for the node set is generated s and edge vector feature E v ; Based on the node scalar feature N for the node set s 、Node vector feature N v , edge scalar feature E s and edge vector feature E v A virtual molecular graph is constructed to determine a molecular feature X of the target molecule based on the virtual molecular graph, so as to determine the solvation free energy based at least on the molecular feature X of the target molecule.

[0006] A second aspect of the present application provides a method for training a solvation free energy prediction model, comprising: inputting a virtual molecular graph determined based on the above method into the solvation free energy prediction model, and adjusting the model parameters to converge the loss function to obtain a trained solvation free energy prediction model, wherein the virtual molecular graph has corresponding solvation free energy annotation information, and the input of the loss function includes the predicted solvation free energy and the solvation free energy in the solvation free energy annotation information.

[0007] The third aspect of the present application provides a method for determining solvation free energy, comprising: using a trained solvation free energy prediction model to process a virtual molecular graph to obtain the solvation free energy for the virtual molecular graph, wherein the virtual molecular graph is a graph generated based on data to be processed, and the data to be processed includes attribute information for each of multiple atoms in the target molecule, and the target molecule includes solute molecules and / or solvent molecules.

[0008] The fourth aspect of the present application provides a design method, comprising: determining the solvation free energy according to the method as described above; and performing drug design or material design based on the solvation free energy.

[0009] The fifth aspect of the present application provides a data processing device, including: a data acquisition module for obtaining data to be processed, which includes attribute information of each of multiple atoms in a target molecule; a set generation module for generating a node set and a node position set for the target molecule in response to the attribute information of each of the multiple atoms, wherein the multiple nodes in the node set represent atoms of a specific atomic type respectively, and the node position set includes coordinate information of each node in the node set in a specific coordinate system; a node and edge feature generation module for generating a node scalar feature N for the node set; s and node vector feature N v , and based on the coordinate information of each node in the node position set, an edge scalar feature E for the node set is generated s and edge vector feature E v ; Virtual molecule building block for node scalar features N based on node sets s 、Node vector feature N v , edge scalar feature E s and edge vector feature E v A virtual molecular graph is constructed to determine a molecular feature X of the target molecule based on the virtual molecular graph, so as to determine the solvation free energy based at least on the molecular feature X of the target molecule.

[0010] The sixth aspect of the present application provides an apparatus for training a solvation free energy prediction model, comprising: a model training module, for inputting a virtual molecular graph determined based on the above method into the solvation free energy prediction model, and obtaining a trained solvation free energy prediction model by adjusting the model parameters so that the loss function converges, wherein the virtual molecular graph has corresponding solvation free energy annotation information, and the input of the loss function includes the predicted solvation free energy and the solvation free energy in the solvation free energy annotation information.

[0011] The seventh aspect of the present application provides a device for determining solvation free energy, including: a free energy prediction module, used to process a virtual molecular graph using a trained solvation free energy prediction model to obtain the solvation free energy for the virtual molecular graph, wherein the virtual molecular graph is a graph generated based on data to be processed, and the data to be processed includes attribute information for each of multiple atoms in the target molecule, and the target molecule includes solute molecules and / or solvent molecules.

[0012] The eighth aspect of the present application provides a design device, which includes: a solvation free energy determination module, used to determine the solvation free energy according to the above method; and a design module, used to perform drug design or material design based on the solvation free energy.

[0013] The ninth aspect of the present application provides an electronic device, comprising: a processor; and a memory on which executable codes are stored, and when the executable codes are executed by the processor, the processor executes the method.

[0014] The tenth aspect of the present application also provides a computer-readable storage medium on which executable code is stored. When the executable code is executed by a processor of an electronic device, the processor executes the above method.

[0015] The eleventh aspect of the present application also provides a computer program product, including executable code, which implements the above method when executed by a processor.

[0016] The data processing method, device, model training method and free energy prediction method provided in the present application convert the data to be processed into a node set and a node position set for the target molecule, so that a node scalar feature N for the node set can be generated. s and node vector feature N v , and based on the coordinate information of each node in the node position set, an edge scalar feature E for the node set is generated s and edge vector feature E v Compared with the low-dimensional descriptors in related technologies, these descriptors that can represent the three-dimensional characteristics of molecules can more completely represent the characteristics of the target molecules and effectively improve the accuracy of the determined solvation free energy.

[0017] In addition, describing the solvent-solute interaction by the matrix product of the solute molecular feature vector and the solvent molecular feature can better and explicitly describe the solvent-solute interaction compared to the simple concatenation, arrangement or addition of the solvent molecular features and the solute molecular features in the related art, and effectively improves the accuracy of the determined solvation free energy.

[0018] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The above and other objects, features and advantages of the present application will become more apparent by describing in more detail exemplary embodiments of the present application in conjunction with the accompanying drawings, wherein the same reference numerals generally represent the same components in the exemplary embodiments of the present application.

[0020] Figure 1 An exemplary system architecture to which a data processing method, device, model training method and free energy prediction method according to an embodiment of the present application can be applied is schematically shown;

[0021] Figure 2 The flowchart of the data processing method according to the embodiment of the present application is schematically shown;

[0022] Figure 3 A flowchart of a method for determining molecular features of a target molecule based on a virtual molecular graph according to an embodiment of the present application is schematically shown;

[0023] Figure 4 A logic diagram for updating node scalar features and node vector features based on a virtual molecular graph according to an embodiment of the present application is schematically shown;

[0024] Figure 5 A flowchart schematically shows another data processing method according to an embodiment of the present application;

[0025] Figure 6 A flowchart of a method for training a solvation free energy prediction model according to an embodiment of the present application is schematically shown;

[0026] Figure 7 The schematic diagram shows the structure of an equivariant graph convolutional network according to an embodiment of the present application;

[0027] Figure 8 A schematic diagram of the structure of a fully connected network according to an embodiment of the present application is shown;

[0028] Fig. 9 A flow chart schematically shows a method for determining solvation free energy according to an embodiment of the present application;

[0029] Fig.10 A schematic diagram showing a correlation diagram between the solvation free energy predicted by the model obtained by training the data set divided by solvent types and the actual solvation free energy on the training set according to an embodiment of the present application;

[0030] Fig.11 A schematic diagram showing a correlation diagram between the solvation free energy predicted by the model obtained by training the data set divided by solvent types and the actual solvation free energy on the test set according to an embodiment of the present application;

[0031] Fig.12 A schematic diagram showing a correlation diagram between the solvation free energy predicted by the model obtained by training the data set divided by solute types and the actual solvation free energy on the training set according to an embodiment of the present application;

[0032] Fig.13 A schematic diagram showing a correlation diagram between the solvation free energy predicted by the model obtained by training the data set divided by solute types and the actual solvation free energy on the test set according to an embodiment of the present application;

[0033] Fig.14 A flowchart of a design method according to an embodiment of the present application is schematically shown;

[0034] Fig.15 A block diagram of a data processing device according to an embodiment of the present application is schematically shown;

[0035] Fig.16 A block diagram of a device for training a solvation free energy prediction model according to an embodiment of the present application is schematically shown;

[0036] Fig.17 A block diagram of a device for determining solvation free energy according to an embodiment of the present application is schematically shown;

[0037] Fig.18 A block diagram of a design device according to an embodiment of the present application is schematically shown;

[0038] Fig.19 A block diagram of an electronic device according to an embodiment of the present application is schematically shown. DETAILED DESCRIPTION

[0039] The embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although the embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to make the present application more thorough and complete, and to fully convey the scope of the present application to those skilled in the art.

[0040] The terms used in this application are only for the purpose of describing specific embodiments and are not intended to limit this application. The terms "comprise", "include", etc. used herein indicate the existence of features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0041] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification, and should not be interpreted in an idealized or overly rigid manner.

[0042] It should be understood that although the terms "first", "second", "third", etc. may be used in this application to describe various information, this information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of this application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of this application, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0043] Before describing the technical solution of the present application, some technical terms in the field involved in the present application are first explained.

[0044] Molecular descriptors represent molecules as data structures that can be processed by computer programs.

[0045] A virtual molecular graph is a molecular descriptor that represents atoms as nodes and the relationships between atoms as edges. Unlike ordinary molecular graphs that establish edges based on the bonding information between atoms, virtual molecular graphs establish edges based on the cutoff radius.

[0046] Cutoff radius: For a certain atom in a molecule, if edges are established between this atom and all other atoms, the number of edges will be large and the amount of calculation will be too large. Considering that the farther away other atoms are from this atom, the smaller the influence on this atom, a cutoff radius is taken, and only this atom is allowed to establish edges with atoms whose distance from it is less than the cutoff radius, and the interactions with atoms outside the cutoff radius are ignored.

[0047] A coordinate system is a reference for describing the position and posture of an object, so it is also called a reference system or a reference system. For example, a coordinate system can be a coordinate system created during a simulation, such as a Cartesian coordinate system.

[0048] Coordinates are used to represent the absolute position of an object in a specific coordinate system. Mathematically, the essence of coordinates is ordered logarithms.

[0049] Solvation is a reaction process driven by the interaction between solute molecules and solvent molecules. It is a key step in processes of interest in drug development, such as crystal nucleation, chemical reactions, drug metabolism, drug interactions, and drug-receptor interactions. The strength of the solvation effect is usually characterized by the solvation free energy. Therefore, the ability to quickly and accurately predict the solvation free energy is extremely important in the field of drug development. The solvation free energy prediction process in the relevant technology can be achieved through two paths.

[0050] One is to use free energy perturbation or thermodynamic integration methods in molecular dynamics simulations based on empirical force fields fitted by experimental and computational data. In the process of practice, the applicant found that although the solvation free energy obtained by this method has a smaller error than the experimental results, the free energy perturbation or thermodynamic integration requires a long simulation time and has a high computational cost, which makes it difficult to achieve the goal of quickly predicting the solvation free energy.

[0051] The other method is to use machine learning methods to construct a structure-solvation free energy model based on existing experimental and computational data. This method can quickly predict the solvation free energy, but the applicant found in practice that in some cases the accuracy of the predicted solvation free energy is not enough to solve problems related to solvation.

