Drug screening model construction method and device, screening method, equipment and medium

Abstract

The application provides a drug screening model construction method and device, a screening method, equipment and a medium, and belongs to the field of drug screening. The method comprises the following steps: obtaining a drug training set, wherein the drug training set comprises a chemical formula of a drug protein, a chemical formula of a small molecule and a classification label; based on the chemical formula, an initial graph network of the drug protein and the small molecule is drawn, wherein atoms are nodes and chemical bonds are edges connecting the nodes; a random initialization vector is used to identify the weight vector of each node in the initial graph network; each node of the initial graph network is reconstructed according to the connection relationship of the initial graph network, to obtain a reconstructed graph network, and the reconstruction step is repeated to obtain at least two layers of reconstructed graph networks; and the initial graph network and the at least two layers of reconstructed graph networks are subjected to deep learning according to the classification label, to construct a drug screening model. Through the processing scheme of the present application, the properties of drug compounds in biological and chemical experiments of different scales can be accurately predicted.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of drug screening, in particular to a drug screening model construction method and device, a drug screening method, equipment and a medium. BACKGROUND

[0002] Drug discovery has long been a time and money consuming process, with the development of computer technology, computing methods are widely used in drug research and development, and drug virtual screening is one of the most valuable technologies. Among them, the virtual screening artificial intelligence method based on protein crystal structure uses the three-dimensional structure information of protein target to predict the binding of protein and small molecule drugs.

[0003] The common virtual screening method based on protein crystal structure often uses molecular docking method, which uses molecular dynamics and quantum chemistry method to simulate the binding process of small molecules and proteins, which involves binding site search and calculation of small molecule space position using heuristic function, and has large computing power consumption and multiple possible results, and has certain uncertainty. And the protein has high flexibility, the traditional simulation method based on fixed three-dimensional structure and the conformation and flexibility of the real physiological structure protein have large gap, and the binding state of the protein and the small molecule cannot be truly reflected, which causes large simulation error.

[0004] Therefore, it is necessary to establish a virtual screening artificial intelligence technology based on protein crystal structure which can overcome the molecular docking method, so as to improve the accuracy and speed of virtual screening. SUMMARY

[0005] Therefore, in order to overcome the shortcomings of the prior art, the present application provides a drug screening model construction method and device, a drug screening method, equipment and a medium which can accurately predict the properties of drug compounds in different scale biological and chemical experiments.

[0006] In order to achieve the above purpose, the present application provides a drug screening model construction method for protein crystal structure screening, comprising: obtaining a drug training set, the drug training set comprising chemical formula of drug protein, chemical formula of small molecule and classification label; based on the chemical formula, drawing an initial graph network of drug protein and small molecule, the atoms in the initial graph network being nodes, and the chemical bonds being edges connecting the nodes; adopting a random initialization vector to identify the weight vector of each node in the initial graph network; reconstructing each node of the initial graph network according to the connection relationship of the initial graph network to obtain a reconstructed graph network, and repeating the reconstruction step to obtain at least two layers of reconstructed graph networks; performing deep learning on the initial graph network and the at least two layers of reconstructed graph networks according to the classification label to construct a drug screening model.

[0007] In one of the embodiments, the drawing of the initial graph network of the drug protein and the small molecule based on the chemical formula comprises: performing molecular docking on the drug protein and the small molecule based on traditional molecular dynamics to obtain a target molecule; and drawing the initial graph network of the drug protein and the small molecule based on the chemical formula of the target molecule.

[0008] In one of the embodiments, the drawing of the initial graph network of the drug protein and the small molecule based on the chemical formula comprises: drawing the initial graph network of the drug protein and the small molecule based on the chemical formula respectively, wherein the initial graph network comprises an initial graph network of the drug protein and an initial graph network of the small molecule.

[0009] In one of the embodiments, the identification of the weight vector of each node in the initial graph network by using the random initialization vector comprises: generating an initialization vector by selecting a value in the interval (0, 1) based on a uniform distribution or a normal distribution, and assigning the initialization vector to the node as its weight vector.

