Method, device, equipment and storage medium for predicting drug-target interaction

Abstract

The present invention relates to the field of neural network technology of artificial intelligence technology. The present invention provides a method, device, equipment and storage medium for predicting drug-target interactions, wherein the method comprises: calling a pre-constructed graph neural network to extract drug features in a molecular graph of a drug, processing the molecular graph according to a restarted random walk algorithm, predicting the similarity of drug features between two adjacent nodes in the molecular graph, obtaining global structural information, inputting the global structural information into a preset deep neural network, obtaining low-dimensional feature information of the drug, acquiring the protein sequence of the target, calling a long short-term memory network to process the protein sequence, obtaining protein features, inputting the low-dimensional feature information and protein features into a preset fully connected layer to predict the results of the interaction between the drug and the target, and extracting the information contained in the drug and protein sequences in a targeted manner, thereby improving the prediction efficiency of the interaction between the drug and the target.

Description

Technical Field

[0001] The present invention relates to the field of neural network technology of artificial intelligence technology. Specifically, the present invention relates to a method, device, equipment and storage medium for predicting the interaction between drugs and targets. Background Art

[0002] Drug development is a systematic process, facing challenges such as high costs, long cycles, and low success rates. The first step in determining whether a molecule can function as a drug is to analyze its ability to interact with a specific target to determine which drug functions need to be enhanced or inhibited. Therefore, predicting drug-target interactions is a key issue in drug discovery and repositioning research. It helps analyze the effectiveness of drug molecules and is a key step in drug development and repurposing. It also helps understand drug mechanisms of action, disease pathology, and minimize adverse side effects caused by drug molecules, thereby assisting in the development of therapeutic drugs and having important implications for disease treatment.

[0003] However, due to limitations in precision and cost, traditional biological experimental methods often find it difficult to obtain large-scale drug-target interaction pairs, resulting in a very limited number of drug-target interaction pairs that have been experimentally identified and verified. Therefore, traditional large-scale verification through biological experimental methods is inefficient and costly. Summary of the Invention

[0004] The main purpose of the present invention is to provide a method, device, equipment and storage medium for predicting drug-target interactions to improve the prediction efficiency of drug-target interactions and reduce costs.

[0005] In order to achieve the above-mentioned object of the invention, the present invention provides a method for predicting the interaction between a drug and a target, which comprises:

[0006] Obtaining a molecular map of a drug;

[0007] Calling a pre-built graph neural network to extract drug features in the molecular graph;

[0008] Processing the molecular graph according to a restarted random walk algorithm to predict the similarity of drug features between two adjacent nodes in the molecular graph to obtain global structural information; wherein the global structural information is used to describe the similarity of drug features between two adjacent nodes in the molecular graph;

[0009] Inputting the global structural information into a preset deep neural network to obtain low-dimensional feature information of the drug;

[0010] Obtaining a target protein sequence, and processing the protein sequence using a long short-term memory network to obtain protein features;

[0011] The low-dimensional feature information and protein features are input into a preset fully connected layer to predict the interaction result between the drug and the target.

[0012] Preferably, inputting the low-dimensional feature information and protein features into a preset fully connected layer to predict the interaction result between the drug and the target includes:

[0013] Inputting the low-dimensional feature information and protein features into a preset fully connected layer to obtain a Boolean value output by the fully connected layer;

[0014] The interaction result between the drug and the target is determined according to the Boolean value.

[0015] Preferably, the obtaining of the molecular graph of the drug comprises:

[0016] Obtaining the SMILES expression of the drug from the database;

[0017] The SMILES expression of the drug is converted into a molecular graph using the RDKit tool, with each atom of the drug as a node and the element bonds between atoms as edges.

[0018] Preferably, the obtaining of the molecular graph of the drug comprises:

[0019] Obtaining the SMILES expression of the drug from the database;

[0020] The SMILES expression is input into a pre-trained molecular graph model to obtain the molecular graph of the drug.

[0021] Furthermore, before inputting the SMILES expression into a pre-trained molecular graph model to obtain the molecular graph of the drug, the method further includes:

[0022] Obtaining a training set; wherein the training set includes standard SMILES expressions of multiple drugs and corresponding standard molecular graphs;

[0023] Training the neural network model according to the training set;

[0024] Obtaining the training results of the neural network model after training, and determining whether the training results meet the requirements;

[0025] If so, the trained neural network model is used as a molecular graph model.

