Molecular conformation generation method based on conditional generative adversarial network and storage medium

Abstract

The application relates to the technical field of molecular conformation, in particular to a molecular conformation generation method based on a conditional generative adversarial network and a storage medium. The molecular conformation generation method comprises the following steps: representing a to-be-generated molecule and a molecular motif corresponding to the molecule by using an undirected graph, and inputting the undirected graph into a graph neural network for coding to obtain a to-be-processed molecule graph and a molecular motif graph; inputting the to-be-processed molecule graph and the molecular motif graph into a trained conditional generative adversarial network to determine the distance between atoms in the molecule in the molecule graph; converting the distance between the atoms in the molecule into corresponding molecular three-dimensional coordinates; and obtaining the molecular conformation of the to-be-generated molecule according to the molecular three-dimensional coordinates. The molecular conformation generation method based on the conditional generative adversarial network can quickly and accurately generate the molecular conformation, improves the efficiency of drug molecule docking, and simultaneously reduces the screening time of the drug molecule.

Description

Technical Field

[0001] This application relates to the field of molecular conformation technology, and in particular to a molecular conformation generation method and storage medium based on conditional generative adversarial networks. Background Technology

[0002] In computational chemistry and computer-aided drug discovery, the three-dimensional structure of molecules is crucial because molecular conformation determines chemical and physical properties. Among related technologies, molecular conformation is primarily based on molecular dynamics (MD) or Monte Carlo (MC) methods. Molecular dynamics (MD) simulations calculate the force of each particle using the gradient of the potential function, and then use the evolution equations based on Newton's second law to obtain the system's position and velocity at the next moment, thus yielding the molecular conformation. However, the main difficulty with this method lies in the fact that higher-energy conformations have a lower probability of being sampled, and it is also significantly influenced by the initial conformation. The Monte Carlo (MC) method obtains new conformations by using bond rotations as degrees of freedom, and then uses the Metropolis algorithm to decide whether to accept the new conformation. The challenge of this method is that the direction and magnitude of random changes are difficult to predict, making it difficult to obtain the next important conformation. Furthermore, both methods require a significant amount of time to sample conformations of large molecules with higher degrees of freedom.

[0003] The commercial software Omega can rapidly generate possible molecular conformations of drug molecules. This method involves searching a molecular fragment library for corresponding fragments of the input molecule, then using a molecular dihedral library to find the corresponding angles between the fragments, and finally combining them to generate the conformation. However, this method only considers the interactions between fragments, not the overall interactions between molecules. Therefore, it may not provide accurate conformations for larger molecules. Summary of the Invention

[0004] The following is an overview of the subject matter described in detail herein. This overview is not intended to limit the scope of the claims. This application aims to at least address one of the technical problems existing in the prior art. To this end, embodiments of this application provide a molecular conformation generation method and storage medium based on conditional generative adversarial networks, which helps to solve the problems of low efficiency and low accuracy in molecular conformation generation. This application encodes the molecule to be generated and the molecular motif using a graph neural network to obtain a molecular graph and a molecular motif graph, and determines the distance between atoms within the molecule using a generative adversarial network. Based on distance geometry theory and minimizing potential energy, the molecular conformation is obtained. This method can generate molecular conformations quickly and accurately, improve the efficiency of drug molecule docking, and reduce the screening time for drug molecules.

[0005] In a first aspect, embodiments of this application provide a molecular conformation generation method based on conditional generative adversarial networks, including:

[0006] The molecule to be generated and the molecular motif corresponding to the molecule are represented by an undirected graph, and the undirected graph is input into a graph neural network for encoding to obtain the molecular graph and molecular motif graph to be processed.

[0007] The molecular graph and molecular motif graph to be processed are input into the trained conditional generative adversarial network to determine the distances between atoms within the molecule in the molecular graph.

[0008] The distances between atoms within the molecule are converted into corresponding three-dimensional molecular coordinates;

[0009] The molecular conformation of the molecule to be generated is obtained based on the three-dimensional coordinates of the molecule.