[0010] In one of the embodiments, the reconstruction of each node of the initial graph network according to the connection relationship of the initial graph network to obtain a reconstructed graph network comprises: obtaining the weight vector of each node and the associated weight vector of the associated node connected to the node according to the connection relationship of the initial graph network; integrating the weight vector and the associated weight vector to obtain the reconstructed weight vector of the node; and generating a reconstructed graph network according to the reconstructed weight vector.

[0011] In one of the embodiments, the deep learning of the initial graph network and the at least two layers of the reconstructed graph network according to the classification label to construct a drug screening model comprises: performing feature extraction on the initial graph network and the at least two layers of the reconstructed graph network respectively to obtain hierarchical network features; and learning the hierarchical network features by using a deep learning network according to the classification label to construct a drug screening model.

[0012] In one of the embodiments, the feature extraction on the initial graph network and the at least two layers of the reconstructed graph network to obtain hierarchical network features comprises: performing feature extraction on the initial graph network and the at least two layers of the reconstructed graph network by using a function to obtain function vector features; and performing normalization operation on the function vector features in a probability space to obtain hierarchical network features.

[0013] A drug screening method, comprising: based on a chemical formula of a target protein and a small molecule, drawing a target initial graph network, atoms in the target initial graph network being nodes, and chemical bonds being edges connecting the nodes; inputting the target initial graph network into a drug screening model, and outputting an analysis result of the target protein and the small molecule, wherein the drug screening model is obtained by training the method.

[0014] A drug screening model construction device, the device comprising: a training set acquisition module for acquiring a drug training set, the drug training set comprising a chemical formula of a drug protein, a chemical formula of a small molecule, and a classification label; a graph network drawing module for drawing an initial graph network of the drug protein and the small molecule based on the chemical formula, atoms in the initial graph network being nodes, and chemical bonds being edges connecting the nodes; a vector identification module for identifying a weight vector of each of the nodes in the initial graph network using a random initialization vector; a reconstruction module for reconstructing each of the nodes of the initial graph network according to a connection relationship of the initial graph network to obtain a reconstructed graph network, and repeating the reconstruction step to obtain at least two layers of reconstructed graph networks; and a model training module for performing deep learning on the initial graph network and the at least two layers of reconstructed graph networks according to the classification label to construct a drug screening model.

[0015] A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the above method when executing the computer program.

[0016] A computer-readable storage medium having a computer program stored thereon, characterized in that the computer program is executed by a processor to implement the steps of the above method.

[0017] Compared with the prior art, the advantages of the present application are that: by modeling and analyzing the initial graph network corresponding to the chemical formula of the drug protein and the chemical formula of the small molecule, and using the initial graph network and the reconstructed graph network instead of the traditional Fingerprint-based method during modeling, only two different graph networks of the same drug are identified through machine learning, which not only effectively reduces the artificial feature design, but also expands the coverage of the features, improves the efficiency and accuracy, and greatly saves the calculation consumption of molecular dynamics, quantum mechanics, quantum chemistry, etc. through the use of big data and deep learning methods, and greatly improves the calculation speed. The present application can consider the high flexibility of the protein, and the simulation method based on the elastic graph neural network can better simulate the flexible characteristics of the protein, and more accurately predict the properties of compounds in different scales of biological and chemical experiments. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the following will briefly introduce the drawings needed to be used in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present disclosure, and for those skilled in the art, other drawings can be obtained based on these drawings without creative labor.

[0019] Figure 1 is a flowchart of the method for constructing a drug screening model in the embodiment of the present application;

[0020] Figure 2 is a flowchart of the method for constructing a drug screening model in the embodiment of the present application;

[0021] Figure 3 is a flowchart of the method for constructing a drug screening model in the embodiment of the present application;

[0022] Figure 4 is a flowchart of the method for constructing a drug screening model in the embodiment of the present application;

[0023] Figure 5 is a structural block diagram of the device for constructing a drug screening model in the embodiment of the present application;

[0024] Figure 6 is an internal structure diagram of the computer device in the embodiment of the present application. DETAILED DESCRIPTION

[0025] The embodiments of the present disclosure will be described in detail below with reference to the drawings.