[0052] After extensive research and analysis by the applicant, it was found that the reasons for the low accuracy of the prediction results using machine learning methods include the following two aspects: First, the machine learning methods in the relevant technology can represent molecules as low-dimensional descriptors such as SMILES, MACCS, Morgan, and hybrid fingerprints, which cannot fully represent the three-dimensional characteristics of molecules. Second, when describing the solvent-solute interaction, the models established by these methods can be achieved by simply splicing, arranging or adding the solvent molecule characteristics and the solute molecule characteristics, without explicitly describing the solvent-solute interaction in a physically meaningful framework.

[0053] In order to achieve fast and accurate prediction of molecular solvation free energy, the embodiments of the present application consider designing descriptors that can represent the three-dimensional characteristics of molecules and / or machine learning models that can explicitly describe solvent-solute interactions, starting from the representation of molecules and / or the description of solvent-solute interactions, so as to improve the accuracy of the predicted solvation free energy.

[0054] The following will be passed Figures 1 to 19 A data processing method, device, model training method and free energy prediction method of the embodiments of the present application are described in detail.

[0055] Figure 1An exemplary system architecture to which the data processing method, device, model training method and free energy prediction method according to an embodiment of the present application can be applied is schematically shown. It should be noted that: Figure 1 What is shown is merely an example of a system architecture to which the embodiments of the present application can be applied, in order to help those skilled in the art understand the technical content of the present application, but it does not mean that the embodiments of the present application cannot be used in other devices, systems, environments or scenarios.

[0056] See also Figure 1 According to the embodiment, the system architecture 100 may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0057] The user can use the terminal devices 101, 102, 103 to interact with other terminal devices and the server 105 through the network 104 to receive or send information, such as sending model training requests, free energy prediction requests and receiving model training results, solvation free energy, etc. The terminal devices 101, 102, 103 can be installed with various communication client applications, such as drug development applications, material design applications, web browser applications, database applications, search applications, instant messaging tools, email clients, social platform software and other applications.

[0058] The terminal devices 101, 102, 103 include but are not limited to smart desktop computers, tablet computers, laptop computers, and other electronic devices that can support functions such as surfing the Internet, modeling, analytical calculation, and design.

[0059] The server 105 can receive model training requests, solvation free energy requests, etc., adjust model parameters, store model topology, model parameters, predict solvation free energy, etc., and can also send solvation free energy to terminal devices 101, 102, 103. For example, the server 105 can be a background management server, a server cluster, etc.

[0060] It should be noted that the number of terminal devices, networks and servers is only illustrative, and any number of terminal devices, networks and clouds may be provided according to implementation requirements.

[0061] Figure 2 The flowchart of the data processing method according to the embodiment of the present application is schematically shown.

[0062] like Figure 2 As shown, this embodiment provides a method for data processing, which includes operations S210 to S240, which are specifically as follows:

[0063] In operation S210 , data to be processed is obtained, where the data to be processed includes attribute information of each of a plurality of atoms in a target molecule.

[0064] In this embodiment, the data to be processed may be a character string. The attribute information may be used to characterize the attributes of the target molecule and at least some of the atoms in the target molecule. The attributes include, but are not limited to, spatial position attributes, molecular species, atomic species, etc. The spatial position attributes may be coordinates in a Cartesian coordinate system or a polar coordinate system. The molecular species may include solute molecules and solvent molecules. The atomic species may be determined based on the number of protons and / or neutrons in the atom. For example, protium, deuterium, and tritium may be considered to be the same atomic species or different atomic species.

[0065] Specifically, the data to be processed may be a three-dimensional conformation of a solute molecule or a solvent molecule represented by a string in the format of x, y, z. For example, the string may include the three-dimensional conformation of the molecule and the x, y, z coordinates of each atom in the molecule.

[0066] For example, the x, y, z character string of a molecule is read to obtain the atomic type information and atomic position information contained in the molecule.

[0067] In addition, the same coordinate system, particularly the spatial coordinate system, can be used in the process of calculating the molecular solvation free energy and performing drug design. For example, the geometric center coordinates of the solute molecule, the atomic coordinates in the solute molecule, the geometric center coordinates of the solvent molecule, and the atomic coordinates in the solvent molecule are partly or entirely coordinates in the same coordinate system. It should be understood that the geometric center coordinates of the solute molecule, the atomic coordinates in the solute molecule, the geometric center coordinates of the solvent molecule, and the atomic coordinates in the solvent molecule may also be partly or entirely coordinates in different coordinate systems, but the coordinates in each coordinate system may be mutually converted.

[0068] In operation S220, in response to the respective attribute information of the plurality of atoms, a node set and a node position set for the target molecule are generated, wherein the plurality of nodes in the node set respectively represent atoms of a specific atomic type, and the node position set includes coordinate information of each node in the node set in a specific coordinate system.

[0069] In this embodiment, the corresponding generated node set {A i} and the node position set {(x i ,y i ,z i )},{(x i ,y i ,z i )}, where i = 1, 2, ... N, N represents the number of atoms contained in the molecule, and A represents the type of atom.

[0070] In operation S230, a node scalar feature N for a node set is generated. s and node vector feature N v , and based on the coordinate information of each node in the node position set, an edge scalar feature E for the node set is generated s and edge vector feature E v .

[0071] In this embodiment, the target molecule includes N atoms, and each of the multiple nodes in the node set has an F-dimensional feature.

[0072] Correspondingly, the node scalar feature N s The dimensions include N×F×1, node vector feature N v The dimensions include N×F×3 dimensions, edge scalar features E s The dimensions include N×1×1 dimensions, edge vector features E v The dimensions include N×3×1 dimensions.

[0073] Specifically, the feature dimension F is set, such as a value of F not less than 64. Optionally, the value of F is an integer power of 2. The elements in the node set are embedded and encoded according to the node type (such as the atomic type), and the node set is represented as an N×F×1-dimensional matrix, which represents the node scalar feature Ns. At the same time, an N×F×3-dimensional matrix with all zero elements is initialized, which represents the node vector feature N v Among them, embedding coding can refer to randomly representing a node as an F-dimensional vector, which will be updated during subsequent model training.

[0074] Next, for the node set {A i}, we can establish edges between each element i and at least some of the elements j in the remaining elements to form an edge set {(i,j)}, where i = 1, 2, ... N, j∈A i , the number of edges is recorded as E.

[0075] For each edge in the edge set, the distance and position vector between the two nodes corresponding to the edge are calculated to obtain the edge distance set, as shown in formula (1).

[0076]

[0077] The edge position vector set is shown in formula (2).

[0078] {r v，ij |r v，ij =(x i -x j ,y i -y j , zi -z j )}Formula (2)

[0079] The edge distance set is further represented as an E×1×1 matrix representing the edge scalar feature E s , the edge position vector set is represented as an E×3×1 matrix to represent the edge vector feature E v .

[0080] In some embodiments, in order to reduce the computational resources consumed when determining the molecular solvation free energy, the node set {A i} selects a set of adjacent nodes that meet the truncation radius requirements {N i}, then determine the adjacent node set {N i}'s edge set.

[0081] Specifically, the above method may also include the following operations.

[0082] First, after generating a node set and a node position set for a target molecule in response to the attribute information of each of the multiple atoms, a cutoff radius r is determined. cut . Among them, the cutoff radius r cut It can be a preset value, which can be a value determined based on expert experience or simulation results, such as 3 angstroms. wait.

[0083] Then, determine from the node set the distance between nodes that is less than or equal to the truncation radius r cut The target node is obtained by i Among them, the template node set N i That is, the adjacent node set {N i}.

[0084] Specifically, the cutoff radius r can be set first cut , for each element i in the node set, determine the adjacent node set of each element i within the truncation radius , as shown in formula (3).

[0085]

[0086] Establish edges between element i and all nodes j in the adjacent node set to form an edge set {(i,j)}, where i = 1, 2, ... N, j∈A i , the number of edges is recorded as E.

[0087] Accordingly, generating edge scalar features and edge vector features for a node set based on the coordinate information of each node in the node position set includes:

[0088] Generate the target node set N based on the coordinate information of the target node in the node position set i The edge scalar feature E s and edge vector feature E v .

[0089] In some embodiments, since the node set {A i} selects the adjacent node set {N i}, accordingly, the edge scalar feature and E s and edge vector feature E v The dimensions will also change.

[0090] Specifically, the target node set includes E nodes, each of which has F-dimensional features. s The dimensions include N×F×1, node vector feature N v The dimensions include N×F×3 dimensions, edge scalar features E s The dimensions include E×1×1 dimensions, edge vector features E v The dimensions include E×3×1 dimensions.

[0091] In operation S240, based on the node scalar feature N for the node set s 、Node vector feature N v , edge scalar feature E s and edge vector feature E v A virtual molecular graph is constructed to determine a molecular feature X of the target molecule based on the virtual molecular graph, so as to determine the solvation free energy based at least on the molecular feature X of the target molecule.

[0092] Thus, the node scalar feature N s 、Node vector feature N v , edge scalar feature E s With edge vector E v A virtual molecular graph is formed. The method for determining the molecular feature X can adopt a variety of related technologies, such as a method similar to extracting the molecular feature X based on the molecular graph.

[0093] The virtual molecular graph includes the scalar and vector information of atoms, as well as the scalar and vector information between atoms. It is a universal and accurate descriptor that can effectively improve the accuracy of the determined molecular solvation free energy.

[0094] In this embodiment, the three-dimensional information of the molecule is represented as a virtual molecular graph. In addition to not losing the three-dimensional information of the molecule, different molecular conformations can also be strictly distinguished. Compared with the two-dimensional descriptors used in the machine learning methods of the related art, the description of the molecule is more accurate.

[0095] Figure 3The flowchart of the method for determining the molecular features of a target molecule based on a virtual molecular graph according to an embodiment of the present application is schematically shown.

[0096] See also Figure 3 The process of determining the molecular features of the target molecule based on the virtual molecular graph may include operations S310 to S340.

[0097] In operation S310, the node scalar feature N is updated based on the virtual molecular graph. s and node vector feature N v , get the updated node scalar feature New_N s and update node vector feature New_N v .

