Traditional Chinese medicine target relationship prediction method based on multi-attention graph neural network learning

Abstract

The present application belongs to the field of bioinformatics, and specifically provides a traditional Chinese medicine target relationship prediction method based on multi-attention graph neural network learning. First, a heterogeneous network containing multiple entities such as traditional Chinese medicine, targets, diseases, and prescriptions and multiple edge types is constructed to better describe the relationship between traditional Chinese medicine and targets. Then, a method based on multi-attention graph neural network is proposed to learn the fusion feature representation of traditional Chinese medicine and targets based on the heterogeneous network. Finally, the traditional Chinese medicine-target relationship is predicted based on the deep neural network (DNN). The present application uses multi-attention graph neural network to perform fusion feature representation of traditional Chinese medicine and targets based on multi-source heterogeneous data and heterogeneous relationships, considers more comprehensive information, and effectively improves the prediction accuracy while predicting the traditional Chinese medicine-target relationship.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of bioinformatics, and specifically provides a traditional Chinese medicine target relationship prediction method based on multi-attention graph neural network learning. BACKGROUND

[0002] Accurate identification of potential traditional Chinese medicine-target interactions is a key step in traditional Chinese medicine discovery. In the past few decades, many researchers have conducted numerous biological experiments to identify traditional Chinese medicine-target interactions. However, traditional Chinese medicine-target interaction identification based on biological experiments is time-consuming and expensive. Therefore, it is of great significance to develop an effective traditional Chinese medicine-target interaction identification and prediction method.

[0003] Existing computational drug-target interaction identification methods can be mainly divided into three categories: text mining-based methods, biological feature-based methods, and network-based methods. Among them, text mining-based methods extract information from literature and use drug and target descriptions as features to identify drug-target interactions. For example, a semantic similarity-based model using random forest and support vector machine methods is proposed to identify drug-target interactions. Biological feature-based methods extract biological features of drugs and targets and use machine learning models to identify drug-target interactions. For example, an existing technology based on chemical and genomic data of drug-drug, target-target similarity uses a support vector machine method to convert the drug-target interaction identification problem into a binary classification problem. Network-based methods can describe complex interactions between different types of biological entities (such as drugs and proteins). For example, an existing technology integrates various drug-related information and protein-related information to calculate the similarity matrix of drugs and proteins through multiple networks. Then, the similarity matrix is sequentially subjected to restart random walk and singular value decomposition to obtain low-dimensional feature representations of drugs and proteins to identify drug-target interactions. As can be seen, text mining-based methods are affected by differences in semantic expression, and conflicts between different literature limit their performance. Moreover, existing methods do not consider more comprehensive information, such as traditional Chinese medicine-disease associations, traditional Chinese medicine interactions, prescription-traditional Chinese medicine relationships, and disease-protein associations. Typically, features based on each interaction type are extracted and connected together, so these methods cannot model heterogeneous information in a single heterogeneous network. At the same time, existing methods do not consider the inconsistency of feature weights for the same node in different relationships. In addition, existing methods mainly consist of two independent steps: feature extraction and traditional Chinese medicine-target interaction prediction. In the traditional Chinese medicine-target interaction prediction task, the parameters involved in the feature extraction step cannot be optimized, which limits the prediction performance. SUMMARY

[0004] The purpose of the present application is to propose a traditional Chinese medicine target relationship prediction method based on a multi-attention graph neural network, and an isomer network containing various biological entities (including traditional Chinese medicine, target, disease, and prescription) is creatively designed. A multi-attention graph neural network-based model is used to learn the low-dimensional feature representation of drugs and targets, and the learned features are used to predict traditional Chinese medicine-target interactions.