[0010] The technical solution of the first aspect of this application has at least one of the following advantages or beneficial effects: by representing the molecule to be generated and the corresponding molecular motif with an undirected graph, and encoding the undirected graph through a graph neural network, it is convenient to convert the three-dimensional molecule into a data form that can be recognized by a computer; by inputting the molecular graph and molecular motif graph to be processed into a trained generative adversarial network, the distance between atoms within the molecule is determined, and the molecular conformation of the molecule to be generated is obtained based on the distance between atoms within the molecule, which can generate molecular conformations quickly and accurately, improve the efficiency of drug molecule docking, and reduce the screening time of drug molecules.

[0011] Furthermore, after obtaining the molecular conformation based on the molecular three-dimensional coordinates, the method further includes:

[0012] The molecular conformations are clustered using the root mean square error;

[0013] Determine the energy of the molecular conformation after clustering.

[0014] Furthermore, the trained conditional generative adversarial network is obtained through the following method:

[0015] Retrieve molecular conformations from the database for training;

[0016] The molecular conformations used for training are converted into a first molecular diagram and a first molecular motif diagram;

[0017] The first molecular map and the first molecular motif map are used as training features;

[0018] The distance between atoms within the first molecule is determined based on the training features, and the distance between atoms within the first molecule is used as the training label;

[0019] The training features are used as input to the conditional generative adversarial network (GAN), and the training labels are used as output to train the GAN, resulting in a trained GAN.

[0020] Furthermore, the conditional generative adversarial network includes a generator and a discriminator. The conditional generative adversarial network is modulated using a loss function preset by the generator and a loss function preset by the discriminator to improve the ability of the conditional generative adversarial network to generate molecular conformations.

[0021] Furthermore, the step of inputting the molecule to be generated and the molecular motif into a graph neural network for encoding to obtain the molecular graph and molecular motif graph to be processed includes:

[0022] The molecules to be generated and their molecular motifs are represented by undirected graphs.

[0023] The undirected graph is encoded using a graph neural network to obtain the molecular graph and molecular motif graph to be processed;

[0024] The encoding form is to represent the undirected graph as G = (V, E), where V = {v...} i} represents a set of nodes, where each node v i It includes the properties of atoms, including atom type, atomic charge, and chirality; E = {e k} represents the set of edges, where each e k The edges contain the types of chemical bonds.

[0025] Furthermore, the loss function of the conditional generative adversarial network is:

[0026]

[0027]

[0028]

[0029] Where U represents the potential energy of the actual molecular conformation. This represents the potential energy of the molecular conformation generated by the generator, which is composed of the resonant potential of bonding interactions, the Lerner-Jones potential of nonbonding interactions, and the Coulomb potential of electrostatic interactions. ij This represents the distance between atoms i and j. k represents the equilibrium distance between atoms i and j. ij A ij B ij Represents the UFF force field parameters; The loss function of the discriminator is used to evaluate the difference between the conformation of the generated molecule and the true conformation. Let U be the loss function of the generator, used to make the potential energy of the generated molecule closer to the actual molecular conformation, and D(U) be the molecular conformation potential energy output by the discriminator. Let λ be the expectation of D(U), and α be the hyperparameters.

[0030] Furthermore, the step of converting the distances between atoms within the molecule into corresponding three-dimensional molecular coordinates includes:

[0031] Based on distance geometry theory and minimum optimization potential energy, the distances between atoms within a molecule are converted into the corresponding three-dimensional molecular conformation;

[0032] The gradient descent method is used to minimize the distance geometry theory and the minimum optimization potential energy to obtain the corresponding molecular three-dimensional coordinates;

[0033] The distance geometry theory converts the distance between atoms within a molecule into atomic coordinates, and the minimum optimization potential energy optimizes the position of atoms by minimizing the potential energy of the molecule.

[0034] Furthermore, the clustering of the molecular conformations using the root mean square error includes:

[0035] Determine the mean square error between the various generated molecular conformations;

[0036] When the mean square error is less than a preset error value, the molecular conformations whose mean square error is less than the preset error value are determined to be the same molecular conformation.

[0037] Secondly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions for performing the molecular conformation generation method based on conditional generative adversarial networks as described in the first aspect above. Attached Figure Description

[0038] Figure 1 This is a flowchart illustrating the steps of a molecular conformation generation method based on conditional generative adversarial networks provided in an embodiment of this application.