[0026] The embodiments of the present disclosure will be described in detail below with reference to the drawings.

[0027] It should be noted that various aspects of embodiments within the scope of the appended claims are described below. It will be apparent that the aspects described herein can be embodied in a wide variety of forms, and any particular structure and / or function described herein is merely illustrative. Based on this disclosure, those skilled in the art will understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects set forth herein can be used to implement the device and / or practice the method. Additionally, this device and / or method can be implemented using other structures and / or functionalities besides one or more of the aspects set forth herein.

[0028] It should also be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this disclosure. The drawings only show the components related to this disclosure and are not drawn according to the number, shape and size of the components in actual implementation. In actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0029] Furthermore, specific details are provided in the following description to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that the described aspects can be practiced without these specific details.

[0030] like Figure 1 As shown, this disclosure provides a method for constructing a drug screening model, which can be applied to a terminal or a server. The terminal can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable smart devices. The server can be a standalone server or a server cluster composed of multiple servers. The method includes the following steps:

[0031] Step 101: Obtain the drug training set, which includes the chemical formulas of drug proteins, the chemical formulas of small molecules, and classification labels.

[0032] The server can obtain a drug training set, which includes the chemical formulas of drug proteins, small molecules, and classification tags. The drug proteins in the training set can be validated drugs; for example, they can be protein portions with protein-ligand co-crystal structures compiled from a PDB (protein data bank) database. Small molecules are small molecule compounds with enzyme activity records from proteins within the PDB database. Classification tags can be text or letters describing the drug's properties, such as content related to efficacy.

[0033] Step 102, based on the chemical formula, draw the initial graph network of the drug protein and small molecules, the atoms in the initial graph network are nodes, and the chemical bonds are edges connecting the nodes.

[0034] The server draws the initial graph network based on the chemical formula. The atoms in the initial graph network are nodes, and the chemical bonds are edges connecting the nodes. When there is a chemical bond between two atoms, whether it is a single bond, a double bond, or a triple bond, it is drawn as an edge. The initial graph network can be a function set organized in a graph structure in a topological space for relationship reasoning. The graph structure can be composed of two sets: a node set (Node) and an edge set (Edge), where the edge set describes how the nodes are connected to each other.

[0035] In one embodiment, based on the chemical formula, the initial graph network of the drug protein and small molecules is drawn, including: based on traditional molecular dynamics, the drug protein and small molecules are molecularly docked to obtain a target molecule; and based on the chemical formula of the target molecule, the initial graph network of the drug protein and small molecules is drawn. As shown in Figure 2 The server can perform molecular docking of the drug protein and small molecules based on traditional molecular dynamics, and draw the initial graph network after docking to obtain the initial graph network.

[0036] In one embodiment, based on the chemical formula, the initial graph network of the drug protein and small molecules is drawn, including: based on the chemical formula, the initial graph network of the drug protein and small molecules is drawn respectively, at this time the initial graph network includes the initial graph network of the drug protein and the initial graph network of the small molecules. As shown in Figure 3 The server can also draw the initial graph network of the drug protein and small molecules based on the chemical formula respectively, at this time the initial graph network includes the initial graph network of the drug protein and the initial graph network of the small molecules.

[0037] Step 103, using a random initialization vector to identify the weight vector of each node in the initial graph network.

[0038] The server uses a random initialization vector to identify the weight vector of each node in the initial graph network. The server can randomly generate a weight value for any edge, and then assign a corresponding weight vector to each node based on the generated weight value. In one embodiment, using a random initialization vector to identify the weight vector of each node in the initial graph network includes: using a uniform distribution or a normal distribution to select a value in the interval (0, 1) to generate an initialization vector, which is assigned to the node as its weight vector. For example, the server can assign a weight value to each edge in a uniform distribution, and then assign a corresponding weight vector to each node based on the generated weight value. Uniform distribution means that each value in the vector is equally distributed in the interval (0, 1). Normal distribution means that each value in the vector is normally distributed in the interval (0, 1).