[0005] To achieve the above-mentioned purpose, the technical scheme adopted by the present application is:

[0006] A traditional Chinese medicine target relationship prediction method based on multi-attention graph neural network learning, characterized by the following steps:

[0007] Step 1. Collect data and build an isomer network;

[0008] Obtain prescription data, traditional Chinese medicine data, target data, and disease data from the ETCM database, and build a traditional Chinese medicine-target association network A mt , a traditional Chinese medicine-disease association network A md , a traditional Chinese medicine-prescription association network A mf , a traditional Chinese medicine interaction network A mm , a target interaction network A tt , a traditional Chinese medicine-traditional Chinese medicine similarity network S mm , a target-disease association network A td , a target-target similarity network S tt There are eight networks in total, and they are combined to form an isomer network;

[0009] Step 2. Initialize the feature vector of the traditional Chinese medicine node of the isomer network;

[0010] Step 3. Calculate the similarity of the traditional Chinese medicine node and the target node of the isomer network respectively, and obtain the adjacency matrix of the isomer network;

[0011] Step 4. Build a graph convolutional neural network model based on the graph attention network GAT, and extract the node feature representation of each node in the isomer network at each layer of the graph convolutional layer according to the adjacency matrix;

[0012] Step 5. Use the self-attention mechanism to perform multi-layer graph convolution fusion on the node feature representation of the isomer network, and obtain the fusion feature representation of each traditional Chinese medicine node and target node in the isomer network;

[0013] Step 6. Concatenate the fusion feature representation of the traditional Chinese medicine node and the target node to be predicted to form a traditional Chinese medicine-target feature vector and input it into the traditional Chinese medicine-target relationship prediction model, and output the traditional Chinese medicine target relationship prediction result from the model.

[0014] Further, in step 2, the feature data of the traditional Chinese medicine node includes nature, taste, meridian tropism and toxicity, specifically "cold, hot, warm, cool, neutral; sour, bitter, sweet, pungent, salty; lung, pericardium, heart, large intestine, triple energizer, small intestine, stomach, gallbladder, bladder, spleen, liver, kidney; toxicity";

[0015] For the description of the modifier class feature, 2 -1 , 2 0 and 2 1 are used for quantitative representation; for the attribution class feature, binary is used for quantitative representation, thereby obtaining the initial feature vector of each traditional Chinese medicine node.

[0016] Further, in step 3, the similarity calculation process of the traditional Chinese medicine node is as follows:

[0017] For any two traditional Chinese medicine nodes, the cosine similarity of the two is calculated according to the feature vector, specifically represented as:

[0018]

[0019] Wherein, and respectively represent the feature vector of the traditional Chinese medicine node a and the traditional Chinese medicine node b, and k represents the feature serial number in the feature vector;

[0020] Set the similarity threshold θ1 of the traditional Chinese medicine node, if sim(a, b) > θ1, it is determined that the traditional Chinese medicine node a and the traditional Chinese medicine node b have similarity, otherwise, they do not have similarity.

[0021] Further, in step 3, the similarity calculation process of the target node is as follows:

[0022] For any two targets, the target similarity is calculated by BLAST;

[0023] Set the target similarity threshold θ2, if the target similarity is greater than the threshold θ2, it is determined that the two targets have similarity, otherwise, they do not have similarity.

[0024] Further, in step 3, the adjacency matrix is specifically represented as:

[0025] Corresponding to the traditional Chinese medicine-target association network A mt , the traditional Chinese medicine-disease association network A md , the traditional Chinese medicine-prescription association network A mf , the traditional Chinese medicine interaction network A mm , the target interaction network A tt and the target-disease association network A td , the original association relationship is represented as a [0, 1] matrix, 0 represents no association relationship, and 1 represents an association relationship; corresponding to the traditional Chinese medicine-traditional Chinese medicine similarity network S mmTarget-target similarity network S tt The similarity relationship is expressed as a [0, 1] matrix, 0 indicating no similarity, and 1 indicating similarity.

[0026] Further, in step 4, the graph convolutional neural network model adopts a three-layer stacked structure, wherein the feature representation of each layer of graph convolutional layer is:

[0027]

[0028] wherein i represents node information, l represents layer information in the graph convolutional neural network, r represents edge relationship type information, and a ij represents an attention coefficient, represents a normalization constant, represents a symmetric square root normalization constant, and φ(·) represents a Relu activation function. represents the neighborhood set of node i with respect to edge relationship r, represents the neighborhood set of node j with respect to edge relationship r. represents the training parameter of the lth layer of graph convolutional layer with edge relationship r, represents the feature representation of node i in the lth layer of graph convolutional layer, represents the neighborhood feature representation of node j in the lth layer of graph convolutional layer, and the original feature representation of each node is initialized as a one-hot vector. represents the feature representation of node i in the (l+1)th layer of graph convolutional layer;a (l) represents the training parameter of the lth layer of graph convolutional layer, and σ(·) represents a sigmoid activation function.