[0039] Figure 2 This is a flowchart of another molecular conformation generation method based on conditional generative adversarial networks provided in the embodiments of this application;

[0040] Figure 3 This is a flowchart illustrating the steps of training a generative adversarial network according to an embodiment of this application;

[0041] Figure 4 This is a flowchart of another molecular conformation generation method based on conditional generative adversarial networks provided in the embodiments of this application;

[0042] Figure 5 yes Figure 1 Flowchart of steps in S100;

[0043] Figure 6 yes Figure 1 Flowchart of steps in the S300 process;

[0044] Figure 7 yes Figure 1 Flowchart of the steps in the S500;

[0045] Figure 8 This is a schematic diagram of the structure of a controller provided in an embodiment of this application. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0047] In the description of this application, "multiple" refers to two or more. The use of "first" and "second" is for the purpose of distinguishing technical features only and should not be construed as indicating or implying relative importance, or implicitly indicating the number of technical features indicated, or the order in which the technical features are indicated.

[0048] In related technologies, molecular conformation is mainly based on molecular dynamics (MD) or Monte Carlo (MC) methods. Molecular dynamics (MD) simulations calculate the force of each particle using the gradient of the potential function, and then use the evolution equations based on Newton's second law to obtain the system's position and velocity at the next moment, thus obtaining the molecular conformation. However, this method has a lower probability of sampling conformations with higher energies and is significantly affected by the initial conformation. The Monte Carlo (MC) method obtains new conformations by using bond rotations as degrees of freedom, and then uses the Metropolis algorithm to decide whether to accept the new conformation. However, it is difficult to predict the direction and magnitude of random changes, making it difficult to obtain the next important conformation. Furthermore, both methods require a significant amount of time to sample conformations of large molecules with higher degrees of freedom. The commercial software Omega searches for corresponding molecular fragments of the input molecule in a molecular fragment library, then uses a molecular dihedral angle library to find the corresponding angles between the fragments, and finally combines them to generate a conformation. However, this method only considers the interactions between fragments, not the overall interactions between molecules. Therefore, it may not provide accurate conformations for larger molecules.

[0049] To address this issue, this application provides a molecular conformation generation method and storage medium based on conditional generative adversarial networks (GANs), which helps solve the problems of low efficiency and low accuracy in molecular conformation generation. This application uses a graph neural network to encode the molecule to be generated and its molecular motif, obtaining a molecular graph and a molecular motif graph. Then, it uses a GAN to determine the distances between atoms within the molecule. Based on distance geometry theory and minimizing potential energy, the molecular conformation is obtained. This method can generate molecular conformations quickly and accurately, improving the efficiency of drug molecule docking while reducing drug molecule screening time.

[0050] Reference Figure 1 , Figure 1 This is a flowchart illustrating the steps of a molecular conformation generation method based on conditional generative adversarial networks provided in this application embodiment, including steps S100 to S400. Specifically,

[0051] S100: Represent the molecule to be generated and the corresponding molecular motif using an undirected graph, and input the undirected graph into a graph neural network for encoding to obtain the molecular graph and molecular motif graph to be processed;

[0052] S200: Input the molecular graph and molecular motif graph to be processed into the trained conditional generative adversarial network to determine the distance between atoms in the molecular graph;

[0053] S300: Converts the distances between atoms within a molecule into corresponding three-dimensional molecular coordinates;

[0054] S400: Obtain the molecular conformation of the molecule to be generated based on the three-dimensional coordinates of the molecule.

[0055] By representing the molecule to be generated and its corresponding molecular motif using an undirected graph, and encoding the undirected graph using a graph neural network, it is easy to convert three-dimensional molecules into a data format that computers can recognize. The molecular graph and molecular motif graph to be processed are input into a trained generative adversarial network to determine the distances between atoms within the molecule. Based on the distances between atoms within the molecule, the corresponding three-dimensional coordinates of the molecule are obtained, and the molecular conformation of the molecule to be generated is obtained from the three-dimensional coordinates. Encoding the undirected graph using a graph neural network and processing the molecular graph and molecular motif graph using a trained generative adversarial network enables rapid and accurate generation of molecular conformations, improving the efficiency of drug molecule docking and reducing drug molecule screening time.

[0056] It should be noted that the trained generative adversarial network includes a generator and a discriminator. The generator is used to generate the distances between atoms within a molecule. The preset loss functions of the generator and the discriminator modulate the generative adversarial network to improve its ability to generate molecular conformations.