[0039] Step 104, reconstructing each node of the initial graph network according to the connection relationship of the initial graph network to obtain a reconstructed graph network, and repeating the reconstruction step to obtain at least two layers of reconstructed graph networks.

[0040] The server reconstructs each node of the initial graph network according to the connection relationship of the initial graph network to obtain a reconstructed graph network, and repeats the reconstruction step to obtain at least two layers of reconstructed graph networks. Preferably, the number of layers of the reconstructed graph network is 3-6 layers. The server repeats the reconstruction step based on the newly constructed reconstructed graph network each time, so that the reconstructed graph network each time is different, and the information contained in each node is also gradually enriched.

[0041] Step 105, deep learning of the initial graph network and the at least two layers of reconstructed graph networks according to the classification label to construct a drug screening model.

[0042] The server deep learns the initial graph network and the at least two layers of reconstructed graph networks according to the classification label to construct a drug screening model. The deep learning network can be a BP neural network (Back Propagation), a convolutional network, a neural network, etc. The server determines the network features in the initial graph network and the at least two layers of reconstructed graph networks through learning, and trains the network features corresponding to the classification label to construct a drug screening model.

[0043] The above method can effectively reduce the design of artificial features, expand the coverage of features, improve efficiency and accuracy, and greatly save the calculation consumption of molecular dynamics, quantum mechanics, quantum chemistry, etc. by using big data and deep learning methods, and greatly improve the calculation speed. The present application can consider the high flexibility of proteins, and the simulation method based on the elastic graph neural network can better simulate the flexible characteristics of proteins, and more accurately predict the properties of compounds in different scales of biological and chemical experiments.

[0044] As shown in Figure 4 In one embodiment, reconstructing each node of the initial graph network according to the connection relationship of the initial graph network to obtain a reconstructed graph network includes the following steps:

[0045] Step 401, obtaining the weight vector of each node and the associated weight vector of the associated node connected to the node according to the connection relationship of the initial graph network.

[0046] The server obtains the weight vector of each node and the associated weight vector of the associated node connected to the node according to the connection relationship of the initial graph network.

[0047] At step 402, the weight vector and the associated weight vector are integrated to obtain a reconstructed weight vector of the node.

[0048] The server integrates the weight vector and the associated weight vector to obtain a reconstructed weight vector of the node. In an embodiment, the integration formula is where N(v i ) is a set of nodes connected to all nodes i, v i is the i-th node, i and j are node numbers, is a vector of the i-th and j-th nodes in the k-th layer network, is a vector of the edge, and f k is the k-th layer network.

[0049] At step 403, a reconstructed graph network is generated according to the reconstructed weight vector.

[0050] The server generates a reconstructed graph network according to the reconstructed weight vector.

[0051] In one embodiment, the initial graph network and the at least two layers of reconstructed graph networks are subjected to deep learning according to the classification label to construct a drug screening model, including: respectively extracting features of the initial graph network and the at least two layers of reconstructed graph networks to obtain hierarchical network features; and learning the hierarchical network features by using a deep learning network according to the classification label to construct a drug screening model.

[0052] The server respectively extracts features of the initial graph network and the at least two layers of reconstructed graph networks to obtain hierarchical network features. In an embodiment, the server can respectively extract picture features of the initial graph network and the reconstructed graph network to analyze picture details in the graph network. In an embodiment, the server can respectively extract vector features of the initial graph network and the reconstructed graph network to further analyze the graph network. Specifically, the server can extract the vector features according to the formula where R is a readout function (feature reading function). The server can select the vector features in two ways: summation and averaging.

[0053] In one embodiment, the initial graph network and the at least two layers of reconstructed graph networks are respectively subjected to feature extraction to obtain hierarchical network features, including: respectively extracting features of the initial graph network and the at least two layers of reconstructed graph networks by using a function to obtain function vector features; and performing a normalization operation on the function vector features in a probability space to obtain the hierarchical network features.