[0029] Further, in step 5, the query vector, key vector and value vector in the self-attention mechanism are defined as:

[0030]

[0031] wherein q l , k l , and v l represent the query vector, key vector and value vector of the attention mechanism, respectively, W Q , and W K represent the training parameters of the self-attention mechanism.

[0032] For any two layers of graph convolutional layer, the attention coefficient is represented as:

[0033]

[0034] wherein e mn represents the attention coefficient of the mth layer of graph convolutional layer and the nth layer of graph convolutional layer, m, n ∈ (1, 2, 3).

[0035] Then, for node i, the multi-layer graph convolution fusion feature is represented as:

[0036]

[0037] wherein, represents the feature representation of node i in the lth layer of the graph convolutional layer of the graph convolutional neural network, l e (1, 2, 3) ;

[0038] Thus, the fusion feature representation of each traditional Chinese medicine node and each target node is obtained.

[0039] Further, in step 6, the traditional Chinese medicine-target relationship prediction model adopts a three-layer DNN network.

[0040] Based on the above technical solution, the beneficial effects of the present application are:

[0041] The present application provides a method for predicting the relationship between traditional Chinese medicine and targets based on a multi-attention graph neural network. First, a heterogeneous network containing multiple entities such as traditional Chinese medicine, targets, diseases, and prescriptions and multiple edge types is constructed to better describe the relationship between traditional Chinese medicine and targets. Then, a method based on a multi-attention graph neural network is proposed to learn the fusion feature representation of traditional Chinese medicine and targets based on the heterogeneous network. Finally, the relationship between traditional Chinese medicine and targets is predicted based on a deep neural network (DNN).

[0042] Compared with the prior art, the present application uses multi-source heterogeneous data and heterogeneous relationships, calculates the similarity relationship, and considers more comprehensive information such as traditional Chinese medicine-disease association, traditional Chinese medicine interaction, target interaction, prescription-traditional Chinese medicine relationship, traditional Chinese medicine-target association, and disease-target association. The graph neural network is used to learn the node feature representation of the heterogeneous graph, and the heterogeneous information in a single heterogeneous network is modeled and the node weight in different relationships is considered. Finally, the node feature learning is performed through the self-attention mechanism to predict the traditional Chinese medicine-target, and the prediction accuracy is improved by training the deep neural network. In addition, the final traditional Chinese medicine-target interaction prediction task can also optimize the parameters involved in the feature extraction step and improve the model performance. BRIEF DESCRIPTION OF DRAWINGS

[0043] Figure 1 The figure is a flowchart of the method for predicting the relationship between traditional Chinese medicine and targets based on a multi-attention graph neural network learning in the present application.

[0044] Figure 2 The figure is a model structure diagram of the method for predicting the relationship between traditional Chinese medicine and targets based on a multi-attention graph neural network learning in the present application. DETAILED DESCRIPTION

[0045] In order to make the objectives, technical solutions and beneficial effects of the present application clearer, further detailed description will be made to the present application in combination with the drawings and examples.

