Method, device, electronic device and storage medium for determining electron density map

Abstract

The present application provides a method, device, electronic device and storage medium for determining an electron density map, which belongs to the field of computer technology. Among them, the method for determining the electron density map includes: based on the molecular structure of the target object, constructing a three-dimensional coordinate system and performing three-dimensional gridding to obtain each grid point corresponding to the target object; taking each atom and each network point as a node, and based on the three-dimensional coordinates of each atom and the three-dimensional coordinates of each grid point, determining the node features of each node corresponding to the target object, and the edge features between each two nodes; inputting each node feature and each edge feature into a trained electron density map generation model to obtain the electron density of each grid point, respectively, wherein the electron density of each grid point represents the probability of finding an electron at the corresponding grid point. The above method can improve the accuracy of the electron density map.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and more specifically, to a method, device, electronic device, and storage medium for determining an electron density map. Background Art

[0002] The actual role of proteins in organisms (such as causing certain genetic diseases, or having immunity to specific diseases) is largely determined by their three-dimensional structure. Therefore, how to accurately and efficiently obtain the three-dimensional structure of proteins through experiments or calculations has a vital impact on understanding the functions and effects of proteins in organisms. At present, the protein three-dimensional structure determination method based on cryo-electron microscopy, in the process of parsing the three-dimensional structure of proteins based on cryo-electron microscopy electron density map data, a key step is to simulate and generate the corresponding electron density maps for a series of candidate protein three-dimensional structures, and then determine the optimization direction of the candidate protein three-dimensional structure by comparing the consistency between the simulated electron density map and the electron density map obtained by the actual experiment, so as to obtain the protein three-dimensional structure that is more consistent with the experimental data through iterative optimization as the final parsed structure.

[0003] Generating electron density maps from protein 3D structure simulations is crucial for protein structure analysis based on cryo-EM. If the simulated electron density maps are inaccurate, it will be impossible to select candidate protein 3D structures that better match experimental data, or to determine optimization directions for candidate protein 3D structures.

[0004] Existing methods often simulate and generate corresponding cryo-EM electron density maps from protein 3D structures based on a single Gaussian assumption. This single Gaussian assumption is insufficient to fully fit the correspondence between protein 3D structures and electron density maps, and it also fails to fully account for interatomic interactions and the mutual influence of electron density at different grid points, resulting in low accuracy. Summary of the Invention

[0005] To solve the technical problems existing in the related art, the embodiments of the present application provide a method, device, electronic device and storage medium for determining an electron density map, so as to improve the accuracy of the electron density map.

[0006] To achieve the above objectives, the technical solution of the embodiment of the present application is implemented as follows:

[0007] In a first aspect, an embodiment of the present application provides a method for determining an electron density map, the method comprising:

[0008] Based on the molecular structure of the target object, a three-dimensional coordinate system is constructed and three-dimensional gridding is performed to obtain grid points corresponding to the target object, wherein the molecular structure contains atoms of the target object, and the grid points are vertices corresponding to each grid after the three-dimensional coordinate system is divided into a plurality of three-dimensional grids;

[0009] Taking each atom and each grid point as a node, and determining the node features of each node corresponding to the target object and the edge features between every two nodes based on the three-dimensional coordinates of each atom and the three-dimensional coordinates of each grid point, the node features are used to characterize the elements that the nodes affect the electron density, and the edge features are used to characterize the elements that the interactions between nodes affect the electron density;

[0010] The node features and the edge features are input into a trained electron density map generation model to obtain the electron density of each grid point, wherein the electron density of each grid point represents the probability of finding an electron at the corresponding grid point.

[0011] In a second aspect, an embodiment of the present application provides a device for determining an electron density map, the device comprising:

[0012] a construction unit for constructing a three-dimensional coordinate system and performing three-dimensional gridding based on the molecular structure of the target object to obtain grid points corresponding to the target object, wherein the molecular structure contains atoms of the target object, and the grid points are vertices corresponding to each grid after the three-dimensional coordinate system is divided into a plurality of three-dimensional grids;

[0013] a feature unit, configured to treat each atom and each grid point as a node, and determine, based on the three-dimensional coordinates of each atom and the three-dimensional coordinates of each grid point, a node feature of each node corresponding to the target object, and an edge feature between every two nodes, wherein the node feature is used to characterize the elements of the node that affect the electron density, and the edge feature is used to characterize the elements of the interaction between nodes that affect the electron density;

[0014] A model unit is used to input the node features and the edge features into a trained electron density map generation model to obtain the electron density of each grid point, wherein the electron density of each grid point represents the probability of finding an electron at the corresponding grid point.

[0015] In an optional embodiment, the feature unit is specifically used to:

[0016] For each grid point, perform the following operations:

[0017] Determine a neighborhood of one of the grid points; the neighborhood represents a space within a set distance around the one grid point;

[0018] determining each atom within the neighborhood and each other grid point within the neighborhood;

[0019] Determining, based on the measurement data, a grid feature of the one grid point, an atomic feature of each atom in the neighborhood, and a grid feature of each other grid point in the neighborhood, and using the obtained grid features and atomic features as node features in the neighborhood;

[0020] Based on the three-dimensional coordinates of the one grid point, the three-dimensional coordinates of each atom in the neighborhood, and the three-dimensional coordinates of each other grid point in the neighborhood, edge features between every two nodes in the neighborhood are determined.

[0021] In an optional embodiment, the edge features within the neighborhood include any one or any combination of the following:

[0022] The edge features between the atoms in the neighborhood, the edge features between the one grid point and each other grid point in the neighborhood, and the edge features between the one grid point and each atom in the neighborhood.

[0023] In an optional embodiment, the edge features between atoms in the neighborhood include: the distance between every two atoms in the neighborhood, and the relative coordinates between every two atoms in the neighborhood;

[0024] The edge features between the one grid point and each other grid point in the neighborhood include: the distance between the one grid point and each other grid point in the neighborhood, and the relative coordinates between the one grid point and each other grid point in the neighborhood;

[0025] The edge features between the grid point and each atom in the neighborhood include: the distance between the grid point and each atom in the neighborhood, and the relative coordinates between the grid point and each atom in the neighborhood.