[0026] Preferably, determining whether the training result meets the requirements includes:

[0027] Calculating the loss value of the trained neural network model based on the training results and a preset loss function;

[0028] Determining whether the loss value is lower than a preset loss value;

[0029] If so, determining that the training result meets the requirements;

[0030] If not, it is determined that the training result does not meet the requirements.

[0031] Preferably, the predicting the similarity of drug features between two adjacent nodes in the molecular graph includes:

[0032] Converting the drug features between two adjacent nodes in the molecular graph into vectors to obtain a first vector and a second vector corresponding to the two adjacent nodes;

[0033] Calculating the cosine of the angle between the first vector and the second vector;

[0034] The similarity of drug features between two adjacent nodes in the molecular graph is determined according to the cosine value.

[0035] The present invention also provides a device for predicting drug-target interaction, comprising:

[0036] An acquisition module, used to obtain the molecular graph of the drug;

[0037] A first calling module is used to call a pre-built graph neural network to extract drug features in the molecular graph;

[0038] a prediction module, configured to process the molecular graph according to a restarted random walk algorithm, predict the similarity of drug features between two adjacent nodes in the molecular graph, and obtain global structural information; wherein the global structural information is used to describe the similarity of drug features between two adjacent nodes in the molecular graph;

[0039] A first input module is used to input the global structure information into a preset deep neural network to obtain low-dimensional feature information of the drug;

[0040] The second calling module is used to obtain the target protein sequence, call the long short-term memory network to process the protein sequence, and obtain protein features;

[0041] The second input module is used to input the low-dimensional feature information and protein features into a preset fully connected layer to predict the interaction result between the drug and the target.

[0042] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0043] The present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above methods are implemented.

[0044] The present invention provides a method, device, equipment and storage medium for predicting drug-target interactions. The method obtains a molecular graph of the drug, calls a pre-built graph neural network to extract drug features in the molecular graph, processes the molecular graph according to a restarted random walk algorithm, predicts the similarity of drug features between two adjacent nodes in the molecular graph, obtains global structural information, inputs the global structural information into a preset deep neural network, obtains low-dimensional feature information of the drug, obtains the protein sequence of the target, calls a long short-term memory network to process the protein sequence, obtains protein features, inputs the low-dimensional feature information and protein features into a preset fully connected layer to predict the results of the drug-target interaction, combines the characteristics of the graph neural network and the long short-term memory network to deal with the problem of drug-target interaction, and specifically extracts the information contained in the drug and protein sequences, thereby improving the prediction efficiency of drug-target interaction and reducing costs. In addition, by using the restarted random walk algorithm to capture the various relationships between different nodes in the molecular graph, global structural information of similarity is obtained, which helps to improve prediction accuracy and prediction performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 Schematic diagram of a process for predicting drug-target interactions according to an embodiment of the present invention;

[0046] Figure 2 Schematic diagram of a process for predicting drug-target interactions according to another embodiment of the present invention;

[0047] Figure 3 Schematic diagram of a process for predicting drug-target interactions according to another embodiment of the present invention;

[0048] Figure 4 Schematic diagram of a process for predicting drug-target interaction according to another embodiment of the present invention;

[0049] Figure 5 Schematic diagram of a process for predicting drug-target interactions according to another embodiment of the present invention;

[0050] Figure 6 Schematic diagram of a process for predicting drug-target interaction according to another embodiment of the present invention;

[0051] Figure 7 Schematic diagram of a process for predicting drug-target interactions according to another embodiment of the present invention;

[0052] Figure 8 This is a schematic block diagram of the structure of a device for predicting drug-target interaction according to one embodiment of the present invention;

[0053] Figure 9 The figure is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0054] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0056] This invention proposes a method for predicting drug-target interactions. Embodiments of the invention can acquire and process relevant data based on artificial intelligence technology. 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, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results.

[0057] Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interaction systems, and mechatronics. AI software technologies primarily encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0058] The present invention proposes a method for predicting drug-target interactions, which uses a server as the execution entity. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0059] This drug-target interaction prediction method is used to solve the technical problem of low efficiency and high cost in obtaining large-scale drug-target interaction pairs through traditional biological experimental methods. Figure 1 In one embodiment, the method for predicting the interaction between a drug and a target comprises the following steps S11-S16:

[0060] S11, obtaining a molecular map of the drug;

[0061] S12. Calling a pre-built graph neural network to extract drug features in the molecular graph;

[0062] S13. Processing the molecular graph according to a restarted random walk algorithm to predict the similarity of drug features between two adjacent nodes in the molecular graph to obtain global structural information; wherein the global structural information is used to describe the similarity of drug features between two adjacent nodes in the molecular graph;

[0063] S14, inputting the global structure information into a preset deep neural network to obtain low-dimensional feature information of the drug;

[0064] S15. Obtain a target protein sequence, and use a long short-term memory network to process the protein sequence to obtain protein features;

[0065] S16. Inputting the low-dimensional feature information and protein features into a preset fully connected layer to predict the interaction result between the drug and the target.