[0098] In operation S320, the node scalar feature New_N is updated. s and update node vector feature New_N v As the current node scalar feature Now_N s and the current node vector feature Now_N v .

[0099] In operation S330, the current node scalar feature Now_N is used. s 、Current node vector feature Now_N v , edge scalar feature E s and edge vector feature E v Build and update the virtual molecular graph.

[0100] In operation S340, the node scalar feature New_N is updated based on the updated virtual molecule graph. s and update node vector feature New_N v .

[0101] Repeat operations S320 to S340 until the specified number of loops num_conv is reached, so that the updated node scalar feature New_N obtained when the specified number of loops num_conv is reached is converted into s , as the molecular feature X.

[0102] Specifically, the number of convolution layers num_conv can be set. For the input virtual molecular graph (solvent or solute molecule), keep E s and E v No change, use NewN s and NewN v Update N respectively s and N v , and iterate num_conv times, get NewN for the num_conv time sCompress the last dimension and convert it into an N×F matrix to represent the molecular feature X.

[0103] The following updates the node scalar feature New_N s and update node vector feature New_N v An exemplary description is given.

[0104] Figure 4 The following schematically shows a logic diagram for updating node scalar features and node vector features based on a virtual molecular graph according to an embodiment of the present application.

[0105] See also Figure 4 The update process can be composed of four basic operations: matrix linear transformation operation (Linear), activation operation (such as ReLU), matrix corresponding multiplication operation, such as Hadamard product and matrix sum operation (Sum). Among them, these four basic operations have been maturely implemented in the pytorch program framework.

[0106] Among them, the matrix linear transformation operation (Linear) transforms the input into a feature space, extracts useful information from the input and retains it.

[0107] The activation operation (such as ReLU) is a nonlinear mapping that gives the network nonlinear expression capabilities.

[0108] The matrix multiplication operation is to perform a one-to-one product of the matrix elements of two matrices of the same dimension, which plays the role of feature scaling.

[0109] The matrix addition operation (Sum) is to add the matrix elements of two matrices of the same dimension one by one, which plays the role of feature fusion.

[0110] The inner product operation (Inner) is the inner product of two vectors, converting vector information into a scalar.

[0111] Specifically, the above-mentioned node scalar feature N is updated based on the virtual molecular graph. s and node vector feature N v , get the updated node scalar feature New_N s and update node vector feature New_N v , which may include the following processes.

[0112] First, the node scalar feature N s The first linear operation, the second activation function and the second linear operation are performed in sequence to obtain the first sub-processing result Q1, and the edge scalar feature E s Perform the third linear operation to obtain the second sub-processing result Q2. The operation of obtaining the first sub-processing result Q1 realizes the extraction of N sThe useful information in E is not linearly mapped to the feature space. The operation of obtaining the second sub-processing result Q2 realizes the extraction of E s The useful information in is linearly mapped to the feature space.

[0113] Then, the first matrix corresponding multiplication operation is performed on the first sub-processing result Q1 and the second sub-processing result Q2 to obtain the third sub-processing result Q3. The operation of obtaining the third sub-processing result Q3 uses the scalar feature of the edge to scale the feature of the node, and integrates the information of the edge into the node to make the node feature more expressive.

[0114] Next, based on the third sub-processing result Q3 and the node vector feature N v Perform the corresponding multiplication operation of the second matrix to obtain the fourth sub-processing result Q4, and based on the third sub-processing result Q3 and the edge vector feature E v The corresponding multiplication operation of the third matrix is ​​performed to obtain the fifth sub-processing result Q5. Among them, the operation of obtaining the fourth sub-processing result Q4 uses the vector features of the node to scale the features of the node, and integrates the vector information into the node to make the node features more expressive. The operation of obtaining the fifth sub-processing result Q5 uses the vector features of the edge to scale the features of the node, and integrates the vector information of the edge into the node to make the node features more expressive.

[0115] Then, the first matrix addition operation is performed on the fourth sub-processing result Q4 and the fifth sub-processing result Q5 to obtain the sixth sub-processing result Q6. The operation of obtaining the sixth sub-processing result Q6 realizes the fusion of the vector features of the nodes and the edges.

[0116] Next, the sixth sub-processing result Q6 is subjected to the fourth linear operation and the fifth linear operation respectively to obtain the seventh sub-processing result Q7 and the eighth sub-processing result Q8. The seventh sub-processing result Q7 is a vector feature used to interact with the scalar feature. The eighth sub-processing result Q8 is a vector feature used to interact with the vector feature.

[0117] Then, the seventh sub-processing result Q7 is subjected to the sixth linear operation, the second activation function, and the seventh linear operation in sequence to obtain the ninth sub-processing result Q9, and the inner product operation Inner is performed based on the third sub-processing result Q3, the seventh sub-processing result Q7 and the eighth sub-processing result Q8 to obtain the tenth sub-processing result Q10; the fifth matrix corresponding multiplication operation is performed on the eighth sub-processing result Q8 and the ninth sub-processing result Q9 to obtain the updated node vector feature NewN v The operation of obtaining the ninth sub-processing result Q9 realizes the updating of the scalar feature with the vector information. The operation of obtaining the tenth sub-processing result Q10 realizes the conversion of the vector information into the scalar information. The updated node vector feature NewN is obtained. vThe operation implements updating vector features with scalar information.

[0118] Next, the fourth matrix corresponding multiplication operation is performed on the ninth sub-processing result Q9 and the tenth sub-processing result Q10 to obtain the eleventh sub-processing result Q11. The operation of obtaining the eleventh sub-processing result Q11 realizes scaling the scalar feature by the scalar information obtained by the vector inner product operation.

[0119] Then, the ninth sub-processing result Q9 and the eleventh sub-processing result Q11 are added by the second matrix to obtain the updated node scalar feature NewN s .

[0120] In this embodiment, the scalar and vector features of the edge interact with the scalar and vector features of the node to update the node features and output new scalar and vector features of the node. Specifically, the edge information is integrated into the node information to form a new feature, which improves the feature NewN. s With NewN v The ability to represent the structure makes it easier for the model to extract information related to the solvation free energy, which ultimately makes the prediction results more accurate. It should be noted that the node scalar feature New_N is updated based on the update of the virtual molecular graph. s and update node vector feature New_N v The logic and Figure 4 The logic shown is similar and will not be described in detail here.

[0121] In the manner shown above, a molecule can be represented by a virtual molecular graph as a descriptor. The virtual molecular graph includes relatively complete three-dimensional features of the molecule, which helps to improve the accuracy of the determined molecular solvation free energy.

[0122] In this embodiment, in addition to scalar features, vector features are also used in convolution, which makes it easier and more accurate to extract molecular features compared to the method of using only scalar features in the related art.

[0123] In some embodiments, the solute molecule characteristics for the solute molecules and the solvent molecule characteristics for the solvent molecules can be determined respectively based on the above-mentioned method of determining the molecular characteristics. It should be noted that there may be an interaction force between a solute molecule and a plurality of adjacent solvent molecules. The interaction force between a solute molecule and a solvent molecule can be determined first, and then the molecular solvation free energy relative to the plurality of solvent molecules can be determined. In addition, the molecular solvation free energy of a solute molecule relative to a plurality of adjacent solvent molecules can also be directly determined.

[0124] Specifically, the target molecule may be a solute molecule and / or a solvent molecule. Specifically, the data to be processed may include molecular attribute information, such as the data to be processed is data of a solute molecule and / or data of a solvent molecule.

[0125] Accordingly, the above method may further include the following operation: determining the solute molecule characteristics of the solute molecule and the solvent molecule characteristics of at least one solvent molecule associated with the solute molecule, so as to determine the solvation free energy based on the solute molecule characteristics of the solute molecule and the solvent molecule characteristics of at least one solvent molecule associated with the solute molecule. It should be noted that the characteristic dimensions of the solute molecule characteristics and the solvent molecule characteristics may be the same.

[0126] Figure 5 The following schematically shows a flow chart of another data processing method according to an embodiment of the present application.

[0127] See also Figure 5 The above method may further include operations S510 to S520.

[0128] In operation S510 , after determining solute molecule features of the solute molecules and solvent molecule features of at least one solvent molecule associated with the solute molecules, a matrix product of the solvent molecule features and the solute molecule features is used as a solvation matrix between the solvent molecules and the solute molecules.

[0129] In operation S520, a solvation characteristic is determined based on the solvation matrix.

[0130] In this embodiment, since the related art does not explicitly describe the solvent-solute interaction, the solvent-solute interaction is described by the matrix product of the solute molecular characteristics and the solvent molecular characteristics in this embodiment, thereby achieving an explicit description of the solvent-solute interaction, which helps to improve the accuracy of the determined solvation characteristics, and further improve the accuracy of the determined molecular solvation free energy.

[0131] In some embodiments, determining the solvation characteristics based on the solvation matrix may include the following operations.

[0132] First, the solvent characteristics corresponding to the preset solute weights are calculated based on the solvation matrix, and the solute characteristics corresponding to the preset solvent weights are calculated based on the solvation matrix.

[0133] Then, the solvent characteristics and the solute characteristics are respectively converted into one-dimensional row vectors including F elements.

[0134] Next, the row vectors are concatenated to obtain the solvation characteristics.

[0135] For example, first, read the solvent molecule feature X Mand solute molecular characteristics X N , X M is a matrix of dimension M×F, X N It is an N×F dimensional matrix, where M and N are the number of atoms contained in the solvent molecules and solute molecules, respectively.

[0136] Calculates the matrix product of solvent and solute molecular features

[0137] Calculate the solvent characteristic X′ under solute weight M =X MN ·X N Solute characteristics with solvent weights

[0138] According to the array element weight, X′ M and X′ N Sum the weights and convert X′ M and X′ N Convert to a one-dimensional row vector containing F elements, and finally concatenate the two row vectors into a 2F-dimensional row vector I MN is the solvation characteristic. Assume that X′ M =(1,2,3,…,F),X′ N =(1,2,3,…,F), then I MN =(1,2,3,…,F,1,2,3,…,F). The array element weights may be determined based on an attention mechanism.