[0026] In an optional embodiment, the electron density map generation model includes N feature update layers and electron density prediction layers, the feature update layer is a deep graph neural network, and N is a positive integer.

[0027] In an optional embodiment, the model unit is specifically used to:

[0028] For each grid point, perform the following operations:

[0029] Inputting each node feature and each edge feature in the neighborhood of one of the grid points into the trained electron density map generation model;

[0030] Based on the N feature update layers, performing multi-layer nonlinear transformation on each of the node features and each of the edge features to obtain corresponding target node features and target edge features;

[0031] Based on the electron density prediction layer, the obtained target node features and target edge features are calculated to obtain the electron density of the one grid point.

[0032] In an optional embodiment, the method further comprises a training unit, configured to train the electron density map generation model according to the following process:

[0033] Obtaining training samples and experimental electron density corresponding to the training samples;

[0034] The electron density map generation model is iteratively trained based on the training samples and the experimental electron densities corresponding to the training samples until a set training end condition is reached, thereby obtaining a trained electron density map generation model. One iterative process includes:

[0035] Inputting the training sample into an electron density map generation model to determine the training electron density of the training sample;

[0036] Determining a loss function according to the experimental electron density and the training electron density;

[0037] Parameters of the electron density map generation model are adjusted according to the loss function.

[0038] In an optional embodiment, the feature update layer is an information transfer neural network.

[0039] In a third aspect, an embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for determining the electron density map of the first aspect is implemented.

[0040] In a fourth aspect, an embodiment of the present application further provides an electronic device comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the computer program is executed by the processor, the processor implements the method for determining the electron density map of the first aspect.

[0041] The embodiment of the present application constructs a three-dimensional coordinate system and performs three-dimensional gridding based on the molecular structure of the target object to obtain each grid point corresponding to the target object. Wherein, the molecular structure contains each atom of the target object, and the grid points are the vertices corresponding to each grid after the three-dimensional coordinate system is divided into a number of three-dimensional grids. Each atom and each network point is regarded as a node, and based on the three-dimensional coordinates of each atom and the three-dimensional coordinates of each grid point, the node features of each node corresponding to the target object and the edge features between each two nodes are determined. Wherein, the node features are used to characterize the elements that affect the electron density of the node, and the edge features are used to characterize the elements that affect the electron density of the interaction between nodes. Each node feature and each edge feature are input into the trained electron density map generation model to obtain the electron density of each grid point, respectively, wherein the electron density of each grid point represents the probability of finding an electron at the corresponding grid point. In this way, by modeling the atomic set in the three-dimensional space where the target object is located and the corresponding grid point set in the electron density map as a heterogeneous graph structure, and then using the algorithm model to model and learn the interactions between atoms, the interactions between atoms and grid points, and the interactions between grid points, the interactions between atoms and the electron densities of different grid points are fully considered, which can better simulate the electron density map and improve the accuracy and precision generated from the electron density map. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0043] Figure 1 A schematic diagram of an application scenario of a method for determining an electron density map provided in an embodiment of the present application;

[0044] Figure 2 A flowchart of a method for determining an electron density map provided in an embodiment of the present application;

[0045] Figure 3 A cryo-electron microscopy electron density map and a schematic diagram of the corresponding protein three-dimensional structure provided in an embodiment of the present application;

[0046] Figure 4 A schematic diagram of the structure of an electron density map generation model provided in an embodiment of the present application;

[0047] Figure 5 A schematic diagram of the training of the electron density map generation model provided in an embodiment of the present application;

[0048] Figure 6 A schematic diagram comparing the accuracy of different cryo-EM electron density map generation methods;

[0049] Figure 7 A schematic structural diagram of an apparatus for determining an electron density map provided in an embodiment of the present application;

[0050] Figure 8 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0051] To make the objectives, technical solutions, and advantages of this application more clear, this application will be further described in detail below with reference to the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of this application.

[0052] The word “exemplary” is used hereinafter to mean “serving as an example, example, or illustration.” Any embodiment described as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments.

[0053] The terms "first" and "second" are used for descriptive purposes only and should not be construed as explicitly or implicitly indicating relative importance or the number of the technical features indicated. Therefore, features specified as "first" or "second" may explicitly or implicitly include one or more of such features. In the description of the embodiments of this application, unless otherwise specified, "plurality" means two or more.

[0054] The following explains some of the terms used in the embodiments of the present application to facilitate understanding by those skilled in the art.

[0055] Three-dimensional structure of protein: Protein is generally composed of dozens to thousands of amino acids, each amino acid is composed of hydrogen, carbon, nitrogen, oxygen and sulfur atoms. The three-dimensional structure of protein is determined by the three-dimensional coordinates of all its atoms in space.

[0056] Cryo-electron microscopy electron density map: Cryo-electron microscopy is one of the three mainstream methods for determining the three-dimensional structure of proteins by experimental means (the other two are nuclear magnetic resonance and X-ray crystallography). The experimental result is a three-dimensional grid of space with electron density values ​​at all grid points, which is called an electron density map (which can be understood as N x ×N y ×N z 3D tensor of .

[0057] Electron density, also known as electron beam density, indicates the probability of finding electrons at a specific location around an atom or molecule. Electrons are generally more likely to be found in areas of high electron density. Atoms or groups with lower electron density indicate that some aspect of the molecular structure is displacing negative charge. When observing materials with a transmission electron microscope, denser areas of materials with strong electron scattering properties appear darker; these areas are generally referred to as having high electron density. Electron density maps are the three-dimensional distribution of electron density in a crystal. Contour surfaces are often used to visualize electron density.

[0058] MPNN (Message Passing Neural Network) model: Strictly speaking, MPNN is not a model, but a framework. To demonstrate that models applied to chemical prediction tasks can directly learn molecular features from molecular graphs and are not affected by graph isomorphism, we call this supervised learning framework applied to graphs MPNN. This framework abstracts some common features from currently popular neural network models that support graph data, with the goal of understanding the relationships between them.