[0057] It should be noted that in the embodiments of this application, the molecular motif is obtained by splitting the molecule into molecular fragments. The splitting is based on the Recap splitting method, which can simulate the forward synthesis process in the laboratory to perform reverse operation, and perform a series of transformations and decompositions on the molecule to obtain the molecular fragments corresponding to the molecule.

[0058] Reference Figure 2 , Figure 2 This is a flowchart of another molecular conformation generation method based on conditional generative adversarial networks provided in this application embodiment, including steps S100 to S600, specifically,

[0059] S100: Represent the molecule to be generated and the corresponding molecular motif using an undirected graph, and input the undirected graph into a graph neural network for encoding to obtain the molecular graph and molecular motif graph to be processed;

[0060] S200: Input the molecular graph and molecular motif graph to be processed into the trained conditional generative adversarial network to determine the distance between atoms in the molecular graph;

[0061] S300: Converts the distances between atoms within a molecule into corresponding three-dimensional molecular coordinates;

[0062] S400: Obtain the molecular conformation of the molecule to be generated based on the three-dimensional coordinates of the molecule;

[0063] S500: Cluster the molecular conformations using the root mean square error;

[0064] S600: Determines the energy of the molecular conformation after clustering.

[0065] In one embodiment of this application, after obtaining the molecular conformation of the molecule to be generated based on the three-dimensional coordinates of the molecule, the molecular conformation generation method based on conditional generative adversarial networks further includes clustering the molecular conformations using root mean square error and determining the energy of the clustered molecular conformations. By clustering the generated molecular conformations, similar molecular conformations can be clustered, redundant conformations can be reduced, the efficiency of drug molecule docking can be improved, and the screening time for drug molecules can be reduced.

[0066] It should be noted that, in the embodiments of this application, the energy of the clustered molecular conformations can be determined by a semi-empirical method or by a practical quantum mechanical method. The embodiments of this application do not limit the calculation method for determining the energy of the clustered molecular conformations.

[0067] In one embodiment of this application, the distribution of molecular conformational energy is calculated by Boltzmann averaging, and the results are output.

[0068] Reference Figure 3 , Figure 3This is a flowchart illustrating the steps of training a generative adversarial network according to an embodiment of this application, including steps S700 to S740. Specifically,

[0069] S700: Retrieve molecular conformations for training from the database;

[0070] S710: Converts the molecular conformation used for training into a first molecular diagram and a first molecular motif diagram;

[0071] S720: Use the first molecule graph and the first molecule motif graph as training features;

[0072] S730: Determine the distance between atoms within the first molecule based on the training features, and use the distance between atoms within the first molecule as the training label;

[0073] S740: The training features are used as input to the conditional generative adversarial network (GAN) and the training labels are used as output to train the GAN, resulting in the trained GAN.

[0074] This application obtains molecular conformations for training from a database and converts them into a first molecular map and corresponding molecular motifs. The first molecular map and first molecular motif map are used as training features, and the distances between atoms within the first molecule are used as training labels. The training features are used as input to a conditional generative adversarial network (GAN), and the training labels are used as the output of the GAN for training, resulting in a trained GAN. After training, the GAN only needs to input the molecular map to be generated and the corresponding molecular motif into the trained GAN to output the distances between atoms within the molecule, thus obtaining the molecular conformation to be generated. This reduces the time required for drug molecule conformation generation, improves the efficiency of drug molecule docking, and reduces drug molecule screening time.

[0075] It should be noted that, in the embodiments of this application, determining the distance between atoms within the first molecule includes determining the distance d between atoms of the first molecule. 12 d 13 d 14 Among them, d 12 d represents the distance between adjacent atoms. 13 This represents the distance between the first and third atoms within the first molecule, where the first and second atoms are connected, the second and third atoms are connected, and the first and third atoms form a predetermined angle α. 1-2-3 . d 14 This represents the distance between the first and fourth atoms, and the first and fourth atoms form a dihedral angle β. 1-2-3-4 .

[0076] It should be noted that in the embodiments of this application, the distance between atoms within the molecule is calculated during the training process. It is not necessary to calculate it when generating the molecular conformation to be generated in real time. The distance between atoms within the molecule can be obtained simply by inputting the molecular image to be generated and the corresponding molecular motif into the trained conditional generation network.

[0077] It should be noted that the embodiments of this application do not limit the number of atoms in the molecule, and the method for calculating the distance between atoms in the first molecule is based on the number of atoms in the molecule to be generated.