[0139] In some embodiments, converting the solvent characteristics and the solute characteristics into one-dimensional row vectors including F elements may include the following operations.

[0140] First, a first element weight of an element corresponding to an atom of a solvent molecule in a solvent feature is determined, and a second element weight of an element corresponding to an atom of a solute molecule in a solute feature is determined.

[0141] Then, the solvent features are weighted summed based on the first array element weights to obtain a one-dimensional first row vector including F elements, and the solute features are weighted summed based on the second array element weights to obtain a one-dimensional second row vector including F elements.

[0142] For example, by using the attention mechanism, we can calculate X′ M and X′ N The attention coefficient of each atom in X′ is calculated according to the attention coefficient. M and X′ N Sum the weights and convert X′ M and X′ NConvert to a one-dimensional row vector containing F elements, and finally concatenate the two row vectors into a 2F-dimensional row vector representing the solvation characteristic I MN .

[0143] In certain embodiments, for an input solvation characteristic I MN , by analyzing the solvation characteristics I MN By weighted summation and biasing of the elements in the equation, the molecular solvation free energy can be obtained. MN , we can get the molecular solvation free energy.

[0144] In this embodiment, an attention mechanism is used in the solvation attention network to perform matrix product of solvent features and solute features and then perform summation, which reflects the physical meaning of solute weight and solvent weight, explicitly describes the solvation effect, and improves the accuracy of solvation free energy prediction.

[0145] Another aspect of the present application provides a method for training a solvation free energy prediction model.

[0146] In this embodiment, the method for training a solvation free energy prediction model may include: inputting a virtual molecular graph determined based on the above method into the solvation free energy prediction model, and adjusting the model parameters so that the loss function converges to obtain a trained solvation free energy prediction model, wherein the virtual molecular graph has corresponding solvation free energy annotation information, and the input of the loss function includes the predicted solvation free energy and the solvation free energy in the solvation free energy annotation information.

[0147] In certain embodiments, the solvation free energy prediction model may include at least one of the following networks.

[0148] The molecular encoding network is configured to convert each training data in a training data set including solute molecule data and / or solvent molecule data into a virtual molecular graph for the solute molecule data and / or for the solvent molecule data, respectively, wherein the training data has solvation free energy annotation information.

[0149] An equivariant graph convolutional network is configured to convert the virtual molecular graph into solute molecular features and / or solvent molecular features.

[0150] A solvation network is configured to convert solute molecule features and solvent molecule features into solvation features.

[0151] Fully connected network configured to convert solvation features into solvation free energies.

[0152] Accordingly, the above training method may include: inputting training data into a molecular encoding network, and adjusting model parameters (such as network parameters) to converge the loss function, wherein the input of the loss function includes the solvation free energy output by the fully connected network and the solvation free energy in the solvation free energy annotation information.

[0153] In some embodiments, the solvation network includes a self-attention network, which is configured to determine a first element weight of an element corresponding to an atom of a solvent molecule in a solvent feature, and to determine a second element weight of an element corresponding to an atom of a solute molecule in a solute feature, so as to fuse the elements corresponding to each atom of a solvent molecule in the solvent feature according to the first element weight and fuse the elements corresponding to each atom of a solute molecule in the solute feature according to the second element weight, wherein the solvent feature and the solute feature are determined based on a solvation matrix, and the solvation matrix is ​​determined based on solute molecule features and solvent molecule features. For details, please refer to the relevant part of the data processing method, which will not be described in detail here.

[0154] In some embodiments, the above method may further include the following operations.

[0155] First, the training data set is divided into sub-training data sets of a specified number of portions. For example, the specified number of portions can be determined based on expert experience or the accuracy of predicting the solvation free energy of molecules. For example, the specified number of portions can be 3 portions, 5 portions, 8 portions, 10 portions, 13 portions, 18 portions, 20 portions, etc.

[0156] Then, the same number of solvation free energy prediction models as the specified number are constructed. In this way, multiple solvation free energy prediction models can be trained, and a model with good accuracy in predicting solvation free energy can be selected from them, or the average of the output results of multiple models can be used as the final prediction result.

[0157] Correspondingly, inputting the training data into the molecular coding network includes: inputting the training data in each sub-training data set into the molecular coding network of different solvation free energy prediction models respectively, so as to perform model training on different solvation free energy prediction models respectively, and obtain multiple trained solvation free energy prediction models of the same number as the specified number.

[0158] Figure 6 The flowchart of the method for training a solvation free energy prediction model according to an embodiment of the present application is schematically shown.

[0159] See also Figure 6, collect several data of the real value of the solvation free energy of the solute in the solvent obtained by experimental measurement or theoretical calculation. For each data, store the three-dimensional conformation of the solute molecule and the solvent molecule in the data set as a string in the format of x, y, z (for example, store the three-dimensional conformation of the molecule as the x, y, z coordinates of each atom), and store the corresponding solvation free energy in the same data set as a floating point number.

[0160] Initialize the solvation free energy prediction model (such as the equivariant graph neural network model). The model consists of four parts: a molecular encoding network that converts the molecular x, y, z string into a virtual molecular graph. An equivariant graph convolutional network that converts the virtual molecular graph into molecular features. A solvation attention network that converts the solute molecule and solvent molecule features into solvation features by matrix product and attention aggregation (after matrix product of solute molecules and solvent molecules, the atomic-level features in the molecule are obtained, and attention aggregation is to aggregate these atomic-level features into molecular-level features through the attention mechanism), and a fully connected network that converts the solvation features into solvation free energy.

[0161] Set the loss function (such as mean square error loss function, absolute difference loss function, Huber loss function, etc.). Divide the data set into ten equal parts and train the model using a ten-fold cross validation method. Until the loss function of the validation set no longer decreases (i.e., the loss function converges, and the difference between the two loss functions is less than the preset value, which can be 0.0005, 0.001, 0.002, etc., i.e., converges), ten equivariant graph neural network models are obtained. It should be noted that a five-fold cross validation method or a k-fold cross validation method can also be used.

[0162] The calculation method of the loss function L is shown in formula (4).

[0163]

[0164] Among them, G i,pred is the predicted value of the solvation free energy, G i,true is the true value of the solvation free energy, and n is the number of solvent-solute pairs used in training.

[0165] The topological structure of the solvation free energy prediction model is exemplified below.

[0166] Figure 7 The structural diagram of the equivariant graph convolutional network according to an embodiment of the present application is schematically shown.

[0167] See also Figure 7 , the equivariant graph convolutional network includes convolutional layers with a specified number of cycles num_conv layers, where the output of the current convolutional layer serves as part of the input of the next adjacent convolutional layer.

[0168] The input of the first convolutional layer (see the first convolutional layer) includes: node scalar features N s 、Node vector feature N v , edge scalar feature E s and edge vector feature E v The output of the first convolutional layer includes: Update node scalar feature New_N s and update node vector feature New_N v .

[0169] The inputs of convolutional layers other than the first convolutional layer (see the second convolutional layer, the third convolutional layer, the fourth convolutional layer, etc.) include: Update node scalar feature New_N s and update node vector feature New_N v , edge scalar feature E s and edge vector feature E v The outputs of the convolutional layers other than the first convolutional layer include: Update node scalar feature New_N s and update node vector feature New_N v .

[0170] Atomic features can be converted into molecular features through equivariant graph convolutional networks.

[0171] Specifically, each convolutional layer can achieve feature conversion in the following way.

[0172] Please also see Figure 7 and Figure 4 , the equivariant graph convolutional network is composed of four basic operations: Linear, ReLU, Hadamard and Sum. Among them, Linear is a matrix linear transformation operation, ReLU is an activation operation, Hadamard is a matrix corresponding multiplication operation, and Sum is a matrix addition operation.

[0173] Specifically, the convolution layer is configured to perform the following operations. It should be noted that the first linear operation can be implemented by a first linear layer, and the second linear operation can be implemented by a second linear layer. The first linear layer and the second linear layer can be the same layer or different layers.

[0174] Scalar feature N for nodes s The first linear operation, the second activation function and the second linear operation are performed in sequence to obtain the first sub-processing result Q1, and the edge scalar feature E s A third linear operation is performed to obtain a second sub-processing result Q2.

[0175] A first matrix corresponding multiplication operation is performed on the first sub-processing result Q1 and the second sub-processing result Q2 to obtain a third sub-processing result Q3.

[0176] Based on the third sub-processing result Q3 and the node vector feature N v Perform the corresponding multiplication operation of the second matrix to obtain the fourth sub-processing result Q4, and based on the third sub-processing result Q3 and the edge vector feature E v Perform a multiplication operation corresponding to the third matrix to obtain a fifth sub-processing result Q5.

[0177] The first matrix addition operation is performed on the fourth sub-processing result Q4 and the fifth sub-processing result Q5 to obtain a sixth sub-processing result Q6.

[0178] The sixth sub-processing result Q6 is subjected to the fourth linear operation and the fifth linear operation respectively to obtain a seventh sub-processing result Q7 and an eighth sub-processing result Q8.

[0179] The seventh sub-processing result Q7 is sequentially subjected to the sixth linear operation, the second activation function, and the seventh linear operation to obtain a ninth sub-processing result Q9, and an inner product operation Inner is performed based on the third sub-processing result Q3, the seventh sub-processing result Q7, and the eighth sub-processing result Q8 to obtain a tenth sub-processing result Q10.

[0180] The corresponding multiplication operation of the fourth matrix is ​​performed on the ninth sub-processing result Q9 and the tenth sub-processing result Q10 to obtain an eleventh sub-processing result Q11.

[0181] Perform the second matrix addition on the ninth sub-processing result Q9 and the eleventh sub-processing result Q11 to obtain the updated node scalar feature NewN s ;

[0182] The fifth matrix corresponding multiplication operation is performed on the eighth sub-processing result Q8 and the ninth sub-processing result Q9 to obtain the updated node vector feature NewN v .

[0183] The functions and effects of the above operations can be referred to the relevant contents in the above embodiments, which will not be described in detail here.

[0184] In some embodiments, atoms of solute molecules or solvent molecules in the training data each have F-dimensional features.

[0185] Figure 8 The schematic diagram shows the structure of a fully connected network according to an embodiment of the present application.