[0059] MLP (Multi-layer Perceptron) model: It is a forward-structured artificial neural network that maps a set of input vectors to a set of output vectors. MLP can be regarded as a directed graph consisting of multiple node layers, and each layer is fully connected to the next layer. Except for the input node, each node is a neuron with a nonlinear activation function. The MLP is trained using the supervised learning method of the BP back-propagation algorithm. MLP is a generalization of the perceptron, which overcomes the weakness of the perceptron that it cannot recognize linearly inseparable data. The most typical MLP consists of three layers: input layer, hidden layer and output layer, and different layers are fully connected (fully connected means that any neuron in the previous layer is connected to all neurons in the next layer).

[0060] The present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0061] Living organisms contain a vast array of diverse molecules, such as proteins, carbohydrates, and lipids. These molecules possess diverse physical and chemical properties and undergo complex interactions and biochemical reactions in a wide variety of ways. Molecules are multi-particle systems composed of atomic nuclei and electrons. These particles interact in complex ways, including Coulomb interactions between atoms, electrons, and electrons and nuclei, as well as spin-spin and spin-orbit interactions. These interactions determine the motion of atoms and electrons and, consequently, the properties of molecules.

[0062] Electron density represents the probability of finding an electron at a specific location around an atom or molecule. Its value is influenced by factors such as atoms, interactions between atoms, interactions between electrons, and interactions between electrons and atoms. Cryo-electron microscopy is an experimental method for determining the three-dimensional structure of proteins. The experimental result is a three-dimensional grid of space with electron density values ​​at all grid points, which is collectively called an electron density map.

[0063] Existing methods often simulate and generate corresponding cryo-EM electron density maps from protein 3D structures based on a single Gaussian assumption. Specifically, these methods assume that the electron density in the neighborhood of an atom follows a Gaussian distribution related to the distance to the atom center, that is:

[0064] ρ c (x g |x i )=α i ·e(-β‖x g -x i ‖ 2 )...Formula 1

[0065] Among them, ρ c (x g |x i ) is the grid point x g Considering only the atom x i The electron density when it affects it, β=[π / (2.4+0.8R0)] 2 , R0 is the resolution of the electron density map, α i =m i (β / π) 1.5 , m i is atom x i For the grid point x g , its final electron density is determined by all atoms in its neighborhood under a certain distance threshold, that is:

[0066]

[0067] in, is the grid point x g The set of all atoms in the neighborhood.

[0068] A major problem with these methods is that a single Gaussian assumption is insufficient to fully fit the correspondence between a protein's three-dimensional structure and its electron density map. It also fails to fully account for interatomic interactions and the influence of electron density at different grid points. For example, if a sulfur atom can form a disulfide bond with another nearby sulfur atom, the corresponding electron density distribution will be significantly different from that in the absence of a disulfide bond. Existing methods do not account for this issue.

[0069] To address the low accuracy of electron density map generation in the aforementioned methods, the present invention provides a method, apparatus, electronic device, and storage medium for determining an electron density map. The present invention relates to artificial intelligence (AI) and machine learning technologies, and is designed based on computer vision (CV) and machine learning (ML) within AI.

[0070] Artificial intelligence (AI) refers to the theories, methods, techniques, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, to perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive field of computer science that seeks to understand the essence of intelligence and produce new intelligent machines that can respond in a manner similar to human intelligence. AI also studies the design principles and implementation methods of various intelligent machines, enabling them to possess the capabilities of perception, reasoning, and decision-making. AI technologies primarily encompass computer vision, speech processing, and machine learning / deep learning.

[0071] With the research and advancement of artificial intelligence technology, artificial intelligence has been studied and applied in many fields, such as common smart homes, image retrieval, video surveillance, smart speakers, smart marketing, unmanned driving, automatic driving, drones, robots, smart medical care, etc. It is believed that with the development of technology, artificial intelligence will be applied in more fields and play an increasingly important role.

[0072] Computer vision technology is a key application of artificial intelligence. It studies related theories and technologies, attempting to build AI systems that can extract information from images, videos, or multidimensional data to replace human visual interpretation. Typical computer vision technologies typically include image processing and video analysis.

[0073] Machine learning is a multi-disciplinary interdisciplinary subject involving probability theory, statistics, approximation theory, convex analysis, algorithmic complexity theory and other disciplines. It specializes in studying how computers simulate or implement human learning behavior to acquire new knowledge or skills, and reorganize existing knowledge structures to continuously improve their own performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications are spread across all areas of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning and other technologies. In the process of generating cryo-electron microscopy electron density maps, the embodiment of the present application uses a deep graph neural network model to learn the heterogeneous graph structure composed of atoms and grid points, and calculates the electron density of each grid point through a multi-layer perceptron.

[0074] The embodiment of the present application introduces the idea of ​​deep graph learning, by modeling the set of atoms in the three-dimensional space where the protein is located and the set of grid points in the corresponding electron density map as a heterogeneous graph structure, and then using the deep graph model for the heterogeneous graph structure to explicitly model and learn the interactions between atoms and the mutual influence between grid points, thereby improving the generation accuracy of cryo-electron microscopy electron density maps generated from the three-dimensional structure of the protein. It should be noted that the electron density map determination method in the embodiment of the present application is not only applicable to the electron density map of proteins, but also to the generation of electron density maps of other molecules such as nucleic acids, lipids, carbohydrates, etc. The embodiment of the present application is only illustrated by taking proteins as an example.

[0075] An application scenario of the method for determining the electron density map provided in the embodiment of the present application can be found in Figure 1 , which is a schematic diagram of the application architecture of the method for determining an electron density map in an embodiment of the present application, includes a server 100 and a terminal device 200.

[0076] The terminal device 200 is an electronic device that can install various applications and display the running interface of the installed applications. The electronic device can be mobile or fixed, for example, a mobile phone, a tablet computer, various wearable devices, a vehicle-mounted device, or other electronic devices capable of implementing the above functions.