[0066] As described in the above step S11, the drug is a drug whose interaction with the target needs to be predicted. The molecular graph of the drug can be converted through the RDKit toolkit. The molecular graph of the drug uses the atoms that constitute the drug as nodes and the connections between atoms as edges; the molecular graph can be constructed and stored in the form of a two-dimensional adjacency matrix, the rows and columns of the matrix represent atoms, and the values ​​of the matrix represent whether there is a connection between two atoms. For example, the feature of each node is a 78-bit vector encoding, of which 11 bits represent the type of atom, 11 bits represent the degree of the atomic node, 11 bits represent the total number of connections between the atom and hydrogen atoms in the compound, 11 bits represent the number of hidden hydrogen atoms connected to the atom, and the remaining 1 bit indicates whether the atom is on the aromatic ring.

[0067] As described in step S12 above, a pre-built graph neural network can be called to extract drug features in the molecular graph, where the drug features are the molecular features of the nodes in the molecular graph. Among them, the graph neural network (GNN) is a type of model for deep learning on graph structure data. The input of a graph neural network is usually a graph structure, and its final output generally depends on the specific task. Taking graph property prediction as an example, the graph neural network trains the implicit vector representation of each node in the graph based on the graph structure and input node attributes. Its goal is to make the vector representation contain sufficiently powerful expression information so that it can help each node extract information. Finally, through methods such as average pooling, the information vector representation of the entire graph can be obtained.

[0068] As described in steps S13-S14 above, after the drug is converted into a molecular graph, the molecular graph is processed using the restarted random walk algorithm. The restarted random walk algorithm was originally an algorithm for image segmentation, which iteratively explores the overall structure of the molecular graph to estimate the similarity or affinity score between two nodes. Starting from a node, each step faces two choices: move to a randomly selected adjacent node or jump back to the starting node. The algorithm only contains a fixed parameter r, called the "restart probability" (1-r represents the probability of moving to an adjacent node). After the iteration reaches stability, the stable probability vector contains the similarity of all nodes in the molecular graph with the starting node. The probability distribution obtained after stabilization can be regarded as a distribution affected by the starting node.

[0069] This embodiment can capture the multifaceted relationships between two nodes in the molecular graph by restarting the random walk algorithm, and then obtain global structural information describing the similarity of drug characteristics between two adjacent nodes in the entire molecular graph, which is then fed into a preset deep neural network to extract low-dimensional feature information of the drug.

[0070] As described in step S15 above, for the target protein sequence, the present invention uses a long short-term memory network LSTM with a self-attention mechanism for processing. The linear self-attention mechanism method is used to reduce the running speed and memory consumption, and at the same time construct an unbiased estimator for the attention matrix to improve the effect of protein sequence processing.

[0071] Among them, the long short-term memory network contains two outputs: the output of all time steps O = [O_1, O_2,…, O_D] and the hidden state H_D of the last time step D.

[0072] Since O = [O_1, O_2, ..., O_D] represents the characteristics of a word / phrase, and H_D represents the characteristics of a text, in order to identify the importance of a word to a text, it is necessary to establish a self-attention relationship between H_D and O, that is, to establish the weight of each time step output O_t for H_D. Since the long short-term memory network itself takes position information into account, there is no need to set up an additional position encoding. The implementation method of the self-attention mechanism in the long short-term memory network is additive attention: the hidden state of the last time step and the output of each time step are concatenated as the query, the output of each time step is linearly transformed as the value, the linear transformation matrix is ​​used as the key, and the query and value are multiplied to form the result matrix z. The present invention uses a linear attention mechanism to decompose the attention matrix into the product of a random nonlinear function of the original query and the key, namely the so-called random feature. In this way, the similarity can be encoded more efficiently, and a scalable low-variance, unbiased estimation of the attention mechanism can be provided. While maintaining linear space and time complexity, the accuracy is also guaranteed.

[0073] As described in step S16 above, after extracting low-dimensional feature information from the drug and protein features from the protein, these features are connected and input into a preset fully connected layer, and then the interaction results between the drug and the target are predicted based on the drug-target interaction pairs with known interactions.