[0186] See also Figure 8 , the fully connected network may include: a first linear layer (such as Linear), a first activation function layer (such as ReLU), a second linear layer, a second activation function layer, and a third linear layer connected in sequence, wherein the output dimension of the first linear layer and the second linear layer is F-dimensional, and the output dimension of the third linear layer is 1-dimensional. The input of the first linear layer is a 2F-dimensional row vector, representing the solvation feature IMN For the input solvation feature I MN , which can be converted into molecular solvation free energy through a fully connected network.

[0187] Another aspect of the present application provides a method for determining solvation free energy.

[0188] In this embodiment, the method for determining the solvation free energy may include the following operations: processing a virtual molecular graph using a solvation free energy prediction model trained according to the above method to obtain the solvation free energy for the virtual molecular graph, wherein the virtual molecular graph is a graph generated based on data to be processed, and the data to be processed includes attribute information for each of multiple atoms in the target molecule, and the target molecule includes solute molecules and / or solvent molecules.

[0189] In certain embodiments, the solvation free energy prediction model may include at least one of the following networks.

[0190] The molecular encoding network is configured to convert each training data in a training data set including solute molecule data and / or solvent molecule data into a virtual molecular graph for the solute molecule data and / or for the solvent molecule data, respectively, wherein the training data has solvation free energy annotation information.

[0191] An equivariant graph convolutional network is configured to convert the virtual molecular graph into solute molecular features and / or solvent molecular features.

[0192] A solvation network is configured to convert solute molecule features and solvent molecule features into solvation features.

[0193] Fully connected network configured to convert solvation features into solvation free energies.

[0194] Accordingly, the above method may include the following operations: using a trained solvation free energy prediction model to process the data to be processed to obtain the solvation free energy for the data to be processed, wherein the data to be processed includes property information for each of multiple atoms in the target molecule, and the target molecule includes solute molecules and / or solvent molecules.

[0195] Fig. 9 The flowchart of the method for determining the solvation free energy according to an embodiment of the present application is schematically shown.

[0196] See also Fig. 9 , network parameters can be input into the solvation free energy prediction model so as to process the solvent conformation and the solute conformation using the trained neural network. Specifically, the solvent conformation (such as can be represented as an xyz string for the solvent molecule) and the solute conformation (such as can be represented as an xyz string for the solute molecule) can be used as inputs of the molecular encoding network.

[0197] In some embodiments, the above method may further include the following operations.

[0198] First, the virtual molecular graph or the data to be processed are respectively input into different trained solvation free energy prediction models of a specified number to obtain the solvation free energy of the specified number, wherein the specified number can be the number of trained solvation free energy prediction models.

[0199] Then, the weighted average of the specified number of solvation free energies is taken as the solvation free energy corresponding to the data to be processed.

[0200] For example, the solvent molecules and solute molecules to be predicted are input into ten models in the format of x, y, z respectively, and ten predicted values ​​of solvation free energy are obtained, and their average is taken as the final prediction result.

[0201] In a specific embodiment, first, 48,776 molecular conformations of 11,940 molecules are collected (for example, molecular conformations can be collected through online databases such as pubchem), and 15 (only examples, there may be more or less than 15) molecules of water, tetrahydrofuran, chloroform, dichloromethane, dioxane, toluene, methanol, acetone, n-heptane, cyclohexane, ether, acetonitrile, dimethylformamide, dimethyl sulfoxide and methyl tert-butyl ether with only a single molecular conformation (selecting a single conformation molecule here is just for the convenience of calculation. Molecules with multiple conformations can also be selected, but the results of calculations of different conformations need to be statistically averaged) are selected as solvents, and COSMOtherm is used to calculate 731,640 pieces of solvation free energy data of 48,776 conformations in 15 solvents. The 48,776 conformations are stored in the data set in the format of x, y, z, and the 731,640 pieces of solvation free energy data corresponding to the solute conformation and the solvent conformation are stored in the data set as floating point numbers. 48776 systems with water as solvent were selected as the test set, and the other 682864 systems were selected as the training set.

[0202] Then, set the feature dimension F to 128 and the cutoff radius rcut to , the number of convolutional layers num_conv is 3, and the equivariant graph neural network model is initialized.

[0203] Next, the loss function is set to mean square error. The 682864 systems in the training set are divided into ten parts, different random number seeds are set, and the model is trained using a ten-fold cross-validation method until the loss function of the validation set no longer decreases. Ten equivariant graph neural network models are obtained, and the network parameters of the ten models are output.

[0204] Then, the solvent molecules and solute molecules corresponding to the 48776 systems in the test set to be predicted are input into the above ten models in the format of x, y, z in turn, and ten predicted values ​​of solvation free energy are obtained, and their average is taken as the final prediction result. For comparison, the solvent molecules and solute molecules in the training set are also input into the above ten models in the format of x, y, z in turn, and ten predicted values ​​of solvation free energy are obtained, and their average is taken as the final prediction result of the training set. The correlation between the true value of the training set and the test set and the model prediction value is shown in the figure below. Fig.10 , Fig.11 As shown. Among them, MAE is the mean absolute error, RMSE is the root mean square error, R 2 is the coefficient of determination. The smaller the MAE and RMSE, the smaller the model error. 2 is a value between 0 and 1, R 2 The larger the value, the better the model relevance. It can be seen that the relevance of the model on the test set and the training set is basically the same, and the average absolute error of the prediction is less than 1 kJ / mol, which is much lower than the error of the traditional machine learning method. The results are shown in Table 1.

[0205] Table 1

[0206] Model CIGIN Delfos MPNN This application MAE / (kJ / mol) 3.17 4.97 4.81 0.77

[0207] In another specific embodiment, first, 48,776 conformations of 11,940 molecules are collected, and 15 molecules with only a single conformation, including water, tetrahydrofuran, chloroform, dichloromethane, dioxane, toluene, methanol, acetone, n-heptane, cyclohexane, ether, acetonitrile, dimethylformamide, dimethyl sulfoxide and methyl tert-butyl ether, are selected as solvents. COSMOtherm is used to calculate 731,640 pieces of solvation free energy data of 48,776 conformations in 15 solvents. The 48,776 conformations are stored in the data set in x, y, z format, and the 731,640 pieces of solvation free energy data corresponding to the solute conformation and the solvent conformation are stored in the data set as floating point numbers. 41,475 pieces of solute-solvent-solvation free energy data of 2,765 conformational systems of 740 molecules are selected as the test set, and the other 690,165 systems are used as the training set.

[0208] Then, set the feature dimension F to 128 and the truncation radius r cut for , the number of convolutional layers num_conv is 3, and the equivariant graph neural network model is initialized.

[0209] Next, the loss function is set to mean square error. The training set of 690165 systems is divided into ten parts, different random number seeds are set, and the model is trained using a ten-fold cross-validation method until the loss function of the validation set no longer decreases. Ten equivariant graph neural network models are obtained, and the network parameters of the ten models are output.

[0210] Then, the solvent and solute molecules of the 41,475 systems in the test set to be predicted are input into ten models in the format of x, y, z, and ten predicted values ​​of solvation free energy are obtained. Their average is taken as the final prediction result. For comparison, the solvent and solute molecules of the training set are also input into ten models in the format of x, y, z, and ten predicted values ​​of solvation free energy are obtained. Their average is taken as the final prediction result of the training set. The correlation between the true value of the training set and the test set and the predicted value of the model is shown in the figure below. Fig.12 , Fig.13 As shown in the figure, it can be seen that the correlation between the model on the test set and the training set is basically consistent, and the average absolute error of the prediction is less than 1 kJ / mol.

[0211] In this embodiment, in view of the defects and deficiencies of predicting the solvation free energy of molecules in the related art, a method based on an equivariant graph neural network is proposed to predict the solvation free energy. Among them, in view of the problem that the related art cannot fully represent the three-dimensional characteristics of molecules, this embodiment uses a virtual molecular graph as a descriptor to represent molecules. In view of the fact that the related art does not explicitly describe the solvent-solute interaction, this embodiment describes the solvent-solute interaction by the matrix product of the solute molecule feature vector and the solvent molecule feature vector. Specifically, it consists of four steps: molecular encoding, equivariant graph convolution, feature interaction and free energy prediction. The molecular encoding step represents the solvent and solute molecules as a virtual molecular graph with feature encoding. The equivariant graph convolution step converts the virtual molecular graph into a feature representation in the form of a matrix. The feature interaction step performs a matrix product on the feature representation of the solvent and the solute to obtain a feature representation of the solvation effect. The free energy prediction step predicts the molecular solvation free energy through a fully connected neural network based on the feature representation of the solvation effect, which effectively improves the accuracy of the predicted molecular solvation free energy.

[0212] Another aspect of the present application provides a design method.

[0213] Fig.14 The flowchart of the design method according to the embodiment of the present application is schematically shown.

[0214] See also Fig.14 , the design method may include operation S1410 and operation S1420.

[0215] In operation S1410, the solvation free energy is determined according to the method described above.

[0216] In operation S1420, drug design or material design is performed based on the solvation free energy.

[0217] Another aspect of the present application provides a data processing device.

[0218] Fig.15 A block diagram of a data processing device according to an embodiment of the present application is schematically shown.

[0219] See also Fig.15 The data processing device may include: a to-be-processed data acquisition module 1510 , a set generation module 1520 , a node and edge feature generation module 1530 and a virtual molecule construction module 1540 .

[0220] The module 1510 for obtaining data to be processed is used to obtain data to be processed, where the data to be processed includes property information of each of multiple atoms in the target molecule.

[0221] The set generation module 1520 is used to generate a node set and a node position set for the target molecule in response to the respective attribute information of multiple atoms, wherein the multiple nodes in the node set respectively represent atoms of a specific atomic type, and the node position set includes the coordinate information of each node in the node set in a specific coordinate system.

[0222] The node and edge feature generation module 1530 is used to generate node scalar features N for a node set. s and node vector feature N v , and based on the coordinate information of each node in the node position set, an edge scalar feature E for the node set is generated s and edge vector feature E v .