[0077] The terminal device 200 and the server 100 can be connected via the Internet to enable communication between them. Optionally, the above-mentioned Internet uses standard communication technologies and / or protocols. The Internet is typically the Internet, but it can also be any network, including but not limited to a local area network (LAN), a metropolitan area network (MAN), a wide area network (WAN), a mobile, wired or wireless network, a private network or any combination of a virtual private network. In some embodiments, technologies and / or formats including Hypertext Markup Language (HTML) and Extensible Markup Language (XML) are used to represent data exchanged over the network. In addition, conventional encryption technologies such as Secure Socket Layer (SSL), Transport Layer Security (TLS), Virtual Private Network (VPN), and Internet Protocol Security (IPsec) can also be used to encrypt all or some links. In other embodiments, customized and / or dedicated data communication technologies can also be used to replace or supplement the above-mentioned data communication technologies.

[0078] The server 100 can provide various network services for the terminal device 200, wherein the server 100 can be a single server, a server cluster consisting of several servers, or a cloud computing center.

[0079] Specifically, the server 100 may include a processor 110 (Center Processing Unit, CPU), a memory 120, an input device 130 and an output device 140, etc. The input device 130 may include a keyboard, a mouse, a touch screen, etc., and the output device 140 may include a display device, such as a liquid crystal display (LCD), a cathode ray tube (CRT), etc.

[0080] The memory 120 may include a read-only memory (ROM) and a random access memory (RAM), and provides program instructions and data stored in the memory 120 to the processor 110. In an embodiment of the present invention, the memory 120 may be used to store the program of the electron density map determination method in an embodiment of the present invention.

[0081] The processor 110 calls the program instructions stored in the memory 120 , and the processor 110 is configured to execute the steps of any one of the electron density map determination methods in the embodiments of the present invention according to the obtained program instructions.

[0082] It should be noted that, in the embodiment of the present invention, the electron density map determination method is mainly executed by the server 100 side. For example, with respect to the electron density map determination method, the terminal device 200 can obtain cryo-electron microscopy experimental data and the corresponding protein three-dimensional structure from the database, and send them to the server 100. The server 100 generates training and test data for model training, trains the electron density map generation model, and returns the training results to the terminal device 200. Figure 1 The application architecture shown is explained by taking the application on the server 100 side as an example. Of course, the electron density map determination method in the embodiment of the present invention can also be executed by the terminal device 200. For example, the terminal device 200 can obtain a trained electron density map generation model from the server 100 side, and then generate an electron density map of the protein based on the electron density map generation model. This is not limited in the embodiment of the present invention.

[0083] In addition, the application architecture diagram in the embodiment of the present invention is intended to more clearly illustrate the technical solution in the embodiment of the present invention, and does not constitute a limitation on the technical solution provided in the embodiment of the present invention. Of course, it is not limited to the application of biological macromolecules. For other application architectures and business applications, the technical solution provided in the embodiment of the present invention is also applicable to similar problems.

[0084] The various embodiments of the present invention are applied to Figure 1 The application architecture diagram shown is used as an example for schematic description.

[0085] Figure 2 FIG. 1 is a flow chart showing a method for determining an electron density map according to an embodiment of the present application. Figure 2 As shown, the method includes the following steps:

[0086] Step S201 : constructing a three-dimensional coordinate system based on the molecular structure of the target object and performing three-dimensional gridding to obtain grid points corresponding to the target object.

[0087] The molecular structure includes atoms of the target object, and the grid points are vertices corresponding to each grid after the three-dimensional coordinate system is divided into a number of three-dimensional grids.

[0088] Let's take protein as an example. The molecular structure of a protein refers to the spatial arrangement of the molecule. Protein molecules contain multiple atoms, primarily carbon, hydrogen, oxygen, and nitrogen. They are an important class of biological macromolecules. Protein molecules are covalent polypeptide chains formed by the end-to-end condensation of amino acids. However, natural protein molecules are not random, loose polypeptide chains. Each natural protein has its own unique spatial structure, or three-dimensional structure, which is often referred to as the protein's molecular structure.

[0089] During the specific implementation process, the three-dimensional structure of the protein is determined by cryo-electron microscopy. Cryo-electron microscopy freezes the biological molecules in motion, constructs a three-dimensional coordinate system for the three-dimensional structure of the protein, and performs three-dimensional gridding. Figure 3 A set of cryo-electron microscopy electron density maps and a schematic diagram of the corresponding protein three-dimensional structure are shown, where the solid circles represent some atoms in the protein. A three-dimensional coordinate system is established based on the three-dimensional structure of the protein. Figure 3 The three-dimensional coordinate system is gridded by dotted lines, and the intersection of the dotted lines, that is, the vertex corresponding to each grid, is used as the grid point. Figure 3 Indicated by a dotted circle.

[0090] In step S202 , each atom and each network point is regarded as a node, and based on the three-dimensional coordinates of each atom and the three-dimensional coordinates of each grid point, the node features of each node corresponding to the target object and the edge features between every two nodes are determined.

[0091] Among them, the node feature is used to characterize the elements that affect the electron density due to the node, and the edge feature is used to characterize the elements that affect the electron density due to the interaction between nodes.

[0092] In the specific implementation process, the atoms and grid points in the three-dimensional grid structure are regarded as nodes, such as Figure 3 The solid dots and dotted circles shown in the figure are all used as nodes to determine the node characteristics of each node.

[0093] Among them, the three-dimensional coordinates of an atom can use the position of the nucleus or the center position of the atom as the position of the atom to determine the coordinates of the position in the three-dimensional coordinate system; the three-dimensional coordinates of the grid points are the coordinates of the vertices corresponding to each grid in the three-dimensional coordinate system.

[0094] Since nodes have an impact on electron density, node features are used to characterize the elements whose nodes affect electron density. Specific node features can include atomic mass, element type, three-dimensional coordinates of atoms, three-dimensional coordinates of grid points, etc. These data can be directly obtained from cryo-electron microscopy experimental data.

[0095] On the other hand, the interactions between nodes, such as the interactions between atoms and the interactions between electron densities at different grid points, can be characterized using edge features. These edge features include edge features between individual atoms, edge features between individual grid points, and edge features between atoms and grid points. Specifically, edge features can be the distance between two atoms, the relative coordinates between two atoms, the distance between an atom and a grid point, the relative coordinates from an atom to a grid point, the distance between two grid points, the relative coordinates between two grid points, and so on. These edge features can be calculated based on experimental data from cryo-electron microscopy.