[0074] The present invention provides a method for predicting drug-target interactions, which obtains a molecular graph of the drug, calls a pre-built graph neural network to extract drug features in the molecular graph, processes the molecular graph according to a restarted random walk algorithm, predicts the similarity of drug features between two adjacent nodes in the molecular graph, obtains global structural information, inputs the global structural information into a preset deep neural network, obtains low-dimensional feature information of the drug, obtains the protein sequence of the target, calls a long short-term memory network to process the protein sequence, obtains protein features, inputs the low-dimensional feature information and protein features into a preset fully connected layer to predict the results of the drug-target interaction, combines the characteristics of the graph neural network and the long short-term memory network to deal with the problem of drug-target interaction, and specifically extracts the information contained in the drug and protein sequences, thereby improving the prediction efficiency of drug-target interactions and reducing costs; in addition, by using the restarted random walk algorithm to capture the various relationships between different nodes in the molecular graph, global structural information of similarity is obtained, which helps to improve prediction accuracy and prediction performance.

[0075] In one embodiment, reference Figure 2 As shown, the low-dimensional feature information and protein features are input into a preset fully connected layer to predict the interaction result between the drug and the target, which may specifically include the following steps S21-S22:

[0076] S21, inputting the low-dimensional feature information and protein features into a preset fully connected layer to obtain a Boolean value output by the fully connected layer;

[0077] S22. Determine the interaction result between the drug and the target according to the Boolean value.

[0078] In this embodiment, the low-dimensional feature information and protein features of the drug are input into a preset fully connected layer, and the feature vector of the drug is output through the branch-pretrained molecular graph model in the fully connected layer. The protein feature outputs the feature vector corresponding to the protein feature through the branch-pretrained protein sequence language model in the fully connected layer. The neural network in the fully connected layer further predicts whether the drug can effectively act on the target based on the feature vector of the drug and the feature vector corresponding to the protein feature, and obtains the Boolean value output by the fully connected layer. For example, if the drug can effectively act on the target, the fully connected layer outputs True, otherwise it outputs False. Therefore, only the molecular graph and protein sequence of the drug are needed to directly output the interaction results between the drug and the target. This end-to-end learning does not require sequence alignment, is more convenient to run, has good generalization ability, etc., and therefore has great potential in the task of predicting the interaction between drugs and targets.

[0079] In one embodiment, reference Figure 3 As shown, the method of obtaining the molecular graph of the drug may specifically include the following steps S31-S32:

[0080] S31, obtaining a SMILES expression of the drug from a database;

[0081] S32. Using the RDKit tool, convert the SMILES expression of the drug into a molecular graph, with each atom of the drug as a node and the element bonds between atoms as edges.

[0082] The molecular formula of a drug can be pre-stored in a database and can be a string representation of the drug's molecular structure, for example, a SMILES expression. SMILES (Simplified Molecular Input Line Entry Specification) is a specification that explicitly describes a molecular structure using ASCII strings. The molecular formula of a drug is unique and varies from drug to drug. Accordingly, if the molecular formula is a SMILES expression, the SMILES expression of the drug is also unique and varies from drug to drug.

[0083] For drug molecules, the present invention uses the RDKit tool to model the SMILES expression of the drug as a molecular graph and uses a graph neural network to extract the drug features therein.

[0084] RDKit is an open-source cheminformatics and machine learning toolkit that converts SMILES expressions into graph data formats. Graph data consists of nodes and edges. Nodes contain entity information (such as atoms in a drug molecule), and edges contain relationship information between entities (such as chemical bonds between atoms in a drug molecule). By treating each atom as a node in the graph and atomic bonds as edges, a molecule can be viewed as a graph.

[0085] Among them, in the graph structure data of the drug, the nodes have the attribute characteristics of the atoms, and the edges have the chemical bond attribute characteristics corresponding to the chemical bonds connecting the atoms, which collectively constitute the drug characteristics of the drug.

[0086] In one embodiment, reference Figure 4 As shown, the method of obtaining the molecular graph of the drug may specifically include the following steps S41-S42:

[0087] S41, obtaining a SMILES expression of the drug from a database;

[0088] S42. Input the SMILES expression into a pre-trained molecular graph model to obtain a molecular graph of the drug.

[0089] This embodiment uses a pre-trained molecular graph model to convert the SMILES expression of the drug into a molecular graph, thereby improving the conversion efficiency of the molecular graph. The molecular graph model is a neural network model, specifically a graph neural network (GNN) model.