[0223] The virtual molecule construction module 1540 is used to construct a virtual molecule based on the node scalar feature N of the node set. s 、Node vector feature N v , edge scalar feature E s and edge vector feature E v A virtual molecular graph is constructed to determine a molecular feature X of the target molecule based on the virtual molecular graph, so as to determine the solvation free energy based at least on the molecular feature X of the target molecule.

[0224] In some embodiments, the target molecule includes N atoms, and each of the plurality of nodes in the node set has an F-dimensional feature.

[0225] Node scalar feature N s The dimensions include N×F×1, node vector feature N v The dimensions include N×F×3 dimensions, edge scalar features E s The dimensions include N×1×1 dimensions, edge vector features Ev The dimensions include N×3×1 dimensions.

[0226] In some embodiments, the apparatus 1500 may further include: a cutoff radius determination module and a target node set determination module.

[0227] The cutoff radius determination module is configured to determine the cutoff radius r after generating a node set and a node position set for the target molecule in response to the attribute information of each of the plurality of atoms. cu t.

[0228] The target node set determination module is configured to determine from the node set the distance between nodes that is less than or equal to the truncation radius r cut The target node is obtained by i .

[0229] Accordingly, the node and edge feature generation module 1530 is specifically configured to generate a node for the target node set N based on the coordinate information for the target node in the node position set. i The edge scalar feature E s and edge vector feature E v .

[0230] In some embodiments, the target node set includes E nodes, each of the E nodes has F-dimensional features.

[0231] Node scalar feature N s The dimensions include N×F×1, node vector feature N v The dimensions include N×F×3 dimensions, edge scalar features E s The dimensions include E×1×1 dimensions, edge vector features E v The dimensions include E×3×1 dimensions.

[0232] In some embodiments, the apparatus 1500 further includes a feature updating module and a circulation module.

[0233] The feature updating module is configured to update the node scalar feature N based on the virtual molecular graph s and node vector feature N v , get the updated node scalar feature New_N s and update node vector feature New_N v .

[0234] The loop module is configured to repeat the following units until the specified number of loops num_conv is reached, so as to update the node scalar feature New_N obtained when the specified number of loops num_conv is reached. s , as the molecular feature X.

[0235] The feature replacement unit is configured to update the node scalar feature New_N s and update node vector feature New_N v As the current node scalar feature Now_N s and the current node vector feature Now_N v .

[0236] The feature calculation unit is configured to use the current node scalar feature Now_N s 、Current node vector feature Now_N v , edge scalar feature E s and edge vector feature E v Build and update the virtual molecular graph.

[0237] A feature updating unit is configured to update the node scalar feature New_N based on the updated virtual molecular graph s and update node vector feature New_N v .

[0238] In some embodiments, the feature update module is specifically configured to perform the following operations.

[0239] Scalar feature N for nodes s The first linear operation, the second activation function and the second linear operation are performed in sequence to obtain the first sub-processing result Q1, and the edge scalar feature E s A third linear operation is performed to obtain a second sub-processing result Q2.

[0240] A first matrix corresponding multiplication operation is performed on the first sub-processing result Q1 and the second sub-processing result Q2 to obtain a third sub-processing result Q3.

[0241] Based on the third sub-processing result Q3 and the node vector feature N v Perform the corresponding multiplication operation of the second matrix to obtain the fourth sub-processing result Q4, and based on the third sub-processing result Q3 and the edge vector feature E v Perform a multiplication operation corresponding to the third matrix to obtain a fifth sub-processing result Q5.

[0242] The first matrix addition operation is performed on the fourth sub-processing result Q4 and the fifth sub-processing result Q5 to obtain a sixth sub-processing result Q6.

[0243] The sixth sub-processing result Q6 is subjected to the fourth linear operation and the fifth linear operation respectively to obtain a seventh sub-processing result Q7 and an eighth sub-processing result Q8.

[0244] The seventh sub-processing result Q7 is sequentially subjected to the sixth linear operation, the second activation function, and the seventh linear operation to obtain a ninth sub-processing result Q9, and an inner product operation Inner is performed based on the third sub-processing result Q3, the seventh sub-processing result Q7, and the eighth sub-processing result Q8 to obtain a tenth sub-processing result Q10.

[0245] The corresponding multiplication operation of the fourth matrix is ​​performed on the ninth sub-processing result Q9 and the tenth sub-processing result Q10 to obtain an eleventh sub-processing result Q11.

[0246] The fifth matrix corresponding multiplication operation is performed on the eighth sub-processing result Q8 and the ninth sub-processing result Q9 to obtain the updated node vector feature NewN v .

[0247] Perform the second matrix addition on the ninth sub-processing result Q9 and the eleventh sub-processing result Q11 to obtain the updated node scalar feature NewN s .

[0248] The functions and effects of the above operations can be referred to the relevant contents in the above embodiments, which will not be described in detail here.

[0249] In certain embodiments, the target molecules are solute molecules and / or solvent molecules.

[0250] The above-mentioned device 1500 also includes: a solute solvent molecule characteristic determination module, which is configured to determine the solute molecule characteristics of the solute molecule and the solvent molecule characteristics of at least one solvent molecule associated with the solute molecule, so as to determine the solvation free energy based on the solute molecule characteristics of the solute molecule and the solvent molecule characteristics of at least one solvent molecule associated with the solute molecule.

[0251] In certain embodiments, the apparatus 1500 further comprises: a solvation matrix determination module and a solvation feature determination module.

[0252] The solvation interaction matrix determination module is configured to, after determining the solute molecule characteristics of the solute molecules and the solvent molecule characteristics of at least one solvent molecule associated with the solute molecules, use the matrix product of the solvent molecule characteristics and the solute molecule characteristics as the solvation interaction matrix between the solvent molecules and the solute molecules.

[0253] The solvation signature determination module is configured to determine a solvation signature based on the solvation matrix.

[0254] In certain embodiments, the solvation characteristic determination module includes: a solvent characteristic determination unit, a solute characteristic determination unit, and a solvation characteristic determination unit.

[0255] The solvent characteristic determination unit is configured to calculate the solvent characteristic corresponding to the preset solute weight based on the solvation matrix, and calculate the solute characteristic corresponding to the preset solvent weight based on the solvation matrix.

[0256] The solute characteristic determination unit is configured to convert the solvent characteristic and the solute characteristic into a one-dimensional row vector including F elements, respectively.

[0257] The solvation feature determination unit is configured to concatenate row vectors to obtain a solvation feature.

[0258] In some embodiments, the solute characteristic determination unit includes an array element weight determination subunit and a weighted summation subunit.

[0259] The array element weight determination subunit is configured to determine first array element weights of array elements corresponding to atoms of solvent molecules in the solvent feature, and determine second array element weights of array elements corresponding to atoms of solute molecules in the solute feature.

[0260] The weighted summation subunit is configured to perform weighted summation on the solvent features based on the first array element weight to obtain a one-dimensional first row vector including F elements, and to perform weighted summation on the solute features based on the second array element weight to obtain a one-dimensional second row vector including F elements.

[0261] Another aspect of the present application also provides a device for training a solvation free energy prediction model.

[0262] Fig.16 A block diagram of an apparatus for training a solvation free energy prediction model according to an embodiment of the present application is schematically shown.

[0263] The above-mentioned device 1600 includes: a model training module 1610, which is used to input the virtual molecular graph determined based on the above-mentioned method into the solvation free energy prediction model, and obtain a trained solvation free energy prediction model by adjusting the model parameters so that the loss function converges, wherein the virtual molecular graph has corresponding solvation free energy annotation information, and the input of the loss function includes the predicted solvation free energy and the solvation free energy in the solvation free energy annotation information.

[0264] In certain embodiments, the solvation free energy prediction model comprises: an isovariant graph convolutional network configured to convert a virtual molecular graph into solute molecular features and / or solvent molecular features.

[0265] The equivariant graph convolution network includes a convolutional layer with a specified number of cycles num_conv, where the output of the current convolutional layer is used as part of the input of the next convolutional layer; the input of the first convolutional layer includes: node scalar feature N s 、Node vector feature N v, edge scalar feature E s and edge vector feature E v , the output of the first convolutional layer includes: updating the node scalar feature New_N s and update node vector feature New_N v ; The input of the convolutional layer other than the first convolutional layer includes: Update node scalar feature New_N s and update node vector feature New_N v , edge scalar feature E s and edge vector feature E v ; The output of the convolutional layer other than the first convolutional layer includes: updating the node scalar feature New_N s and update node vector feature New_N v .

[0266] In some embodiments, the convolutional layer is configured to perform the following operations.

[0267] Scalar feature N for nodes s The first linear operation, the second activation function and the second linear operation are performed in sequence to obtain the first sub-processing result Q1, and the edge scalar feature E s A third linear operation is performed to obtain a second sub-processing result Q2.

[0268] A first matrix corresponding multiplication operation is performed on the first sub-processing result Q1 and the second sub-processing result Q2 to obtain a third sub-processing result Q3.

[0269] Based on the third sub-processing result Q3 and the node vector feature N v Perform the corresponding multiplication operation of the second matrix to obtain the fourth sub-processing result Q4, and based on the third sub-processing result Q3 and the edge vector feature E v Perform a multiplication operation corresponding to the third matrix to obtain a fifth sub-processing result Q5.

[0270] The first matrix addition operation is performed on the fourth sub-processing result Q4 and the fifth sub-processing result Q5 to obtain a sixth sub-processing result Q6.

[0271] The sixth sub-processing result Q6 is subjected to the fourth linear operation and the fifth linear operation respectively to obtain a seventh sub-processing result Q7 and an eighth sub-processing result Q8.

[0272] The seventh sub-processing result Q7 is sequentially subjected to the sixth linear operation, the second activation function, and the seventh linear operation to obtain a ninth sub-processing result Q9, and an inner product operation Inner is performed based on the third sub-processing result Q3, the seventh sub-processing result Q7, and the eighth sub-processing result Q8 to obtain a tenth sub-processing result Q10.

[0273] The corresponding multiplication operation of the fourth matrix is ​​performed on the ninth sub-processing result Q9 and the tenth sub-processing result Q10 to obtain an eleventh sub-processing result Q11.