[0096] In step S203 , each node feature and each edge feature is input into the trained electron density map generation model to obtain the electron density of each grid point, where the electron density of each grid point represents the probability of finding an electron at the corresponding grid point.

[0097] During the specific implementation process, based on the characteristic data extracted or calculated in step S201 and step S202, the electron density of each grid point is determined using the electron density map generation model. The electron density is the probability of finding an electron at the corresponding grid point position.

[0098] The electron density map generation model is trained using training and test data, which also include node and edge features of proteins. Furthermore, the training and test data also include the experimental electron density of grid points. This means that the experimental results obtained through cryo-electron microscopy experiments contain the experimental electron density of each grid point. This allows the experimental electron density to be used as the true value to train the electron density map generation model, resulting in a trained electron density map generation model.

[0099] In this way, by modeling the atomic set in the three-dimensional space where the target object is located and the corresponding grid point set in the electron density map as a heterogeneous graph structure, and then using the algorithm model to model and learn the interactions between atoms, the interactions between atoms and grid points, and the interactions between grid points, the interactions between atoms and the electron densities of different grid points are fully considered, which can better simulate the electron density map and improve the accuracy and precision generated from the electron density map.

[0100] In a preferred embodiment, since the interactions between atoms, between grid points, and between atoms and grid points are greatly affected by distance, the characteristic data determined based on the above-mentioned cryo-electron microscopy experimental results can be obtained within the grid area.

[0101] For each grid point, perform the following operations:

[0102] Determine a neighborhood of a grid point among the grid points; the neighborhood represents the space within a set distance around a grid point;

[0103] Identify each atom in the neighborhood and each other grid point in the neighborhood;

[0104] Based on the measurement data, the grid features of a grid point, the atomic features of each atom in the neighborhood, and the grid features of each other grid point in the neighborhood are determined, and the obtained grid features and atomic features are used as node features in the neighborhood;

[0105] Based on the three-dimensional coordinates of a grid point, the three-dimensional coordinates of each atom in the neighborhood, and the three-dimensional coordinates of each other grid point in the neighborhood, the edge characteristics between each two nodes in the neighborhood are determined.

[0106] In the specific implementation process, whether it is training data or prediction data, feature data is obtained or determined within the range of a grid point. Specifically, for each set of cryo-EM electron density maps and the corresponding protein three-dimensional structure, each grid point in the electron density map is taken as the center, and a preset distance threshold (e.g., 5A, i.e., 5×10 -10 m) Divide the neighborhood, and then count all atoms in the neighborhood of the grid point and the related information of the grid point.

[0107] Among them, the node features include the grid features of the grid point (such as the three-dimensional coordinates of the grid point), the atomic features of each atom in the field (such as atomic mass, three-dimensional coordinates of the atom, etc.), and the grid features of each other grid point in the neighborhood (such as the three-dimensional coordinates of other grid points). These data can be obtained directly from the measurement data or determined based on the measurement data.

[0108] On the other hand, edge features between nodes in a neighborhood can include edge features between atoms in the neighborhood, edge features between the grid point and other grid points in the neighborhood, and edge features between the grid point and atoms in the neighborhood. These edge features are calculated based on the 3D coordinates of the grid point, the 3D coordinates of atoms in the neighborhood, and the 3D coordinates of other grid points in the neighborhood.

[0109] Furthermore, the edge features between atoms in the neighborhood include: the distance between every two atoms in the neighborhood, and the relative coordinates between every two atoms in the neighborhood. Here, the edge features between atoms can be calculated based on the three-dimensional coordinates of each atom in the neighborhood.

[0110] The edge features between the grid point and each other grid point in the neighborhood include: the distance between the grid point and each other grid point in the neighborhood, and the relative coordinates between the grid point and each other grid point in the neighborhood. Here, the edge features between the grid point and each other grid point in the neighborhood can be calculated based on the three-dimensional coordinates of the grid point and the three-dimensional coordinates of each other grid point in the neighborhood.

[0111] The edge features between the grid point and each atom in the neighborhood include: the distance between the grid point and each atom in the neighborhood, and the relative coordinates between the grid point and each atom in the neighborhood. Here, the edge features between the grid point and each atom in the neighborhood can be calculated based on the three-dimensional coordinates of the grid point and the three-dimensional coordinates of each atom in the neighborhood.

[0112] Below Figure 3 The grid points in are used as an example to introduce. Figure 3 Each grid point in the grid is taken as the center, and the neighborhood is divided according to the preset distance threshold (for example, 5A). Then, the relevant information of all atoms in the neighborhood is counted (including element type, atomic mass, relative coordinates of atoms to grid points, and distances from atoms to grid points). Figure 3 The cryo-electron microscopy electron density map and the corresponding protein three-dimensional structure, the final extracted data include:

[0113] 1. Experimental electron density at all grid points;

[0114] 2. The three-dimensional coordinates of all grid points;

[0115] 3. Relevant information of all atoms in the neighborhood of each grid point (element type, atomic mass, relative coordinates of the atom to the grid point, distance between the atom and the grid point, etc.);

[0116] 4. All relevant information between two atoms whose distance is less than the preset interatomic distance threshold (element type, atomic mass, interatomic distance and atomic coordinates of the two atoms, etc.).

[0117] Among them, data 2, 3, and 4 can be used as prediction data, and data 1 connected with data 2, 3, and 4 constitute the training data for the electron density map generation model.

[0118] It should be noted that to facilitate subsequent training and calculations, all data corresponding to a grid point, namely the experimental electron density at that grid point, the three-dimensional coordinates of that grid point, and all relevant feature data within the grid point area, can be stored as a set of data. In this way, when training or prediction is required, data can be directly extracted according to the grid point. Feature data that is repeated between different grid points can be obtained or calculated repeatedly, or it can be calculated only once without repeated calculation.