[0046] The present embodiment provides a traditional Chinese medicine target point relationship prediction method based on multi-attention graph neural network learning, the flow thereof is as shown in Figure 1 The model structure thereof is as shown in Figure 2 The specific steps include the following steps:

[0047] Step 1. Collecting data and constructing a heterogeneous network;

[0048] Obtaining prescription data, traditional Chinese medicine data, target point data and disease data from ETCM database, and constructing eight networks including traditional Chinese medicine-target point association network A mt , traditional Chinese medicine-disease association network A md , traditional Chinese medicine-prescription association network A mf , traditional Chinese medicine-traditional Chinese medicine interaction network A mm , target point-target point interaction network A tt , traditional Chinese medicine-traditional Chinese medicine similarity network S mm , target point-disease association network A td , target point-target point similarity network S tt , and combining them to form a heterogeneous network;

[0049] Step 2. Initializing feature vectors of traditional Chinese medicine nodes of the heterogeneous network;

[0050] Statistically analyzing the characteristics of traditional Chinese medicines, and initializing the characteristics of traditional Chinese medicines; the traditional characteristics of traditional Chinese medicines include nature, taste, meridian tropism and toxicity, specifically “cold, hot, warm, cool, neutral; sour, bitter, sweet, pungent, salty; lung, pericardium, heart, large intestine, triple energizer, small intestine, stomach, gallbladder, bladder, spleen, liver, kidney; toxicity”, a total of 23 characteristics; quantifying all the characteristics, for the characteristics described with modifiers, for example, “mild cold”, “cold” and “severe cold”, 2 -1 , 2 0 and 2 1 are used for quantification respectively; for the attribution type characteristics, binary quantification is adopted, that is, 1 represents the existence of the characteristic, and 0 represents the non-existence of the characteristic; thus, the initialized feature vector of each traditional Chinese medicine node is obtained;

[0051] Step 3. Calculating the similarity of traditional Chinese medicine nodes and target points of the heterogeneous network respectively, and obtaining the adjacency matrix of the heterogeneous network;

[0052] For any two traditional Chinese medicine nodes, the cosine similarity of the two nodes is calculated according to the feature vectors, which is specifically represented as:

[0053]

[0054] wherein, sim(a,b) represents the similarity between traditional Chinese medicine node a and traditional Chinese medicine node b, k represents the feature sequence number in the feature vector; sim(a,b) represents the similarity between traditional Chinese medicine node a and traditional Chinese medicine node b, k represents the feature sequence number in the feature vector;

[0055] A similarity threshold θ1 is set for traditional Chinese medicine nodes. If sim(a,b) > θ1, it is determined that traditional Chinese medicine node a and traditional Chinese medicine node b have similarity, otherwise, they do not have similarity;

[0056] For any two target points, the BLAST is used to calculate the target point similarity. The similarity between two protein sequences is identified by finding local sequence similarity. The higher the similarity score output by BLAST, the higher the similarity of the two target points;

[0057] A target point similarity threshold θ2 is set. If the target point similarity is greater than the threshold θ2, it is determined that the two target points have similarity, otherwise, they do not have similarity;

[0058] According to the similarity determination of the above traditional Chinese medicine nodes and target points and the original association relationship, the adjacency matrix of the heterogeneous network is constructed; wherein, corresponding to the traditional Chinese medicine-target point association network A mt , the traditional Chinese medicine-disease association network A md , the traditional Chinese medicine-prescription association network A mf , the traditional Chinese medicine interaction network A mm , the target point interaction network A tt and the target point-disease association network A td , the original association relationship is represented as a [0,1] matrix, 0 represents no association relationship, and 1 represents an association relationship; corresponding to the traditional Chinese medicine-traditional Chinese medicine similarity network S mm and the target point-target point similarity network S tt , the similarity relationship is represented as a [0,1] matrix, 0 represents no similarity, and 1 represents similarity;

[0059] Step 4. Construct a graph convolutional neural network model based on a graph attention network GAT, and extract the node feature representation of each node in the heterogeneous network at each layer of the graph convolutional layer according to the adjacency matrix;