[0078] In one embodiment of this application, the conditional generative adversarial network includes a generator and a discriminator. The generator and discriminator each contain multiple fully connected layers constructed from neural networks. The conditional generative adversarial network is modulated using a preset loss function for the generator and a preset loss function for the discriminator. The loss function of the conditional generative adversarial network is...

[0079]

[0080]

[0081]

[0082] Where U represents the potential energy of the actual molecular conformation. This represents the potential energy of the molecular conformation generated by the generator, which is composed of the resonant potential of bonding interactions, the Lerner-Jones potential of nonbonding interactions, and the Coulomb potential of electrostatic interactions. ij This represents the distance between atoms i and j. k represents the equilibrium distance between atoms i and j. ij A ij B ij q i and q j For UFF force field parameters, The loss function of the discriminator is used to evaluate the difference between the conformation of the generated molecule and the true conformation. Let U be the loss function of the generator, used to make the potential energy of the generated molecule closer to the actual molecular conformation, and D(U) be the molecular conformation potential energy output by the discriminator. Let λ be the expectation of D(U), and α be the hyperparameters.

[0083] By modulating the conditional generative adversarial network (GAN) with preset loss functions for the generator and the discriminator, the ability of the GAN to generate molecular conformations is improved, and the accuracy of the generated molecular conformations is also improved.

[0084] It should be noted that the training set used in training the conditional generative adversarial network using the loss function is the publicly available databases GEMO-DRUG and GEMO-QM9. The loss function for training the conditional generative adversarial network is constructed by extracting molecular graphs, interatomic distances, and field parameters of all molecular conformations in the database.

[0085] It should be noted that the force field parameters in the embodiments of this application include the parameters of the bonding interaction resonance potential, the nonbonding interaction Lennard-Jones potential, and the charge interaction Coulomb potential.

[0086] It should be noted that the loss function uses the WGAN-GP algorithm, and the input potential energy is obtained by calculating the resonance potential of bonding interactions, the Lennard-Jones potential of non-bonding interactions, and the Coulomb interaction potential of electrons through the UFF force field parameters.

[0087] Reference Figure 4 , Figure 4 This is a flowchart illustrating another molecular conformation generation method based on conditional generative adversarial networks provided in this application embodiment, which superimposes randomly generated Gaussian noise onto the molecular graph and molecular motif. Figure 1 The inputs are fed into the generator, which determines the distances between atoms within the molecule in the molecular diagram based on the input molecular diagram and molecular motif diagram. Then, based on distance geometry theory and minimum optimization potential energy, the generator converts these distances into the corresponding three-dimensional molecular conformation. The purpose of adding Gaussian noise is to increase the diversity of the generated structures.

[0088] Reference Figure 5 , Figure 5 yes Figure 1 The flowchart of step S100 includes steps S110 to S120, specifically,

[0089] S110: Represent the molecules to be generated and molecular motifs using an undirected graph;

[0090] S120: Encode the undirected graph using a graph neural network to obtain the molecular graph and molecular motif graph to be processed;

[0091] By representing the molecule to be generated and its corresponding molecular motif using an undirected graph, and encoding the undirected graph using a graph neural network, it is easier to convert the three-dimensional molecule into a data form that a computer can recognize. The encoding form represents the undirected graph as G = (V, E), where V = {v...} i} represents a set of nodes, where each node v i It includes the properties of atoms, including atom type, atomic charge, and chirality; E = {e k} represents the set of edges, where each ek The edges contain the types of chemical bonds. By representing the molecule to be generated and its corresponding molecular motif using an undirected graph, and in conjunction with the Recap resolution method, the reverse operation can be performed to simulate the forward synthesis process in the laboratory. This allows for a series of transformations and decompositions of the molecule, improving the accuracy of the generated molecular conformation. At the same time, it increases the efficiency of drug molecule docking and reduces the drug molecule screening time.

[0092] Reference Figure 6 , Figure 6 yes Figure 1 The flowchart of steps S300 includes steps S310 to S320, specifically...

[0093] S310: Based on distance geometry theory and minimum optimization potential energy, the distance between atoms within a molecule is converted into the corresponding three-dimensional molecular conformation;

[0094] S320: The gradient descent method is used to minimize the distance geometry theory and the minimum optimization potential energy to obtain the corresponding molecular three-dimensional coordinates.