[0054] The server respectively extracts features of the initial graph network and the at least two layers of reconstructed graph networks by using a function to obtain function vector features. The server extracts the function vector features according to the formula The function vector feature is obtained. Wherein R is a readout function (feature reading function). The server selects the vector feature by averaging. The server normalizes the function vector feature in the probability space to obtain the hierarchical network feature. The server can use the softmax function as a direct function of the classifier and calculate the normalization operation in the probability space. The specific mathematical formula of the softmax function is Its function is to normalize the calculated result from the entire real number domain to the (0, 1) interval, thereby representing the probability between nodes.

[0055] The above method only retains the most core compound information (such as: atomic number, single bond, double bond, etc. Node or edge information) as the connection atom information, so as to more efficiently complete the model training and extract high-quality model features.

[0056] The application also provides a drug screening method, comprising: based on the chemical formula of the target protein and the small molecule, drawing a target initial graph network, the atoms in the target initial graph network are nodes, and the chemical bonds are edges connecting the nodes; inputting the target initial graph network into a drug screening model, and outputting an analysis result of the target protein and the small molecule, wherein the drug screening model is obtained by training the above method.

[0057] In one embodiment, as shown in Figure 5 A drug screening model construction device is provided, the device comprising a training set acquisition module 501, a graph network drawing module 502, a vector identification module 503, a reconstruction module 504, and a model training module 505.

[0058] The training set acquisition module 501 is used to acquire a drug training set, and the drug training set contains the chemical formula of the drug protein, the chemical formula of the small molecule, and the classification label.

[0059] The graph network drawing module 502 is used to draw an initial graph network of the drug protein and the small molecule based on the chemical formula, and the atoms in the initial graph network are nodes, and the chemical bonds are edges connecting the nodes.

[0060] The vector identification module 503 is used to identify the weight vector of each node in the initial graph network by using a random initialization vector.

[0061] The reconstruction module 504 is used to reconstruct each node of the initial graph network according to the connection relationship of the initial graph network to obtain a reconstructed graph network, and the reconstruction step is repeated to obtain at least two layers of reconstructed graph networks.

[0062] The model training module 505 is used to perform deep learning on the initial graph network and the at least two layers of reconstructed graph networks according to the classification label to construct a drug screening model.

[0063] In one of the embodiments, the graph network drawing module comprises:

[0064] The molecule docking unit is configured to dock the drug protein and the small molecule based on traditional molecular dynamics to obtain the target molecule.

[0065] The graph network drawing unit is configured to draw the initial graph network of the drug protein and the small molecule according to the chemical formula of the target molecule.

[0066] In one of the embodiments, the graph network drawing module comprises:

[0067] The graph network drawing unit is configured to draw the initial graph network of the drug protein and the small molecule based on the chemical formula, and the initial graph network comprises the initial graph network of the drug protein and the initial graph network of the small molecule.

[0068] In one of the embodiments, the vector identification module comprises:

[0069] The vector identification unit is configured to generate an initialization vector by selecting a value in the interval (0, 1) in a uniform distribution or a normal distribution, and assign the initialization vector to a node as a weight vector of the node.

[0070] In one of the embodiments, the reconstruction module 504 comprises:

[0071] The weight vector acquisition unit is configured to acquire the weight vector of each node and the associated weight vector of the associated node connected to the node according to the connection relationship of the initial graph network.

[0072] The integration unit is configured to integrate the weight vector and the associated weight vector to obtain a reconstructed weight vector of the node.

[0073] The reconstruction unit is configured to generate a reconstructed graph network according to the reconstructed weight vector.

[0074] In one of the embodiments, the model training module comprises:

[0075] The feature extraction unit is configured to perform feature extraction on the initial graph network and at least two layers of reconstructed graph networks respectively to obtain hierarchical network features.