[0060] The key step of identifying traditional Chinese medicine-target relationship lies in feature extraction, and the heterogeneous graph convolutional neural network generates feature representations of traditional Chinese medicines and targets in each layer; the heterogeneous network constructed in the application has four types of nodes: traditional Chinese medicines, targets, diseases and prescriptions, and contains eight types of edge relationships: traditional Chinese medicine- traditional Chinese medicine interaction, target-target interaction, traditional Chinese medicine-target association, traditional Chinese medicine-disease association, traditional Chinese medicine-prescription association, target-disease association, traditional Chinese medicine-traditional Chinese medicine similarity and target-target similarity; the heterogeneous graph convolutional neural network based on the graph attention network GAT is used for message passing on different types of edges to realize feature aggregation between related nodes and extract features of traditional Chinese medicines and targets in the heterogeneous network; when the number of graph convolutional neural network layers is only one layer, the node feature representation obtained by the model will only aggregate the first-order information of its neighbors, and the stacked multi-layer graph convolutional neural network can effectively convolve the high-order neighbor information of the feature representation, therefore, the three-layer stacked graph convolutional neural network is adopted in the application, and in each layer, the feature representation of each node is generated by aggregating the features of adjacent nodes connected by different types of edges;

[0061] Specifically, in each graph convolutional layer of the graph convolutional neural network, the type information of the edge is modeled as:

[0062]

[0063] Wherein, i represents node information, l represents layer information in the graph convolutional neural network, and r represents edge relationship type information, represents a normalization constant, represents a symmetric square root normalization constant, and φ(·) represents a Relu activation function; represents the neighborhood set of node i with respect to edge relationship r, and similarly, represents the neighborhood set of node j with respect to edge relationship r; represents the training parameter of the lth layer graph convolutional layer with edge relationship r, represents the feature representation of node i in the lth layer graph convolutional layer, represents the neighborhood feature representation of node j in the lth layer graph convolutional layer, and the original feature representation of each node is initialized as a one-hot vector; represents the feature representation of node i in the l+1th layer graph convolutional layer, i.e., the feature representation after feature aggregation;

[0064] GAT gives different weights to adjacent nodes in the node information aggregation process, i.e., different weights are given to different nodes in the heterogeneous network, GAT respectively performs linear transformation on the neighbor nodes, and then calculates the attention coefficient by assigning values to the nodes in the graph, therefore, GAT is used to strengthen the feature learning of the nodes; the attention coefficient is specifically represented as:

[0065]

[0066] wherein, α ij represents an attention coefficient, a (l) represents a training parameter of the lth layer graph convolution layer, and σ(·) represents a sigmoid activation function; therefore, the graph convolution layer based on GAT can be finally represented as:

[0067]

[0068] After the above steps, the feature representation of each layer of each node in the heterogeneous network can be obtained;

[0069] Step 5. Using a self-attention mechanism to perform multi-layer graph convolution fusion on the node feature representation of the heterogeneous network to obtain the fusion feature representation of each traditional Chinese medicine node and target node in the heterogeneous network;

[0070] In a conventional method, the output of the last layer of graph convolution layer is usually taken as the final feature representation of the traditional Chinese medicine node or the target node, or the multi-layer feature representations are spliced to be taken as the final feature representation of the traditional Chinese medicine node or the target node. Different from the conventional method, the self-attention mechanism is used in the present application to prevent information loss;

[0071] Specifically, in the self-attention mechanism, the query vector, the key vector and the value vector are defined as:

[0072]

[0073] wherein, q l , k l , v l represent the query vector, the key vector and the value vector of the attention mechanism respectively, W Q , W K represent the training parameters of the self-attention mechanism;

[0074] For the output of any two layers of graph convolution layer, the attention coefficient thereof is calculated as:

[0075]

[0076] wherein, e mn represents the attention coefficient of the mth layer of graph convolution layer and the nth layer of graph convolution layer, and m, n ∈ (1, 2, 3);

[0077] Therefore, for the node i, the feature representation after multi-layer graph convolution fusion is:

[0078]

[0079] wherein, represents the feature representation of node i in the lth layer of graph convolution layer of the graph convolution neural network, and l ∈ (1, 2, 3).