[0095] In one embodiment of this application, the distances between atoms within a molecule are converted into corresponding three-dimensional molecular conformations based on distance geometry theory and minimum optimization potential energy, as achieved by the following formula.

[0096]

[0097]

[0098] Where R is the three-dimensional coordinate of the molecular conformation, r i and r j Let d represent the three-dimensional coordinates of atom i and atom j respectively. ij This refers to the distance between atoms i and j, U ij This refers to the potential energy between atoms i and j. By minimizing the distance geometry theory and the minimum optimization potential energy through the gradient descent method, the corresponding three-dimensional molecular coordinates R are obtained. Based on the three-dimensional coordinates R, the corresponding molecular conformation is determined, which improves the accuracy of the generated molecular conformation. At the same time, it improves the efficiency of drug molecule docking and effectively reduces the drug molecule screening time.

[0099] It should be noted that distance geometry theory converts the distance between atoms within a molecule into atomic coordinates, while minimum optimization potential energy optimizes the position of atoms by minimizing the potential energy of the molecule.

[0100] Reference Figure 7 , Figure 7 yes Figure 1 The flowchart of steps S500 includes steps S510 to S520, specifically...

[0101] S510: Determine the mean square error between the generated molecular conformations;

[0102] S520: When the root mean square error is less than a preset error value, the molecular conformations with root mean square errors less than the preset error value are determined to be the same molecular conformation.

[0103] In this embodiment, clustering molecular conformations using root mean square error includes determining the mean square error between each generated molecular conformation; when the mean square error is less than a preset error value, molecular conformations with mean square errors less than the preset error value are determined to be the same molecular conformation. By clustering the generated molecular conformations, similar molecular conformations that meet the preset error conditions can be determined to be the same molecular conformation, effectively reducing redundant conformations, improving the efficiency of drug molecule docking, and reducing drug molecule screening time.

[0104] It should be noted that in this embodiment, the generated molecular conformations are also stored in a database used to train the conditional generative network. By clustering the generated molecular conformations, the memory usage of the database can be reduced, and similar molecular conformations can be classified, thereby improving the accuracy and efficiency of generating molecular conformations.

[0105] It should be noted that the preset error value in the embodiments of this application is... One or more of the above, and the present application embodiments do not limit the magnitude of the preset error value.

[0106] In one embodiment of this application, the preset error value is: When the mean square error is less than Then the mean square error is determined to be less than The molecular conformations are the same; when the mean square error is greater than 100%. Determine the mean square error is greater than The molecular conformations are not the same molecular conformations.

[0107] Figure 8This is a schematic diagram of the structure of a controller 1000 provided in an embodiment of this application. It includes a processor 1001, which can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement a molecular conformation generation method based on conditional generative adversarial networks provided in this embodiment of the application; and a memory 1002, which can be implemented using a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM), etc. The memory 1002 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program code is stored in the memory 1002 and is called and executed by the processor 1001. The input / output interface 1003 is used to implement information input and output. The communication interface 1004 is used to realize communication interaction between this device and other devices. Communication can be realized by wired means (e.g., USB, network cable, etc.) or by wireless means (e.g., mobile network, WIFI, Bluetooth, etc.). The bus transmits information between the various components of the device (e.g., processor 1001, memory 1002, input / output interface 1003 and communication interface 1004). The processor 1001, memory 1002, input / output interface 1003 and communication interface 1004 are connected to each other within the device through the bus.

[0108] This application also provides a storage medium, which is a computer-readable storage medium storing a computer program. When executed by a processor, this computer program implements a flowchart of the aforementioned molecular conformation generation method based on a conditional generative adversarial network. As a non-transitory computer-readable storage medium, the memory can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, the memory may include high-speed random access memory and non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0109] The embodiments of this application have been described in detail above with reference to the accompanying drawings. However, this application is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of this application.

Claims

1. A molecular conformation generation method based on conditional generative adversarial networks, characterized in that, include: The molecule to be generated and the molecular motif corresponding to the molecule are represented by an undirected graph, and the undirected graph is input into a graph neural network for encoding to obtain the molecular graph and molecular motif graph to be processed. The molecular graph and molecular motif graph to be processed are input into the trained conditional generative adversarial network to determine the distances between atoms within the molecule in the molecular graph. The distances between atoms within the molecule are converted into corresponding three-dimensional molecular coordinates; The molecular conformation of the molecule to be generated is obtained based on the three-dimensional coordinates of the molecule.