[0076] The model construction unit is configured to learn the hierarchical network features by using a deep learning network according to the classification label to construct a drug screening model.

[0077] In one of the embodiments, the model training module comprises:

[0078] The vector feature extraction unit is configured to perform feature extraction on the initial graph network and at least two layers of reconstructed graph networks by using a function respectively to obtain function vector features.

[0079] A normalization unit is configured to normalize the function vector feature in a probability space to obtain the hierarchical network feature.

[0080] In one embodiment, a drug screening device is provided, which includes a graph network drawing module and a drug analysis module.

[0081] The graph network drawing module is configured to draw a target initial graph network based on the chemical formula of the target protein and the small molecule, wherein atoms are nodes and chemical bonds are edges connecting the nodes in the target initial graph network.

[0082] The drug analysis module is configured to input the target initial graph network into a drug screening model and output an analysis result of the target drug, wherein the drug screening model is trained by the method.

[0083] In one embodiment, a computer device is provided, which can be a server, and an internal structure diagram thereof can be as shown in FIG. 1. Figure 4 The computer device includes a processor, a memory, a network interface and a database connected through a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The database of the computer device is configured to store data such as a drug training set or a drug screening model. The network interface of the computer device is configured to communicate with an external terminal through a network connection. The computer program is executed by the processor to implement a drug screening model construction method or a drug screening method.

[0084] In one embodiment, a computer device is provided, which includes a memory and a processor. The memory stores a computer program, and the processor implements the following steps when executing the computer program: obtaining a drug training set, the drug training set including chemical formulas of drug proteins, chemical formulas of small molecules and classification labels; drawing an initial graph network based on the drug chemical formulas, wherein atoms are nodes and chemical bonds are edges connecting the nodes in the initial graph network; identifying weight vectors of each node in the initial graph network using a random initialization vector; reconstructing each node of the initial graph network according to the connection relationship of the initial graph network to obtain a reconstructed graph network, and repeating the reconstruction step to obtain at least two layers of reconstructed graph networks; and performing deep learning on the initial graph network and the at least two layers of reconstructed graph networks according to the classification labels to construct a drug screening model.

[0085] In one embodiment, the processor implements the initial graph network of the drug protein and the small molecule based on the chemical formula when executing the computer program, and the implementation includes: performing molecular docking on the drug protein and the small molecule based on traditional molecular dynamics to obtain a target molecule; and drawing the initial graph network of the drug protein and the small molecule according to the chemical formula of the target molecule.

[0086] In one embodiment, the processor implements the initial graph network of the drug protein and the small molecule based on the chemical formula when executing the computer program, and the implementation includes: drawing the initial graph network of the drug protein and the small molecule based on the chemical formula, respectively, wherein the initial graph network includes an initial graph network of the drug protein and an initial graph network of the small molecule.

[0087] In one embodiment, the processor implements the initial graph network of the drug protein and the small molecule based on the chemical formula when executing the computer program, and the implementation includes: drawing the initial graph network of the drug protein and the small molecule based on the chemical formula, respectively, wherein the initial graph network includes an initial graph network of the drug protein and an initial graph network of the small molecule.

[0088] In one embodiment, the processor implements the initial graph network of the drug protein and the small molecule based on the chemical formula when executing the computer program, and the implementation includes: drawing the initial graph network of the drug protein and the small molecule based on the chemical formula, respectively, wherein the initial graph network includes an initial graph network of the drug protein and an initial graph network of the small molecule.

[0089] In one embodiment, the processor implements the initial graph network of the drug protein and the small molecule based on the chemical formula when executing the computer program, and the implementation includes: drawing the initial graph network of the drug protein and the small molecule based on the chemical formula, respectively, wherein the initial graph network includes an initial graph network of the drug protein and an initial graph network of the small molecule.

[0090] In one embodiment, the processor implements the initial graph network of the drug protein and the small molecule based on the chemical formula when executing the computer program, and the implementation includes: drawing the initial graph network of the drug protein and the small molecule based on the chemical formula, respectively, wherein the initial graph network includes an initial graph network of the drug protein and an initial graph network of the small molecule.