[0090] In one embodiment, reference Figure 5 As shown, before inputting the SMILES expression into the pre-trained molecular graph model to obtain the molecular graph of the drug, the following steps S51-S54 may be included:

[0091] S51, obtaining a training set; wherein the training set includes standard SMILES expressions of multiple drugs and corresponding standard molecular graphs;

[0092] S52, training the neural network model according to the training set;

[0093] S53, obtaining the training result of the neural network model after training, and determining whether the training result meets the requirements;

[0094] S54: If yes, use the trained neural network model as a molecular graph model.

[0095] The training set of this embodiment includes standard SMILES expressions of multiple drugs and a standard molecular graph corresponding to each standard SMILES expression. Before using the training set to train the neural network model, the iteration conditions of the neural network model can be set. The iteration conditions include the number of training times or the training duration, etc. When the neural network model meets the iteration conditions, the training is terminated. At this time, the training results of the neural network model after the training are obtained, and it is determined whether the training results meet the requirements. If it is determined that the training results meet the requirements, the trained neural network model is used as a molecular graph model to convert the SMILES expression of the drug into the corresponding molecular graph.

[0096] In addition, the training set and the trained molecular graph model of the present invention can be stored in the blockchain, so that the server only needs to broadcast the blockchain and call the trained molecular graph model. The blockchain is a new application model of computer technologies such as distributed data storage, point-to-point transmission, consensus mechanism, and encryption algorithm. The blockchain is essentially a decentralized database, a string of data blocks generated by cryptographic methods. Each data block contains information about a batch of network transactions, which is used to verify the validity of its information (anti-counterfeiting) and generate the next block. The blockchain can include the blockchain underlying platform, the platform product service layer, and the application service layer.

[0097] In one embodiment, reference Figure 6 As shown, the determination of whether the training result meets the requirements may specifically include the following steps S61-S64:

[0098] S61, calculating the loss value of the trained neural network model according to the training results and a preset loss function;

[0099] S62, determining whether the loss value is lower than a preset loss value;

[0100] S63: If yes, determine that the training result meets the requirements;

[0101] S64: If not, determine that the training result does not meet the requirements.

[0102] In this embodiment, after the neural network model is trained, the preset cross-entropy loss function can be used to calculate the loss value of the neural network model after training. When the loss value meets the preset threshold or is less than the preset loss value, that is, the training result of the neural network model meets the requirements, it indicates that the neural network model meets the training requirements and the training of the molecular graph model is completed, so as to improve the conversion accuracy of the molecular graph model to the SMILES expression.

[0103] Among them, the cross entropy loss function is used to evaluate the degree to which the predicted value of the molecular graph model is different from the true value. The better the loss function, the better the performance of the molecular graph model. In addition, the cross entropy loss function is often used in classification problems. In particular, when neural networks are used for classification problems, cross entropy is often used as a loss function. Since cross entropy involves calculating the probability of each category, cross entropy almost always appears together with the sigmoid (or softmax) function. In addition, the loss function of this embodiment is not specifically limited. For example, it can be a mean square error function, a covariance function, etc. In addition, the preset loss value of this embodiment can be determined according to actual conditions, and the preset loss value is different from the loss threshold corresponding to when the molecular graph model is finally trained. Generally, the preset loss value here is greater than the loss threshold corresponding to when the molecular graph model is finally trained.

[0104] When the loss value of the molecular graph model is not less than the preset loss value, forward transmission can be performed in the neural network structure of the molecular graph model according to the loss value, and the relevant parameters of the molecular graph model can be adjusted. The reference image is input to the molecular graph model with the relevant parameters reset to retrain until the loss value of the molecular graph model is less than the preset loss value. At this point, the training of the molecular graph model is completed, and a molecular graph model whose training results meet the requirements is obtained, thereby obtaining a trained molecular graph model.

[0105] In one embodiment, reference Figure 7 As shown, the predicting of the similarity of drug features between two adjacent nodes in the molecular graph may specifically include the following steps S71-S73:

[0106] S71. Converting drug features between two adjacent nodes in the molecular graph into vectors to obtain a first vector and a second vector corresponding to the two adjacent nodes;

[0107] S72. Calculate the cosine value of the angle between the first vector and the second vector;

[0108] S73. Determine the similarity of drug features between two adjacent nodes in the molecular graph according to the cosine value.

[0109] In this embodiment, a common method for measuring the similarity of drug features between two adjacent nodes in a molecular graph is to calculate the cosine value between the drug features. The cosine value can reflect the difference between two vectors in space. Two similar semantic relationships are clustered to complete the clustering of all semantic relationships. The most clustered semantic relationship is then selected as the semantic recognition result of the drug feature. For example, if most semantic relationships are clustered in region A, the semantic relationship closest to the center of region A is selected as the semantic recognition result.