[0119] by Figure 3 Taking the grid point Y in as an example, the neighborhood of the grid point Y is shown as the solid circle in the figure. Figure 3 The radius of the neighborhood of the grid point Y shown is smaller than the distance between two adjacent grid points. Therefore, the neighborhood of the grid point Y contains only atoms and no other grid points. Get the three-dimensional coordinates of the grid point Y; Get Figure 3 The atomic characteristics of each atom in the solid circle, including element type and atomic mass; obtain Figure 3 The edge features of each atom in the solid circle to the grid point Y, including the relative coordinates of each atom to the grid point Y and the distance between each atom and the grid point Y; obtain Figure 3 The edge features of the atomic pairs whose distances in the solid circle are within the interatomic distance threshold, including the interatomic distances and the relative coordinates between atoms, are used as the feature data corresponding to the grid point Y.

[0120] In addition, during the training process, it is also necessary to obtain the experimental electron density of grid point Y, and use the experimental electron density of grid point Y and the corresponding characteristic data as the training data of grid point Y.

[0121] In addition, the radius of the neighborhood of a grid point may also be greater than the distance between two adjacent grid points. In this case, the feature data corresponding to the grid point Y also includes node features and edge features related to other grid points in the neighborhood.

[0122] Based on the acquired feature data, a network model can be constructed based on deep graph learning to generate a cryo-EM electron density map from a protein's three-dimensional structure. In one optional embodiment, the electron density map generation model includes N feature update layers and an electron density prediction layer, where the feature update layer is a deep graph neural network and N is a positive integer.

[0123] In a specific embodiment, the feature update layer is an information transfer neural network. Figure 4 Figure 2 shows a schematic diagram of the structure of the electron density map generation model. Figure 4 As shown, the electron density map generation model includes N information transfer neural network layers ( Figure 4 MPNN Layer) and electron density prediction layer ( Figure 4 In the MLP Network, each node feature and each edge feature is input into the trained electron density map generation model to obtain the electron density of each grid point, including:

[0124] For each grid point, perform the following operations:

[0125] Input each node feature and each edge feature in the neighborhood of a network point in each grid point into the trained electron density map generation model;

[0126] Based on N feature update layers, each node feature and each edge feature is subjected to multi-layer nonlinear transformation to obtain the corresponding target node feature and target edge feature;

[0127] Based on the electron density prediction layer, the obtained target node features and target edge features are calculated to obtain the electron density of a grid point.

[0128] In the specific implementation process, the node features and edge features corresponding to a certain network point are input into the electron density map generation model, and the electron density of the network point is output through the processing of the information transmission neural network layer and the electron density prediction layer in the electron density map generation model. Figure 4 As shown, for the grid points Y, V a is the atomic characteristic, V g is the grid feature, E a,a is the edge feature between atoms, E a,g is the edge feature between atoms, E g,g is the edge feature between grid points. The above features are input into the electron density map generation model, and the electron density of grid point Y is output.

[0129] Furthermore, since the electron density map generation model contains multiple layers of information transfer neural network layers, each layer of information transfer neural network layers performs nonlinear transformation on the node features. For example, the atomic features of the initial input electron density map generation model are The grid features are After processing through a layer of information transfer neural network layer, the atomic features are obtained and the grid characteristics are ...After all the information is passed through the neural network layer, the final target node feature is obtained and target edge features

[0130] In addition, the electron density map generation model in the embodiment of the present application adopts an information transfer neural network model to learn the heterogeneous graph structure composed of atoms and grid points. In other optional implementation processes, the information transfer neural network model can also be replaced by other deep graph neural network models, such as GCN (Graph Convolutional Network), GAN (Graph Attention Network), etc.

[0131] The following describes the training process of the electron density map generation model. The specific model training includes the following steps:

[0132] Obtain training samples and experimental electron density corresponding to the training samples;

[0133] The electron density map generation model is iteratively trained based on the training samples and the experimental electron density corresponding to the training samples until the set training end condition is reached, thereby obtaining a trained electron density map generation model. One iterative process includes:

[0134] Inputting the training samples into the electron density map generation model to determine the training electron density of the training samples;

[0135] Determine the loss function based on the experimental electron density and the training electron density;

[0136] Parameters of the electron density map generation model are adjusted according to the loss function.

[0137] The training process requires not only training samples at the grid points but also the experimental electron density corresponding to those grid points. This experimental electron density is used as the true electron density at the grid points and compared with the training electron density obtained using the training samples to determine the loss function. The model parameters are then adjusted based on the loss function until the set training result condition is reached.

[0138] The specific training process can be as follows Figure 5 As shown, for a certain grid point, the initial atomic features are obtained through multiple MPNN layers. and mesh features Multi-layer nonlinear transformations were performed to obtain the final target atomic features. and target mesh features Then use the MLP model to calculate and obtain the training electron density of the grid point Will train electron density The experimental electron density ρ at this grid point g Calculate the mean square error (MSE) as the loss function of the model to train and update the model parameters.

[0139] Specifically, the MPNN layer updates the atomic features and grid features using the following formula:

[0140]

[0141]

[0142] Among them, M t (·) collects the node features of all atoms and grid points in the neighborhood of the atom (or grid point), U t (·) Based on the collected information, the characteristics of atoms (or grid points) are updated, M t (·) and U t (·) are all implemented in the form of neural networks.

[0143] Through the electron density map determination method proposed in the embodiment of the present application, it is possible to more accurately screen out candidate protein three-dimensional structures that are more consistent with experimental data, and determine the subsequent optimization and adjustment direction for the candidate protein three-dimensional structure, thereby more accurately analyzing the protein three-dimensional structure from the cryo-electron microscopy electron density map.

[0144] Figure 6 The figure shows the accuracy comparison of different cryo-electron microscopy electron density map generation methods. The horizontal axis represents two generation methods based on the single Gaussian hypothesis, and the vertical axis represents the electron density map determination method of the embodiment of the present application. The numerical value represents the consistency with the electron density map obtained by the experiment, and the higher the better. Figure 6 As shown, both the left and right figures are electron density maps of the embodiment of the present application with higher accuracy.

[0145] The method for determining the three-dimensional structure of a protein from a cryo-electron microscopy electron density map proposed in the embodiments of the present application can generate an electron density map with higher precision than the approximate method based on the single Gaussian hypothesis, and the generated electron density map is also more accurate.