[0274] Perform the second matrix addition on the ninth sub-processing result Q9 and the eleventh sub-processing result Q11 to obtain the updated node scalar feature NewN s .

[0275] The fifth matrix corresponding multiplication operation is performed on the eighth sub-processing result Q8 and the ninth sub-processing result Q9 to obtain the updated node vector feature NewN v .

[0276] The functions and effects of the above operations can be referred to the relevant contents in the above embodiments, which will not be described in detail here.

[0277] In certain embodiments, the solvation free energy prediction model comprises: a molecular encoding network.

[0278] The molecular encoding network is configured to convert each training data in a training data set including solute molecule data and / or solvent molecule data into a virtual molecular graph for the solute molecule data and / or for the solvent molecule data, respectively, wherein the training data has solvation free energy annotation information, and wherein the atoms of the solute molecules or solvent molecules in the training data respectively have F-dimensional characteristics.

[0279] In certain embodiments, the solvation free energy prediction model comprises: a solvation network.

[0280] The solvation network is configured to convert solute molecule features and solvent molecule features into solvation features.

[0281] For example, the solvation network includes a self-attention network, which is configured to determine a first element weight of an element corresponding to an atom of a solvent molecule in a solvent feature, and to determine a second element weight of an element corresponding to an atom of a solute molecule in a solute feature, so as to fuse the elements corresponding to each atom of the solvent molecule in the solvent feature according to the first element weight and to fuse the elements corresponding to each atom of the solute molecule in the solute feature according to the second element weight, wherein the solvent feature and the solute feature are determined based on a solvation matrix, and the solvation matrix is ​​determined based on the solute molecule feature and the solvent molecule feature.

[0282] In certain embodiments, the solvation free energy prediction model includes: a fully connected network configured to convert solvation characteristics into solvation free energy.

[0283] The fully connected network includes: a first linear layer, a first activation function layer, a second linear layer, a second activation function layer and a third linear layer connected in sequence, wherein the output dimension of the first linear layer and the second linear layer is F dimension, and the output dimension of the third linear layer is 1 dimension.

[0284] In some embodiments, the apparatus 1600 further includes: a training set segmentation module and a model building module.

[0285] The training set segmentation module is configured to segment the training data set into a specified number of sub-training data sets.

[0286] The model building module is configured to build the same number of solvation free energy prediction models as the specified number of copies.

[0287] The model training module 1610 is specifically configured to input the training data in each sub-training data set into the molecular encoding network of different solvation free energy prediction models, so as to perform model training on different solvation free energy prediction models respectively, and obtain multiple trained solvation free energy prediction models of the same number as the specified number.

[0288] Another aspect of the present application provides a device for determining solvation free energy.

[0289] Fig.17 A block diagram of an apparatus for determining solvation free energy according to an embodiment of the present application is schematically shown.

[0290] The above-mentioned device 1700 includes: a free energy prediction module 1710, which is used to process the data to be processed using a trained solvation free energy prediction model to obtain the solvation free energy for the data to be processed, wherein the data to be processed includes the property information of each of multiple atoms in the target molecule, and the target molecule includes solute molecules and / or solvent molecules.

[0291] For example, the solvation free energy prediction model includes at least one of the following networks: a molecular encoding network, configured to convert each training data in a training data set including solute molecule data and / or solvent molecule data into a virtual molecular graph for the solute molecule data and / or for the solvent molecule data, respectively, wherein the training data has solvation free energy annotation information; an equivariant graph convolutional network, configured to convert the virtual molecular graph into solute molecule features and / or solvent molecule features; a solvation network, configured to convert solute molecule features and solvent molecule features into solvation features; and a fully connected network, configured to convert solvation features into solvation free energy.

[0292] In some embodiments, the apparatus 1700 further includes: a multi-model processing module and a weighted processing module.

[0293] The multi-model processing module is configured to input the data to be processed into different trained solvation free energy prediction models of a specified number, respectively, to obtain the solvation free energies of a specified number;

[0294] The weighted processing module is configured to use the weighted average of the specified number of solvation free energies as the solvation free energy corresponding to the data to be processed. It should be noted that the weights of the specified number of solvation free energies can be the same or different. For example, the weight of the solvation free energy obtained by the model with high prediction accuracy on the test data set can be higher than the weight of the solvation free energy obtained by other models.

[0295] Another aspect of the present application also provides a design device.

[0296] Fig.18 A block diagram of a design device according to an embodiment of the present application is schematically shown.

[0297] See also Fig.18 , the device 1800 may include: a solvation free energy determination module 1810 and a design module 1820 .

[0298] The solvation free energy determination module 1810 is configured to determine the solvation free energy according to the above method.

[0299] The design module 1820 is used for drug design or material design based on solvation free energy.

[0300] Regarding the devices 1500, 1600, 1700, and 1800 in the above embodiments, the specific manner in which each module and unit performs operations has been described in detail in the embodiment of the method, and will not be elaborated again here.

[0301] Another aspect of the present application provides an electronic device.

[0302] Fig.19 A block diagram of an electronic device implementing an embodiment of the present application is schematically shown.

[0303] See also Fig.19 , the electronic device 1900 includes a memory 1910 and a processor 1920 .

[0304] The processor 1920 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc.

[0305] The memory 1910 may include various types of storage units, such as system memory, read-only memory (ROM), and permanent storage devices. Among them, ROM can store static data or instructions required by the processor 1920 or other modules of the computer. The permanent storage device may be a readable and writable storage device. The permanent storage device may be a non-volatile storage device that does not lose the stored instructions and data even after the computer is powered off. In some embodiments, the permanent storage device uses a large-capacity storage device (such as a magnetic or optical disk, flash memory) as a permanent storage device. In some other embodiments, the permanent storage device may be a removable storage device (such as a floppy disk, optical drive). The system memory may be a readable and writable storage device or a volatile readable and writable storage device, such as a dynamic random access memory. The system memory may store some or all instructions and data required by the processor at run time. In addition, the memory 1910 may include any combination of computer-readable storage media, including various types of semiconductor memory chips (such as DRAM, SRAM, SDRAM, flash memory, programmable read-only memory), and disks and / or optical disks may also be used. In some embodiments, the memory 1910 may include a readable and / or writable removable storage device, such as a laser disc (CD), a read-only digital versatile disc (such as a DVD-ROM, a double-layer DVD-ROM), a read-only Blu-ray disc, an ultra-density optical disc, a flash memory card (such as an SD card, a mini SD card, a Micro-SD card, etc.), a magnetic floppy disk, etc. The computer-readable storage medium does not include carrier waves and transient electronic signals transmitted wirelessly or wired.

[0306] The memory 1910 stores executable codes, and when the executable codes are processed by the processor 1920 , the processor 1920 can execute part or all of the methods described above.

[0307] In addition, the method according to the present application may also be implemented as a computer program or a computer program product, which includes computer program code instructions for executing some or all of the steps in the above method of the present application.

[0308] Alternatively, the present application can also be implemented as a computer-readable storage medium (or non-transitory machine-readable storage medium or machine-readable storage medium) on which executable code (or computer program or computer instruction code) is stored. When the executable code (or computer program or computer instruction code) is executed by a processor of an electronic device (or server, etc.), the processor executes part or all of the steps of the above-mentioned method according to the present application.

[0309] The embodiments of the present application have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A data processing method, characterized in that: include: Obtaining data to be processed, wherein the data to be processed includes attribute information of each of a plurality of atoms in a target molecule; In response to the attribute information of each of the multiple atoms, a node set and a node position set for the target molecule are generated, wherein the multiple nodes in the node set respectively represent atoms of a specific atomic type, and the node position set includes coordinate information of each node in the node set in a specific coordinate system; Generate a node scalar feature N for the node set s and node vector feature N v , and based on the coordinate information of each node in the node position set, an edge scalar feature E for the node set is generated s and edge vector feature E v ; Based on the node scalar feature N for the node set s 、Node vector feature N v , edge scalar feature E s and edge vector feature E v A virtual molecular graph is constructed to determine a molecular feature X of the target molecule based on the virtual molecular graph, so as to determine the solvation free energy based at least on the molecular feature X of the target molecule.

2. The method according to claim 1, characterized in that: The target molecule includes N atoms, and each of the plurality of nodes in the node set has an F-dimensional feature; The node scalar feature N s The dimensions include N×F×1 dimensions, the node vector feature N v The dimensions include N×F×3 dimensions, the edge scalar feature E s The dimension includes N×1×1 dimensions, the edge vector feature E v The dimensions include N×3×1 dimensions.

3. The method according to claim 1, characterized in that Also includes: After generating a node set and a node position set for the target molecule in response to the attribute information of each of the plurality of atoms, Determine the cutoff radius r cut ; Determine from the node set that the distance between nodes is less than or equal to the truncation radius r cut The target node is obtained by i ; The step of generating edge scalar features and edge vector features for the node set based on the coordinate information of each node in the node position set includes: Generate the coordinate information of the target node in the node position set for the target node set N i The edge scalar feature E s and edge vector feature E v .

4. The method according to claim 3, characterized in that The target node set includes E nodes, and each of the E nodes has an F-dimensional feature; The node scalar feature N s The dimensions include N×F×1 dimensions, the node vector feature N v The dimensions include N×F×3 dimensions, the edge scalar feature E s The dimensions include E×1×1 dimensions, the edge vector feature E v The dimensions include E×3×1 dimensions.

5. The method according to claim 2 or 4, characterized in that: The determining the molecular feature X of the target molecule based on the virtual molecular graph comprises: Update the node scalar feature N based on the virtual molecular graph s and the node vector feature N v , get the updated node scalar feature New_N s and update node vector feature New_N v ; Repeat the following operation until the specified number of loops num_conv is reached, so that the updated node scalar feature New_N obtained when the specified number of loops num_conv is reached is s , as the molecular feature X: The update node scalar feature New_N s and the updated node vector feature New_N v As the current node scalar feature Now_N s and the current node vector feature Now_N v ; Using the current node scalar feature Now_N s , the current node vector feature Now_N v , the edge scalar feature E s and the edge vector feature E v Construct and update virtual molecular graph; Update the updated node scalar feature New_N based on the updated virtual molecular graph s and the updated node vector feature New_N v .