[0110] When the drug features are described in text form, this embodiment can use the Word2Vec word vector model to convert the drug features between two adjacent nodes in the molecular graph into vectors respectively, obtain the first vector and the second vector corresponding to the two adjacent nodes, and then calculate the cosine value of the first vector and the second vector, and use the cosine value as the similarity. Among them, the Word2Vec word vector model is a model that learns semantic knowledge from a large amount of text in an unsupervised manner. It trains a large amount of text and represents the words in the text in the form of vectors. We call this vector a word vector. By calculating the distance between the word vectors of two words, we can know the connection between the two words.

[0111] Reference Figure 8 In an embodiment of the present invention, a device for predicting drug-target interaction is provided, comprising:

[0112] An acquisition module 11 is used to acquire a molecular graph of a drug;

[0113] A first calling module 12 is used to call a pre-built graph neural network to extract drug features in the molecular graph;

[0114] A prediction module 13 is configured to process the molecular graph according to a restarted random walk algorithm, predict the similarity of drug features between two adjacent nodes in the molecular graph, and obtain global structural information; wherein the global structural information is used to describe the similarity of drug features between two adjacent nodes in the molecular graph;

[0115] A first input module 14 is configured to input the global structural information into a preset deep neural network to obtain low-dimensional feature information of the drug;

[0116] The second calling module 15 is used to obtain the target protein sequence, call the long short-term memory network to process the protein sequence, and obtain protein features;

[0117] The second input module 16 is used to input the low-dimensional feature information and protein features into a preset fully connected layer to predict the interaction result between the drug and the target.

[0118] The drug is a drug whose interaction with the target needs to be predicted. The molecular graph of the drug can be converted through the RDKit toolkit. The molecular graph of the drug uses the atoms that constitute the drug as nodes and the connections between atoms as edges. The molecular graph can be constructed and stored in the form of a two-dimensional adjacency matrix. The rows and columns of the matrix represent atoms, and the values ​​of the matrix represent whether there is a connection between two atoms. For example, the feature of each node is a 78-bit vector encoding, of which 11 bits represent the type of atom, 11 bits represent the degree of the atomic node, 11 bits represent the total number of connections between the atom and hydrogen atoms in the compound, 11 bits represent the number of hidden hydrogen atoms connected to the atom, and the remaining 1 bit indicates whether the atom is on an aromatic ring.

[0119] The present invention can call a pre-built graph neural network to extract drug features in a molecular graph, where drug features are the molecular features of nodes in the molecular graph. Among them, a graph neural network (GNN) is a type of model for deep learning on graph structure data. The input of a graph neural network is usually a graph structure, and its final output generally depends on the specific task. Taking graph property prediction as an example, the graph neural network trains the implicit vector representation of each node in the graph based on the graph structure and input node attributes. Its goal is to make the vector representation contain sufficiently powerful expression information so that it can help each node extract information. Finally, through methods such as average pooling, the information vector representation of the entire graph can be obtained.

[0120] After converting the drugs into a molecular graph, the graph is processed using a restarted random walk algorithm, originally developed for image segmentation. The restarted random walk algorithm iteratively explores the overall structure of a molecular graph to estimate the similarity, or affinity, between two nodes. Starting from a node, each step presents two choices: move to a randomly selected adjacent node or jump back to the starting node. The algorithm incorporates a single fixed parameter, r, called the "restart probability" (1-r represents the probability of moving to an adjacent node). After the iterations reach stability, the stable probability vector contains the similarity of all nodes in the molecular graph to the starting node. The resulting probability distribution can be viewed as a distribution influenced by the starting node.

[0121] This embodiment can capture the multifaceted relationships between two nodes in the molecular graph by restarting the random walk algorithm, and then obtain global structural information describing the similarity of drug characteristics between two adjacent nodes in the entire molecular graph, which is then fed into a preset deep neural network to extract low-dimensional feature information of the drug.

[0122] For the target protein sequence, the present invention uses a long short-term memory network (LSTM) with a self-attention mechanism for processing. The linear self-attention mechanism method is used to reduce the running speed and memory consumption, and at the same time constructs an unbiased estimator for the attention matrix to improve the effect of protein sequence processing.

[0123] Among them, the long short-term memory network contains two outputs: the output of all time steps O = [O_1, O_2,…, O_D] and the hidden state H_D of the last time step D.