[0091] In one embodiment, a computer device is provided, comprising a memory and a processor, the memory storing a computer program, the processor implementing the following steps when executing the computer program: based on the chemical formula of the target protein and the small molecule, drawing a target initial graph network, in which atoms are nodes and chemical bonds are edges connecting the nodes; inputting the target initial graph network into a drug screening model, and outputting an analysis result of the target protein and the small molecule, wherein the drug screening model is trained by the method described above.

[0092] In one embodiment, a computer readable storage medium is provided, which stores a computer program, the computer program being executed by a processor to implement the following steps: obtaining a drug training set, the drug training set comprising the chemical formula of a drug protein, the chemical formula of a small molecule, and a classification label; based on the chemical formula of the drug, drawing an initial graph network, in which atoms are nodes and chemical bonds are edges connecting the nodes; using a random initialization vector to identify the weight vector of each node in the initial graph network; reconstructing each node of the initial graph network according to the connection relationship of the initial graph network to obtain a reconstructed graph network, and repeating the reconstruction step to obtain at least two layers of reconstructed graph networks; performing deep learning on the initial graph network and the at least two layers of reconstructed graph networks according to the classification label to construct a drug screening model.

[0093] In one embodiment, the computer program executed by the processor to implement the drawing of the initial graph network of the drug protein and the small molecule based on the chemical formula comprises: performing molecular docking of the drug protein and the small molecule based on traditional molecular dynamics to obtain a target molecule; and drawing the initial graph network of the drug protein and the small molecule according to the chemical formula of the target molecule.

[0094] In one embodiment, the computer program executed by the processor to implement the drawing of the initial graph network of the drug protein and the small molecule based on the chemical formula comprises: based on the chemical formula, drawing the initial graph network of the drug protein and the small molecule respectively, wherein the initial graph network at this time comprises the initial graph network of the drug protein and the initial graph network of the small molecule.

[0095] In one embodiment, the computer program executed by the processor to implement the use of a random initialization vector to identify the weight vector of each node in the initial graph network comprises: using a uniform distribution or a normal distribution to select a value in the interval (0, 1) to generate an initialization vector, which is assigned to the node as its weight vector.

[0096] In one embodiment, the computer program is executed by the processor to reconstruct nodes of the initial graph network according to the connection relationship of the initial graph network to obtain a reconstructed graph network, including: obtaining a weight vector of each node and an associated weight vector of an associated node connected to the node according to the connection relationship of the initial graph network; integrating the weight vector and the associated weight vector to obtain a reconstructed weight vector of the node; and generating the reconstructed graph network according to the reconstructed weight vector.

[0097] In one embodiment, the computer program is executed by the processor to perform deep learning on the initial graph network and the at least two layers of reconstructed graph networks according to the classification label to construct a drug screening model, including: performing feature extraction on the initial graph network and the at least two layers of reconstructed graph networks respectively to obtain hierarchical network features; and learning the hierarchical network features by using a deep learning network according to the classification label to construct the drug screening model.

[0098] In one embodiment, the computer program is executed by the processor to perform feature extraction on the initial graph network and the at least two layers of reconstructed graph networks respectively to obtain hierarchical network features, including: performing feature extraction on the initial graph network and the at least two layers of reconstructed graph networks by using a function respectively to obtain function vector features; and performing normalization operation on the function vector features in a probability space to obtain the hierarchical network features.

[0099] In one embodiment, a computer readable storage medium is provided, and the computer readable storage medium stores a computer program. The computer program is executed by a processor to perform the following steps: based on a chemical formula of a target protein and a small molecule, a target initial graph network is drawn, atoms in the target initial graph network are nodes, and chemical bonds are edges connecting the nodes; the target initial graph network is input into a drug screening model, and an analysis result of the target protein and the small molecule is output, wherein the drug screening model is obtained by training the method.