[0080] Thus, the fusion feature representation of each traditional Chinese medicine node and each target node is obtained;

[0081] Step 6. The fusion feature representation of the traditional Chinese medicine node to be predicted and the target node is spliced to form a traditional Chinese medicine-target feature vector and is sent into a traditional Chinese medicine-target relationship prediction model, and a traditional Chinese medicine-target relationship prediction result is output by the model;

[0082] After obtaining the fusion feature representation of the traditional Chinese medicine and the target, a deep neural network DNN is used to predict the traditional Chinese medicine-target relationship; specifically, given the fusion feature representation h i of the traditional Chinese medicine node and the fusion feature representation h j of the target node, they are spliced to form a traditional Chinese medicine-target feature vector X = [h i , h j ], and are taken as an input vector of a three-layer DNN model, and a traditional Chinese medicine-target relationship prediction result is output by the model;

[0083] In the three-layer DNN model, 256 nodes are constructed by using a ReLU function in the first layer, a ReLU is selected as an activation function in the second layer, and a sigmoid activation function is used in the last layer to realize binary classification; a binary cross-entropy is used as a loss function, and an RMSProp is used as an optimizer to complete training of the three-layer DNN model; in the loss function, when a predicted value and an actual value are equal, the loss is 0, otherwise, the loss is a positive number, and the greater the difference between the predicted value and the actual value, the greater the loss.

[0084] The beneficial effects of the present application will be described in detail in combination with experiments.

[0085] In this embodiment, known traditional Chinese medicine-target associations are taken as positive samples, the same number of unknown traditional Chinese medicine-target associations are randomly selected as negative samples, and a 10-fold cross-validation method is used for evaluation; for the positive samples, 10% of the entire labeled traditional Chinese medicine-target associations are randomly extracted as a test set, and the remaining 90% are taken as a training set; for the negative samples, the method of generating the training set and the test set is the same as that of generating the positive samples; the area under the receiver operating characteristic curve (AUROC) and the area under the precision-recall curve (AUPR) are used as the evaluation criteria of the model, and the performance results are shown in Tables 1 and 2; as shown in the tables, the AUROC and AUPR values of the present application reach 0.8569 and 0.8655 respectively on the traditional Chinese medicine-target relationship prediction task, proving the effectiveness of the present application.

[0086] Table 1

[0087]

[0088] Table 2

[0089]

[0090]

[0091] The above description is merely that of a specific implementation of the application, any feature in the present description is capable of being, unless specifically stated, replaced by other equivalent or similar features and / or replaced by a combination of some of the disclosed features. Any feature disclosed in the context of one aspect of the application is capable of being applicable to another aspect of the application, unless specifically stated.

Claims

1. A method for predicting target relationships of traditional Chinese medicine based on multi-attention graph neural network learning, characterized in that, The method comprises the following steps: Step 1. Collecting data and constructing a heterogeneous network; Obtaining prescription data, traditional Chinese medicine data, target data and disease data from ETCM database, and constructing traditional Chinese medicine-target correlation network A mt , traditional Chinese medicine-disease correlation network A md , traditional Chinese medicine-prescription correlation network A mf , traditional Chinese medicine interaction network A mm , target interaction network A tt , traditional Chinese medicine-traditional Chinese medicine similarity network S mm , target-disease correlation network A td , target-target similarity network S tt There are eight networks in total, and they are combined to form a heterogeneous network; Step 2. Initializing feature vectors of traditional Chinese medicine nodes in the heterogeneous network; Step 3. Calculating the similarity of traditional Chinese medicine nodes and target nodes in the heterogeneous network respectively to obtain an adjacency matrix of the heterogeneous network; Step 4. Constructing a graph convolutional neural network model based on a graph attention network GAT, and extracting node feature representations of each node in each layer of the graph convolutional layer in the heterogeneous network according to the adjacency matrix; the graph convolutional neural network model adopts a three-layer stacked structure, wherein the feature representation of each layer of the graph convolutional layer is: where i denotes node information, l denotes layer information in the graph convolutional neural network, r denotes edge relation type information, and a denotes attention coefficient information. ij denotes an attention coefficient, denotes a normalization constant, denotes a symmetric square root normalization constant, and φ(·) denotes a Relu activation function. denotes a neighborhood set of node i with respect to edge relation r, denotes a neighborhood set of node j with respect to edge relation r. denotes a training parameter of an l-th layer graph convolutional layer with edge relation r, denotes a feature representation of node i in the l-th layer graph convolutional layer, denotes a neighborhood feature representation of node j in the l-th layer graph convolutional layer, and the original feature representation of each node is initialized as a one-hot vector. denotes a feature representation of node i in an (l+1)-th layer graph convolutional layer;a (l) denotes a training parameter of an l-th layer graph convolutional layer, and σ(·) denotes a sigmoid activation function. Step 5. Using a self-attention mechanism to perform multi-layer graph convolution fusion on the node feature representation of the heterogeneous network to obtain the fusion feature representation of each traditional Chinese medicine node and target node in the heterogeneous network; the query vector, key vector and value vector in the self-attention mechanism are defined as: q l = W Q · v l , k l = W K · v l , l = 1, 2, 3 wherein q l , k l , and v l respectively represent the query vector, the key vector, and the value vector of the attention mechanism, W Q , and W K represent the training parameters of the self-attention mechanism; For any two layers of the graph convolutional layer, the attention coefficient is: wherein e mn denotes the attention coefficient of the mth graph convolution layer and the nth graph convolution layer, m, n ∈ (1, 2, 3); Therefore, for node i, the feature representation after multi-layer graph convolution fusion is: wherein, represents the feature representation of node i at the l-th layer of the graph convolutional layers of the graph convolutional neural network, l e (1, 2, 3); Thus, the fusion feature representation of each traditional Chinese medicine node and each target node is obtained; Step 6. Splicing the fusion feature representation of the traditional Chinese medicine node and the target node to be predicted to form a traditional Chinese medicine-target feature vector and inputting the traditional Chinese medicine-target feature vector into a traditional Chinese medicine-target relationship prediction model to output a traditional Chinese medicine-target relationship prediction result; the traditional Chinese medicine-target relationship prediction model adopts a three-layer DNN network.