[0124] Since O = [O_1, O_2, ..., O_D] represents the characteristics of a word / phrase, and H_D represents the characteristics of a text, in order to identify the importance of a word to a text, it is necessary to establish a self-attention relationship between H_D and O, that is, to establish the weight of each time step output O_t for H_D. Since the long short-term memory network itself takes position information into account, there is no need to set up an additional position encoding. The implementation method of the self-attention mechanism in the long short-term memory network is additive attention: the hidden state of the last time step and the output of each time step are concatenated as the query, the output of each time step is linearly transformed as the value, the linear transformation matrix is ​​used as the key, and the query and value are multiplied to form the result matrix z. The present invention uses a linear attention mechanism to decompose the attention matrix into the product of a random nonlinear function of the original query and the key, namely the so-called random feature. In this way, the similarity can be encoded more efficiently, and a scalable low-variance, unbiased estimation of the attention mechanism can be provided. While maintaining linear space and time complexity, the accuracy is also guaranteed.

[0125] The present invention extracts low-dimensional feature information from drugs and protein features from proteins respectively, connects these features and inputs them into a preset fully connected layer, and then predicts the interaction results between the drug and the target based on the known drug-target interaction pairs.

[0126] As described above, it can be understood that the various components of the drug-target interaction prediction device proposed in the present invention can realize the functions of any of the drug-target interaction prediction methods described above, and the specific structure will not be repeated.

[0127] Reference Figure 9 The present invention also provides a computer device in an embodiment, the internal structure of which can be as follows Figure 9As shown. The computer device includes a processor, a memory, a network interface and a database connected via a system bus. The computer-designed processor is used to provide computing and control capabilities. The memory of the computer device includes a computer-readable storage medium and an internal memory. The computer-readable storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the computer-readable storage medium. The database of the computer device is used to store relevant data of the method for predicting the interaction between the drug and the target. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a method for predicting the interaction between the drug and the target is implemented.

[0128] The processor executes the above-mentioned drug-target interaction prediction method, including:

[0129] Obtaining molecular maps of drugs;

[0130] Calling a pre-built graph neural network to extract drug features in the molecular graph;

[0131] Processing the molecular graph according to a restarted random walk algorithm to predict the similarity of drug features between two adjacent nodes in the molecular graph to obtain global structural information; wherein the global structural information is used to describe the similarity of drug features between two adjacent nodes in the molecular graph;

[0132] Inputting the global structural information into a preset deep neural network to obtain low-dimensional feature information of the drug;

[0133] Obtaining a target protein sequence, and processing the protein sequence using a long short-term memory network to obtain protein features;

[0134] The low-dimensional feature information and protein features are input into a preset fully connected layer to predict the interaction result between the drug and the target.

[0135] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, a method for predicting drug-target interactions is implemented, comprising the steps of:

[0136] Obtaining a molecular map of a drug;

[0137] Calling a pre-built graph neural network to extract drug features in the molecular graph;

[0138] Processing the molecular graph according to a restarted random walk algorithm to predict the similarity of drug features between two adjacent nodes in the molecular graph to obtain global structural information; wherein the global structural information is used to describe the similarity of drug features between two adjacent nodes in the molecular graph;

[0139] Inputting the global structural information into a preset deep neural network to obtain low-dimensional feature information of the drug;

[0140] Obtaining a target protein sequence, and processing the protein sequence using a long short-term memory network to obtain protein features;

[0141] The low-dimensional feature information and protein features are input into a preset fully connected layer to predict the interaction result between the drug and the target.

[0142] Those skilled in the art will appreciate that all or part of the processes in the above-described embodiment methods can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When executed, the computer program can include the processes of the above-described embodiment methods. Among them, any reference to memory, storage, database or other media provided by the present invention and used in the embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct RAM bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM).

[0143] In summary, the greatest beneficial effects of the present invention are:

[0144] The present invention provides a method, device, equipment and storage medium for predicting drug-target interactions. The method obtains a molecular graph of the drug, calls a pre-built graph neural network to extract drug features in the molecular graph, processes the molecular graph according to a restarted random walk algorithm, predicts the similarity of drug features between two adjacent nodes in the molecular graph, obtains global structural information, inputs the global structural information into a preset deep neural network, obtains low-dimensional feature information of the drug, obtains the protein sequence of the target, calls a long short-term memory network to process the protein sequence, obtains protein features, inputs the low-dimensional feature information and protein features into a preset fully connected layer to predict the results of the drug-target interaction, combines the characteristics of the graph neural network and the long short-term memory network to deal with the problem of drug-target interaction, and specifically extracts the information contained in the drug and protein sequences, thereby improving the prediction efficiency of drug-target interaction and reducing costs. In addition, by using the restarted random walk algorithm to capture the various relationships between different nodes in the molecular graph, global structural information of similarity is obtained, which helps to improve prediction accuracy and prediction performance.