[0100] The above merely describes specific embodiments of the present disclosure, but the protection scope of the present disclosure is not limited thereto. Any changes or replacements within the technical scope disclosed by the present disclosure can be easily conceived by those skilled in the art, and should be covered by the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure should be subject to the protection scope of the claims.

Claims

1. A method for constructing a drug screening model for protein crystal structure screening, characterized in that, include: Obtain a drug training set, which includes the chemical formulas of drug proteins, the chemical formulas of small molecules, and classification labels; Based on chemical formulas, an initial graph network of drug proteins and small molecules is drawn, in which atoms are nodes and chemical bonds are edges connecting the nodes; The initial network of maps for drug proteins and small molecules, based on chemical formulas, includes: The drug protein and the small molecule are molecularly docked based on traditional molecular dynamics to obtain the target molecule; an initial graph network of the drug protein and the small molecule is drawn based on the chemical formula of the target molecule. Alternatively, based on the chemical formula, initial graph networks for drug proteins and small molecules can be drawn separately. In this case, the initial graph network includes the initial graph network for drug proteins and the initial graph network for small molecules. The weight vectors of each node in the initial graph network are identified by random initialization vectors; Reconstructing each node of the initial graph network according to the connection relationships of the initial graph network to obtain a reconstructed graph network, repeating the reconstruction steps to obtain at least two layers of reconstructed graph networks; the step of reconstructing each node of the initial graph network according to the connection relationships of the initial graph network to obtain a reconstructed graph network includes: Based on the connection relationships of the initial graph network, obtain the weight vector of each node and the associated weight vector of the associated nodes connected to that node; By integrating the weight vector and the associated weight vector, the reconstructed weight vector of the node is obtained; Generate a reconstructed graph network based on the reconstructed weight vectors; A drug screening model is constructed by performing deep learning on the initial graph network and the at least two layers of the reconstructed graph network based on the classification labels.

2. The method according to claim 1, characterized in that, The step of using random initialization vectors to identify the weight vectors of each node in the initial graph network includes: An initialization vector is generated by selecting values ​​in the (0,1) interval using a uniform or normal distribution, and then assigned to the node as its weight vector.

3. The method according to claim 1, characterized in that, The step of performing deep learning on the initial graph network and the at least two layers of the reconstructed graph network based on the classification labels to construct a drug screening model includes: Feature extraction is performed on the initial graph network and the at least two layers of the reconstructed graph network to obtain hierarchical network features; Based on the classification labels, a deep learning network is used to learn the hierarchical network features to construct a drug screening model.

4. The method according to claim 3, characterized in that, The step of extracting features from the initial graph network and the at least two layers of the reconstructed graph network to obtain hierarchical network features includes: The initial graph network and the at least two layers of the reconstructed graph network are subjected to feature extraction using functions to obtain function vector features. The function vector features are normalized in the probability space to obtain the hierarchical network features.

5. A drug screening method, characterized in that, include: Based on the chemical formulas of the target protein and small molecules, an initial target graph network is drawn, in which atoms are nodes and chemical bonds are edges connecting the nodes; The target initial graph network is input into the drug screening model, and the analysis results of the target protein and small molecule are output, wherein the drug screening model is trained by the method of any one of claims 1 to 4.

6. A drug screening model construction device, characterized in that, The device includes: The training set acquisition module is used to acquire a drug training set, which includes the chemical formulas of drug proteins, the chemical formulas of small molecules, and classification labels. The graph network drawing module is used to draw an initial graph network of drug proteins and small molecules based on chemical formulas. In the initial graph network, atoms are nodes and chemical bonds are edges connecting the nodes. A vector identification module is used to identify the weight vector of each node in the initial graph network using a randomly initialized vector; The reconstruction module is used to reconstruct each node of the initial graph network according to the connection relationship of the initial graph network to obtain a reconstructed graph network. The reconstruction steps are repeated to obtain at least two layers of reconstructed graph networks. The model training module is used to perform deep learning on the initial graph network and the at least two layers of the reconstructed graph network based on the classification labels to construct a drug screening model.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

Images