2. The method of claim 1, wherein, In step 2, the feature data of the traditional Chinese medicine node includes nature, taste, meridian and toxicity, specifically "cold, hot, warm, cool, neutral; sour, bitter, sweet, pungent, salty; lung, pericardium, heart, large intestine, triple energizer, small intestine, stomach, gallbladder, bladder, spleen, liver, kidney; Toxicity"; For the description of the modifier, the 2 -1 , 2 0 and 2 1 are used to quantitatively represent; for the attribution of the class characteristics, the binary is used to quantitatively represent, thus obtaining the initial feature vector of each traditional Chinese medicine node.

3. The method of claim 1, wherein the method comprises: In step 3, the similarity calculation process of the traditional Chinese medicine node is: For any two traditional Chinese medicine nodes, the cosine similarity between the two nodes is calculated according to the feature vectors, and is specifically represented as: wherein, with respectively represent the feature vectors of traditional Chinese medicine node a and traditional Chinese medicine node b, k represents the feature serial number in the feature vector; Set the similarity threshold θ1 of the traditional Chinese medicine node, if sim(a, b) > θ1, it is determined that the traditional Chinese medicine node a and the traditional Chinese medicine node b have similarity, otherwise, they do not have similarity.

4. The method of claim 1, wherein the method comprises: In step 3, the similarity calculation process of the target node is: For any two targets, the target similarity is calculated by BLAST; Set the target similarity threshold θ2, if the target similarity is greater than the threshold θ2, it is determined that the two targets have similarity, otherwise, they do not have similarity.

5. The method of claim 1, wherein, In step 3, the adjacency matrix is specifically represented as: Corresponding to Traditional Chinese Medicine-Target Association Network A mt Traditional Chinese Medicine-Disease Association Network A md Traditional Chinese Medicine-Formula Association Network A mf Traditional Chinese Medicine Interaction Network A mm Target Interaction Network A tt Target-Disease Association Network A td The original association relationship is represented as a [0, 1] matrix, 0 indicating no association relationship, and 1 indicating an association relationship; Corresponding to Traditional Chinese Medicine-Traditional Chinese Medicine Similarity Network S mm Target-Target Similarity Network S tt The similarity relationship is represented as a [0, 1] matrix, 0 indicating no similarity, and 1 indicating similarity.

Images