2. The molecular conformation generation method based on conditional generative adversarial networks according to claim 1, characterized in that, After obtaining the molecular conformation of the molecule to be generated based on the three-dimensional coordinates of the molecule, the method further includes: The molecular conformations are clustered using the root mean square error; Determine the energy of the molecular conformation after clustering.

3. The molecular conformation generation method based on conditional generative adversarial networks according to claim 1, characterized in that, The trained conditional generative adversarial network was obtained through the following method: Retrieve molecular conformations from the database for training; The molecular conformations used for training are converted into a first molecular diagram and a first molecular motif diagram; The first molecular map and the first molecular motif map are used as training features; The distance between atoms within the first molecule is determined based on the training features, and the distance between atoms within the first molecule is used as the training label; The training features are used as input to the conditional generative adversarial network (GAN), and the training labels are used as output to train the GAN, resulting in a trained GAN.

4. The molecular conformation generation method based on conditional generative adversarial networks according to claim 3, characterized in that, The conditional generative adversarial network includes a generator and a discriminator. The conditional generative adversarial network is modulated using a loss function preset by the generator and a loss function preset by the discriminator to improve the ability of the conditional generative adversarial network to generate molecular conformations.

5. The molecular conformation generation method based on conditional generative adversarial networks according to claim 4, characterized in that, The method further includes: Gaussian noise is superimposed on the molecular map and the molecular motif map and input together into the generator to improve the diversity of molecular conformations generated by the generator.

6. The molecular conformation generation method based on conditional generative adversarial networks according to claim 1, characterized in that, The step of inputting the molecule and molecular motif to be generated into a graph neural network for encoding to obtain the molecular graph and molecular motif graph to be processed includes: The molecules to be generated and their molecular motifs are represented by undirected graphs. The undirected graph is encoded using a graph neural network to obtain the molecular graph and molecular motif graph to be processed; The encoding form is to represent the undirected graph as... G =( V , E ), V ={ v i } represents a set of nodes, each node v i It includes the properties of atoms, including atom type, atomic charge, and chirality; E = { e k } represents the set of edges, each e k The edges contain the types of chemical bonds.

7. The molecular conformation generation method based on conditional generative adversarial networks according to claim 4, characterized in that, The loss function of the conditional generative adversarial network is: in, Potential energy representing the actual molecular conformation. The potential energy represents the molecular conformation generated by the generator, which is composed of the bonding interaction harmonic potential, the non-bonding interaction Lerner-Jones potential, and the electrostatic interaction Coulomb potential. Representing atoms i and atoms j The distance between them This represents the equilibrium distance between atoms i and j. , , , and These are UFF force field parameters; The loss function of the discriminator is used to evaluate the difference between the conformation of the generated molecule and the conformation of the real molecule. The loss function of the generator is used to make the potential energy of the generated molecule closer to the conformation of the real molecule. It is the molecular conformational potential energy output by the discriminator. for Expectations and It's a hyperparameter.

8. The molecular conformation generation method based on conditional generative adversarial networks according to claim 6, characterized in that, The step of converting the distances between atoms within the molecule into corresponding three-dimensional molecular coordinates includes: Based on distance geometry theory and minimum optimization potential energy, the distances between atoms within a molecule are converted into the corresponding three-dimensional molecular conformation; The gradient descent method is used to minimize the distance geometry theory and the minimum optimization potential energy to obtain the corresponding molecular three-dimensional coordinates; The distance geometry theory converts the distance between atoms within a molecule into atomic coordinates, and the minimum optimization potential energy optimizes the position of atoms by minimizing the potential energy of the molecule.

9. The molecular conformation generation method based on conditional generative adversarial networks according to claim 2, characterized in that, The method of clustering the molecular conformations using root mean square error includes: Determine the mean square error between the various generated molecular conformations; When the mean square error is less than a preset error value, the molecular conformations whose mean square error is less than the preset error value are determined to be the same molecular conformation.

10. A computer-readable storage medium, characterized in that: The device stores computer-executable instructions for performing the molecular conformation generation method based on conditional generative adversarial networks as described in any one of claims 1 to 9.

Images