[0146] Corresponding to the above method embodiment, the embodiment of the present application further provides a device for determining an electron density map. Figure 7 Schematic diagram of the structure of the device for determining the electron density map provided in the embodiment of the present application; Figure 7 As shown, the device for determining the electron density map includes:

[0147] A construction unit 701 is configured to construct a three-dimensional coordinate system based on the molecular structure of the target object and perform three-dimensional gridding to obtain grid points corresponding to the target object, wherein the molecular structure contains atoms of the target object, and the grid points are vertices corresponding to each grid after the three-dimensional coordinate system is divided into a plurality of three-dimensional grids;

[0148] A feature unit 702 is configured to treat each atom and each grid point as a node, and determine, based on the three-dimensional coordinates of each atom and the three-dimensional coordinates of each grid point, a node feature of each node corresponding to the target object, and an edge feature between every two nodes, wherein the node feature is used to characterize the elements of the node that affect the electron density, and the edge feature is used to characterize the elements of the interaction between nodes that affect the electron density;

[0149] The model unit 703 is used to input the node features and the edge features into the trained electron density map generation model to obtain the electron density of each grid point, wherein the electron density of each grid point represents the probability of finding an electron at the corresponding grid point.

[0150] In an optional embodiment, the feature unit 702 is specifically configured to:

[0151] For each grid point, perform the following operations:

[0152] Determine a neighborhood of one of the grid points; the neighborhood represents a space within a set distance around the one grid point;

[0153] determining each atom within the neighborhood and each other grid point within the neighborhood;

[0154] Determining, based on the measurement data, a grid feature of the one grid point, an atomic feature of each atom in the neighborhood, and a grid feature of each other grid point in the neighborhood, and using the obtained grid features and atomic features as node features in the neighborhood;

[0155] Based on the three-dimensional coordinates of the one grid point, the three-dimensional coordinates of each atom in the neighborhood, and the three-dimensional coordinates of each other grid point in the neighborhood, edge features between every two nodes in the neighborhood are determined.

[0156] In an optional embodiment, the edge features within the neighborhood include any one or any combination of the following:

[0157] The edge features between the atoms in the neighborhood, the edge features between the one grid point and each other grid point in the neighborhood, and the edge features between the one grid point and each atom in the neighborhood.

[0158] In an optional embodiment, the edge features between atoms in the neighborhood include: the distance between every two atoms in the neighborhood, and the relative coordinates between every two atoms in the neighborhood;

[0159] The edge features between the one grid point and each other grid point in the neighborhood include: the distance between the one grid point and each other grid point in the neighborhood, and the relative coordinates between the one grid point and each other grid point in the neighborhood;

[0160] The edge features between the grid point and each atom in the neighborhood include: the distance between the grid point and each atom in the neighborhood, and the relative coordinates between the grid point and each atom in the neighborhood.

[0161] In an optional embodiment, the electron density map generation model includes N feature update layers and electron density prediction layers, the feature update layer is a deep graph neural network, and N is a positive integer.

[0162] In an optional embodiment, the model unit 703 is specifically configured to:

[0163] For each grid point, perform the following operations:

[0164] Inputting each node feature and each edge feature in the neighborhood of one of the grid points into the trained electron density map generation model;

[0165] Based on the N feature update layers, performing multi-layer nonlinear transformation on each of the node features and each of the edge features to obtain corresponding target node features and target edge features;

[0166] Based on the electron density prediction layer, the obtained target node features and target edge features are calculated to obtain the electron density of the one grid point.

[0167] In an optional embodiment, the method further includes a training unit 704, configured to train the electron density map generation model according to the following process:

[0168] Obtaining training samples and experimental electron density corresponding to the training samples;

[0169] The electron density map generation model is iteratively trained based on the training samples and the experimental electron densities corresponding to the training samples until a set training end condition is reached, thereby obtaining a trained electron density map generation model. One iterative process includes:

[0170] Inputting the training sample into an electron density map generation model to determine the training electron density of the training sample;

[0171] Determining a loss function according to the experimental electron density and the training electron density;

[0172] Parameters of the electron density map generation model are adjusted according to the loss function.

[0173] In an optional embodiment, the feature update layer is an information transfer neural network.

[0174] Corresponding to the above method embodiment, an embodiment of the present application also provides an electronic device.

[0175] Figure 8 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application; Figure 8As shown, in the embodiment of the present application, the electronic device 80 includes: a processor 81, a display 82, a memory 83, an input device 86, a bus 85 and a communication device 84; the processor 81, the memory 83, the input device 86, the display 82 and the communication device 84 are all connected through the bus 85, and the bus 85 is used to transmit data between the processor 81, the memory 83, the display 82, the communication device 84 and the input device 86.

[0176] Among them, the memory 83 can be used to store software programs and modules, such as the program instructions / modules corresponding to the image classification method in the embodiment of the present application. The processor 81 executes various functional applications and data processing of the electronic device 80 by running the software programs and modules stored in the memory 83, such as the image classification method provided in the embodiment of the present application. The memory 83 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, at least one application application, etc.; the data storage area may store data created according to the use of the electronic device 80 (such as training samples, feature extraction networks, etc.). In addition, the memory 83 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0177] The processor 81 is the control center of the electronic device 80. It connects the various components of the electronic device 80 using a bus 85 and various interfaces and lines. It executes or runs software programs and / or modules stored in the memory 83 and calls data stored in the memory 83 to perform various functions of the electronic device 80 and process data. Optionally, the processor 81 may include one or more processing units, such as a CPU, a GPU (Graphics Processing Unit), a digital processing unit, etc.

[0178] In the embodiment of the present application, the processor 81 displays the image to the user through the display 82.

[0179] The input device 86 is primarily used to obtain user input operations. Depending on the electronic device, the input device 86 may also vary. For example, if the electronic device is a computer, the input device 86 may be a mouse, keyboard, or other input device; if the electronic device is a portable device such as a smartphone or tablet, the input device 86 may be a touch screen.

[0180] An embodiment of the present application further provides a computer storage medium, in which computer executable instructions are stored. The computer executable instructions are used to implement the method for determining the electron density map described in any embodiment of the present application.

[0181] In some possible embodiments, various aspects of the method for determining an electron density map provided in the present application may also be implemented in the form of a program product, which includes program code. When the program product is run on a computer device, the program code is used to cause the computer device to perform the steps of the method for determining an electron density map according to various exemplary embodiments of the present application described above in this specification. For example, the computer device may perform the following steps: Figure 2 The flow of determining the electron density map in steps S201 to S203 is shown.