6. The method according to claim 5, characterized in that The updating of the node scalar feature N based on the virtual molecular graph s and the node vector feature N v , get the updated node scalar feature New_N s and update node vector feature New_N v ,include: The node scalar feature N s The first linear operation, the second activation function and the second linear operation are performed in sequence to obtain the first sub-processing result Q1, and the edge scalar feature E s Perform a third linear operation to obtain a second sub-processing result Q2; Performing a first matrix corresponding multiplication operation on the first sub-processing result Q1 and the second sub-processing result Q2 to obtain a third sub-processing result Q3; Based on the third sub-processing result Q3 and the node vector feature N v Perform the corresponding multiplication operation of the second matrix to obtain the fourth sub-processing result Q4, and based on the third sub-processing result Q3 and the edge vector feature E v Perform a corresponding multiplication operation on the third matrix to obtain a fifth sub-processing result Q5; Performing a first matrix addition operation on the fourth sub-processing result Q4 and the fifth sub-processing result Q5 to obtain a sixth sub-processing result Q6; The sixth sub-processing result Q6 is subjected to a fourth linear operation and a fifth linear operation respectively to obtain a seventh sub-processing result Q7 and an eighth sub-processing result Q8; The seventh sub-processing result Q7 is subjected to the sixth linear operation, the second activation function, and the seventh linear operation in sequence to obtain a ninth sub-processing result Q9, and an inner product operation Inner is performed based on the third sub-processing result Q3, the seventh sub-processing result Q7, and the eighth sub-processing result Q8 to obtain a tenth sub-processing result Q10; Performing a fourth matrix corresponding multiplication operation on the ninth sub-processing result Q9 and the tenth sub-processing result Q10 to obtain an eleventh sub-processing result Q11; The fifth matrix corresponding multiplication operation is performed on the eighth sub-processing result Q8 and the ninth sub-processing result Q9 to obtain the updated node vector feature NewN v ; A second matrix addition operation is performed on the ninth sub-processing result Q9 and the eleventh sub-processing result Q11 to obtain the updated node scalar feature NewN s .

7. The method according to claim 5, characterized in that The target molecules are solute molecules and / or solvent molecules; The method further comprises: A solute molecule characteristic of a solute molecule and a solvent molecule characteristic of at least one solvent molecule associated with the solute molecule are determined so as to determine the solvation free energy based on the solute molecule characteristic of the solute molecule and the solvent molecule characteristic of at least one solvent molecule associated with the solute molecule.

8. The method according to claim 7, characterized in that Also includes: After determining the solute molecule characteristics of the solute molecule and the solvent molecule characteristics of at least one solvent molecule associated with the solute molecule, taking the matrix product of the solvent molecule characteristics and the solute molecule characteristics as the solvation interaction matrix between the solvent molecules and the solute molecules; A solvation signature is determined based on the solvation matrix.

9. The method according to claim 8, characterized in that in, Determining the solvation characteristics based on the solvation matrix comprises: calculating a solvent characteristic corresponding to a preset solute weight based on the solvation matrix, and calculating a solute characteristic corresponding to a preset solvent weight based on the solvation matrix; Convert the solvent characteristics and the solute characteristics into one-dimensional row vectors including F elements respectively; The row vectors are concatenated to obtain the solvation signature.

10. The method according to claim 9, characterized in that The step of converting the solvent characteristics and the solute characteristics into one-dimensional row vectors including F elements respectively includes: Determine a first element weight of an element in the solvent feature corresponding to an atom of the solvent molecule, and determine a second element weight of an element in the solute feature corresponding to an atom of the solute molecule; The solvent characteristics are weighted summed based on the first array element weights to obtain a one-dimensional first row vector including F elements, and the solute characteristics are weighted summed based on the second array element weights to obtain a one-dimensional second row vector including F elements.

11. A method for training a solvation free energy prediction model, characterized in that: A virtual molecular graph determined based on the method described in any one of claims 1 to 10 is input into the solvation free energy prediction model, and the model parameters are adjusted to converge the loss function to obtain a trained solvation free energy prediction model, wherein the virtual molecular graph has corresponding solvation free energy annotation information, and the input of the loss function includes the predicted solvation free energy and the solvation free energy in the solvation free energy annotation information.

12. The method according to claim 11, characterized in that The solvation free energy prediction model includes: an equivariant graph convolutional network configured to convert the virtual molecular graph into solute molecular features and / or solvent molecular features; The equivariant graph convolution network includes a convolution layer with a specified number of cycles num_conv, where the output of the current convolution layer is used as part of the input of the next convolution layer; the input of the first convolution layer includes: node scalar feature N s 、Node vector feature N v , edge scalar feature E s and edge vector feature E v , the output of the first convolutional layer includes: updating the node scalar feature New_N s and update node vector feature New_N v ; The input of the convolutional layer other than the first convolutional layer includes: updating the node scalar feature New_N s and update node vector feature New_N v , edge scalar feature E s and edge vector feature E v ; The output of the convolutional layer other than the first convolutional layer includes: updating the node scalar feature New_N s and update node vector feature New_N v .

13. The method according to claim 12, characterized in that The convolutional layers are configured as: The node scalar feature N s The first linear operation, the second activation function and the second linear operation are performed in sequence to obtain the first sub-processing result Q1, and the edge scalar feature E s Perform a third linear operation to obtain a second sub-processing result Q2; Performing a first matrix corresponding multiplication operation on the first sub-processing result Q1 and the second sub-processing result Q2 to obtain a third sub-processing result Q3; Based on the third sub-processing result Q3 and the node vector feature N v Perform the corresponding multiplication operation of the second matrix to obtain the fourth sub-processing result Q4, and based on the third sub-processing result Q3 and the edge vector feature E v Perform a corresponding multiplication operation on the third matrix to obtain a fifth sub-processing result Q5; Performing a first matrix addition operation on the fourth sub-processing result Q4 and the fifth sub-processing result Q5 to obtain a sixth sub-processing result Q6; The sixth sub-processing result Q6 is subjected to a fourth linear operation and a fifth linear operation respectively to obtain a seventh sub-processing result Q7 and an eighth sub-processing result Q8; The seventh sub-processing result Q7 is subjected to the sixth linear operation, the second activation function, and the seventh linear operation in sequence to obtain a ninth sub-processing result Q9, and an inner product operation Inner is performed based on the third sub-processing result Q3, the seventh sub-processing result Q7, and the eighth sub-processing result Q8 to obtain a tenth sub-processing result Q10; Performing a fourth matrix corresponding multiplication operation on the ninth sub-processing result Q9 and the tenth sub-processing result Q10 to obtain an eleventh sub-processing result Q11; A second matrix addition operation is performed on the ninth sub-processing result Q9 and the eleventh sub-processing result Q11 to obtain the updated node scalar feature NewN s ; The fifth matrix corresponding multiplication operation is performed on the eighth sub-processing result Q8 and the ninth sub-processing result Q9 to obtain the updated node vector feature NewN v .

14. The method according to claim 11, characterized in that: The solvation free energy prediction model includes: A molecular encoding network is configured to convert each training data in a training data set including solute molecule data and / or solvent molecule data into a virtual molecular graph for the solute molecule data and / or for the solvent molecule data, wherein the training data has solvation free energy annotation information, and atoms of the solute molecules or solvent molecules in the training data have F-dimensional features respectively; and / or The solvation free energy prediction model includes: a solvation network configured to convert solute molecule features and solvent molecule features into solvation features; wherein the solvation network comprises a self-attention network, the self-attention network being configured to determine a first element weight of an element corresponding to an atom of the solvent molecule in a solvent feature, and to determine a second element weight of an element corresponding to an atom of the solute molecule in a solute feature, so as to fuse the elements corresponding to each atom of the solvent molecule in the solvent feature according to the first element weight and fuse the elements corresponding to each atom of the solute molecule in the solute feature according to the second element weight, wherein the solvent feature and the solute feature are determined based on a solvation matrix, and the solvation matrix is ​​determined based on the solute molecule feature and the solvent molecule feature; and / or The solvation free energy prediction model includes: A fully connected network is configured to convert the solvation characteristics into solvation free energy; the fully connected network comprises: a first linear layer, a first activation function layer, a second linear layer, a second activation function layer and a third linear layer connected in sequence, wherein the output dimension of the first linear layer and the second linear layer is F dimension, and the output dimension of the third linear layer is 1 dimension.

15. The method according to claim 14, characterized in that Also includes: Splitting the training data set into a specified number of sub-training data sets; constructing a solvation free energy prediction model of the same number as the specified number; The inputting the training data into the molecular encoding network comprises: The training data in each sub-training data set is respectively input into the molecular encoding network of different solvation free energy prediction models to perform model training on different solvation free energy prediction models respectively, thereby obtaining a plurality of trained solvation free energy prediction models of the same number as the specified number.

16. A method for determining solvation free energy, characterized in that include: A virtual molecular graph is processed using a solvation free energy prediction model trained according to the method described in any one of claims 11 to 15 to obtain the solvation free energy for the virtual molecular graph, wherein the virtual molecular graph is a graph generated based on data to be processed, the data to be processed include attribute information for each of multiple atoms in a target molecule, and the target molecule includes solute molecules and / or solvent molecules.

17. The method according to claim 16, characterized in that Also includes: Inputting the virtual molecular graph or the data to be processed into different trained solvation free energy prediction models of a specified number to obtain the solvation free energies of the specified number; The weighted average of the specified number of solvation free energies is taken as the solvation free energy corresponding to the data to be processed.

18. A design method, characterized in that: The method comprises: The method according to any one of claims 1 to 17, determining the solvation free energy; Drug design or material design is performed based on the solvation free energy.

19. An electronic device, characterized in that: include: processor; as well as A memory having executable codes stored thereon, which, when executed by the processor, causes the processor to execute the method according to any one of claims 1 to 17.

20. A computer-readable storage medium, characterized in that: An executable code is stored thereon, and when the executable code is executed by a processor of an electronic device, the processor is caused to execute the method according to any one of claims 1-17.

Images