[0145] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, apparatus, article, or method comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, apparatus, article, or method. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, apparatus, article, or method comprising the element.

[0146] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for predicting drug-target interaction, characterized in that: The method comprises: Obtaining a molecular map of a drug; Calling a pre-built graph neural network to extract drug features in the molecular graph; Processing the molecular graph according to a restarted random walk algorithm to predict the similarity of drug features between two adjacent nodes in the molecular graph to obtain global structural information; wherein the global structural information is used to describe the similarity of drug features between two adjacent nodes in the molecular graph; Inputting the global structural information into a preset deep neural network to obtain low-dimensional feature information of the drug; Obtaining a target protein sequence, and processing the protein sequence using a long short-term memory network to obtain protein features; Inputting the low-dimensional feature information and protein features into a preset fully connected layer to predict the interaction result between the drug and the target; The step of inputting the low-dimensional feature information and protein features into a preset fully connected layer to predict the interaction between the drug and the target includes: Inputting the low-dimensional feature information and protein features into a preset fully connected layer to obtain a Boolean value output by the fully connected layer; Determining the interaction result between the drug and the target according to the Boolean value; The step of predicting the similarity of drug features between two adjacent nodes in the molecular graph includes: Converting the drug features between two adjacent nodes in the molecular graph into vectors to obtain a first vector and a second vector corresponding to the two adjacent nodes; Calculating the cosine of the angle between the first vector and the second vector; The similarity of drug features between two adjacent nodes in the molecular graph is determined according to the cosine value.

2. The method according to claim 1, characterized in that The obtaining of the molecular graph of the drug comprises: Obtaining the SMILES expression of the drug from the database; The SMILES expression of the drug was converted into a molecular graph using the RDK it tool, with each atom of the drug as a node and the element bonds between atoms as edges.

3. The method according to claim 1, characterized in that The obtaining of the molecular graph of the drug comprises: Obtaining the SMILES expression of the drug from the database; The SMILES expression is input into a pre-trained molecular graph model to obtain the molecular graph of the drug.

4. The method according to claim 3, characterized in that Before inputting the SMILES expression into a pre-trained molecular graph model to obtain the molecular graph of the drug, the method further includes: Obtaining a training set; wherein the training set includes standard SMILES expressions of multiple drugs and corresponding standard molecular graphs; Training the neural network model according to the training set; Obtaining the training results of the neural network model after training, and determining whether the training results meet the requirements; If so, the trained neural network model is used as a molecular graph model.

5. The method according to claim 4, characterized in that Determining whether the training result meets the requirements includes: Calculating the loss value of the trained neural network model based on the training results and a preset loss function; Determining whether the loss value is lower than a preset loss value; If so, determining that the training result meets the requirements; If not, it is determined that the training result does not meet the requirements.

6. A device for predicting drug-target interactions, characterized in that: The device comprises: An acquisition module, used to obtain the molecular graph of the drug; A first calling module is used to call a pre-built graph neural network to extract drug features in the molecular graph; a prediction module, configured to process the molecular graph according to a restarted random walk algorithm, predict the similarity of drug features between two adjacent nodes in the molecular graph, and obtain global structural information; wherein the global structural information is used to describe the similarity of drug features between two adjacent nodes in the molecular graph; A first input module is used to input the global structure information into a preset deep neural network to obtain low-dimensional feature information of the drug; The second calling module is used to obtain the target protein sequence, call the long short-term memory network to process the protein sequence, and obtain protein features; A second input module is used to input the low-dimensional feature information and protein features into a preset fully connected layer to predict the interaction result between the drug and the target; The step of inputting the low-dimensional feature information and protein features into a preset fully connected layer to predict the interaction between the drug and the target includes: Inputting the low-dimensional feature information and protein features into a preset fully connected layer to obtain a Boolean value output by the fully connected layer; Determining the interaction result between the drug and the target according to the Boolean value; The step of predicting the similarity of drug features between two adjacent nodes in the molecular graph includes: Converting the drug features between two adjacent nodes in the molecular graph into vectors to obtain a first vector and a second vector corresponding to the two adjacent nodes; Calculating the cosine of the angle between the first vector and the second vector; The similarity of drug features between two adjacent nodes in the molecular graph is determined according to the cosine value.

7. A computer device, characterized in that: include: processor; Memory; Wherein, the memory stores a computer program, and when the processor executes the computer program, it implements the drug-target interaction prediction method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the method for predicting drug-target interaction according to any one of claims 1 to 5.

Images