[0182] The program product may employ any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or component, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0183] A readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries readable program code. Such propagated data signals may take a variety of forms, including, but not limited to, electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium that can transmit, propagate, or transfer a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0184] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0185] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0186] In addition, all functional units in the embodiments of the present application can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the above-mentioned integrated units can be implemented in the form of hardware or in the form of hardware plus software functional units.

[0187] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. A method for determining an electron density map, characterized in that: The method comprises: Based on the molecular structure of the target object, a three-dimensional coordinate system is constructed and three-dimensional gridding is performed to obtain grid points corresponding to the target object, wherein the molecular structure contains atoms of the target object, and the grid points are vertices corresponding to each grid after the three-dimensional coordinate system is divided into a plurality of three-dimensional grids; A heterogeneous graph is constructed by taking each atom and each network point as a node, and determining, based on the three-dimensional coordinates of each atom and the three-dimensional coordinates of each grid point, node features of each node corresponding to the target object and edge features between every two nodes, wherein the node features are used to characterize the elements of the node that affect the electron density, and the edge features are used to characterize the elements of the interaction between nodes that affect the electron density; the edge features include edge features between atoms, edge features between grid points, and edge features between atoms and grid points; Each node feature and each edge feature is input into a trained electron density map generation model to obtain the electron density of each grid point, where the electron density of each grid point represents the probability of finding an electron at the corresponding grid point. The electron density map generation model is obtained by modeling and learning the interactions between atoms, the interactions between atoms and grid points, and the interactions between grid points. The electron density map generation model is a deep graph neural network model.

2. The method according to claim 1, characterized in that The steps of taking each atom and each grid point as a node and determining the node features of each node corresponding to the target object and the edge features between each two nodes based on the three-dimensional coordinates of each atom and the three-dimensional coordinates of each grid point include: For each grid point, perform the following operations: Determine a neighborhood of one of the grid points; the neighborhood represents a space within a set distance around the one grid point; determining each atom within the neighborhood and each other grid point within the neighborhood; Determining, based on the measurement data, a grid feature of the one grid point, an atomic feature of each atom in the neighborhood, and a grid feature of each other grid point in the neighborhood, and using the obtained grid features and atomic features as node features in the neighborhood; Based on the three-dimensional coordinates of the one grid point, the three-dimensional coordinates of each atom in the neighborhood, and the three-dimensional coordinates of each other grid point in the neighborhood, edge features between every two nodes in the neighborhood are determined.

3. The method according to claim 2, characterized in that The edge features within the neighborhood include any one or any combination of the following: The edge features between the atoms in the neighborhood, the edge features between the one grid point and each other grid point in the neighborhood, and the edge features between the one grid point and each atom in the neighborhood.

4. The method according to claim 3, characterized in that The edge features between atoms in the neighborhood include: the distance between every two atoms in the neighborhood, and the relative coordinates between every two atoms in the neighborhood; The edge features between the one grid point and each other grid point in the neighborhood include: the distance between the one grid point and each other grid point in the neighborhood, and the relative coordinates between the one grid point and each other grid point in the neighborhood; The edge features between the grid point and each atom in the neighborhood include: the distance between the grid point and each atom in the neighborhood, and the relative coordinates between the grid point and each atom in the neighborhood.

5. The method according to claim 2, 3 or 4, characterized in that The electron density map generation model includes N feature update layers and electron density prediction layers, the feature update layer is a deep graph neural network, and N is a positive integer.

6. The method according to claim 5, characterized in that Inputting each node feature and each edge feature into the trained electron density map generation model to obtain the electron density of each grid point includes: For each grid point, perform the following operations: Inputting each node feature and each edge feature in the neighborhood of one of the grid points into the trained electron density map generation model; Based on the N feature update layers, performing multi-layer nonlinear transformation on each of the node features and each of the edge features to obtain corresponding target node features and target edge features; Based on the electron density prediction layer, the obtained target node features and target edge features are calculated to obtain the electron density of the one grid point.

7. The method according to claim 5, characterized in that The training of the electron density map generation model includes the following steps: Obtaining training samples and experimental electron density corresponding to the training samples; The electron density map generation model is iteratively trained based on the training samples and the experimental electron densities corresponding to the training samples until a set training end condition is reached, thereby obtaining a trained electron density map generation model. One iterative process includes: Inputting the training sample into an electron density map generation model to determine the training electron density of the training sample; Determining a loss function according to the experimental electron density and the training electron density; Parameters of the electron density map generation model are adjusted according to the loss function.

8. The method according to claim 5, characterized in that The feature update layer is an information transfer neural network.

9. A device for determining an electron density map, characterized in that: The device comprises: a construction unit for constructing a three-dimensional coordinate system and performing three-dimensional gridding based on the molecular structure of the target object to obtain grid points corresponding to the target object, wherein the molecular structure contains atoms of the target object, and the grid points are vertices corresponding to each grid after the three-dimensional coordinate system is divided into a plurality of three-dimensional grids; a feature unit for constructing a heterogeneous graph using each of the atoms and each of the network points as a node, and determining, based on the three-dimensional coordinates of each atom and the three-dimensional coordinates of each grid point, node features of each node corresponding to the target object, and edge features between every two nodes, wherein the node features are used to characterize the elements of the node that affect the electron density, and the edge features are used to characterize the elements of the interaction between nodes that affect the electron density; the edge features include edge features between atoms, edge features between grid points, and edge features between atoms and grid points; A model unit is used to input each node feature and each edge feature into a trained electron density map generation model to obtain the electron density of each grid point, wherein the electron density of each grid point represents the probability of finding an electron at the corresponding grid point; the electron density map generation model is obtained by modeling and learning the interactions between atoms, the interactions between atoms and grid points, and the interactions between grid points, and the electron density map generation model is a deep graph neural network model.

10. A computer-readable storage medium storing a computer program, wherein: When the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

11. An electronic device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the computer program is executed by the processor, the processor implements the method according to any one of claims 1 to 8.

Images