Gene and drug association prediction method

Abstract

The embodiment of the invention provides a gene and drug association prediction method, and belongs to the technical field of medical care informatics, and the method specifically comprises the steps: constructing an association matrix of genes and drugs, and randomly initializing the feature vector of each gene and drug; the diffusion process of genes and drugs is simulated through thermonuclear, and the self-adaptive aggregation weights of gene nodes and drug nodes are dynamically adjusted; according to the adaptive aggregation weight, updating feature vectors of the gene nodes and the drug nodes by using the graph neural network and the information of the neighbor nodes; calculating prediction scores of genes and drugs, calculating a BPR loss function, and updating parameters of the initial model through back propagation of the BPR loss function; obtaining a prediction model until the model is fitted; and inputting the feature vectors of the gene nodes and the drug nodes in the updated unknown association pair into the prediction model to execute an inner product operation to obtain an association score matrix of the genes and the drugs. Through the scheme disclosed by the invention, the prediction efficiency, accuracy and adaptability are improved.

Description

Technical Field

[0001] The disclosed embodiments relate to the technical field of medical care informatics, and in particular to a method for predicting the association between genes and drugs. Background Art

[0002] Currently, the process of developing and approving new drugs is time-consuming, labor-intensive, and costly, with a high probability of failure. Drug repurposing, the use of existing therapies or drugs that have been discontinued to treat other diseases, is an effective strategy to improve efficiency. However, a major challenge lies in the scarcity of data on existing gene and drug interactions. Studies have shown that different drugs that are effective for the same disease have similar effects on gene expression; similarly, different cells with similar gene expression patterns respond similarly to the same drug.

[0003] It can be seen that there is an urgent need for a gene and drug association prediction method with high prediction efficiency, accuracy and adaptability. Summary of the invention

[0004] In view of this, the embodiments of the present disclosure provide a gene and drug association prediction method, which at least partially solves the problems of poor prediction efficiency, accuracy and adaptability in the prior art.

[0005] The present disclosure provides a method for predicting the association between genes and drugs, including:

[0006] Step 1, using the construction module of the initial model, construct the gene-drug association matrix according to the gene-drug association pairs in the database, randomly initialize the feature vectors of each gene and drug, and obtain the feature vector matrix of the gene and the feature vector matrix of the drug;

[0007] Step 2: Using the dynamic adjustment module of the initial model, based on the association matrix, the gene feature vector matrix, the drug feature vector matrix and the graph convolutional network framework, the heat kernel is used to simulate the diffusion process of genes and drugs, and the prior knowledge is integrated into the feature propagation process between nodes to dynamically adjust the adaptive aggregation weights of gene nodes and drug nodes.

[0008] Step 3: Using the update module of the initial model, the feature vectors of the gene nodes and drug nodes are updated using the graph neural network and the information of neighbor nodes according to the adaptive aggregation weights;

[0009] Step 4: Input the updated node feature vector into the initial model, calculate the prediction scores of genes and drugs using the inner product-based calculation method, calculate the BPR loss function based on the prediction score and data label, and update the parameters of the initial model through the back propagation of the BPR loss function;

[0010] Step 5, repeat steps 2 to 4 until the model is fitted and a prediction model is obtained;

[0011] Step 6: Input the updated feature vectors of the gene node and the drug node in the unknown association pair into the prediction model to perform an inner product operation to obtain the gene-drug association score matrix.

[0012] According to a specific implementation of the embodiment of the present disclosure, step 1 specifically includes:

[0013] Define the gene set G in the database as {g1, g2, ..., g m} and drug set D = {d1,d2,...,d n}, where m and n represent the number of genes and drugs, respectively, to construct the association matrix If the gene g i With drug d j If there is a known relationship between i,j =1, otherwise 0, for each gene node g i and drug node d j Assign d-dimensional feature vector z(g) i and z(d) j , forming the gene feature vector matrix and the drug eigenvector matrix

[0014] According to a specific implementation of the embodiment of the present disclosure, step 2 specifically includes:

[0015] The heat kernel simulates the diffusion process of genes and drugs, and integrates prior knowledge into the feature propagation process between nodes, expressed as a differential equation

[0016]

[0017] in, and Represents the gene node g at time t i and drug node d j The characteristics of g, N(i) is i The set of neighbor nodes of

[0018] Solve the differential equation to get the node g after diffusion time t i Features:

[0019]

[0020] Among them, H t =e -(I-K)t , H t represents the heat kernel, I is the identity matrix, and K is the Laplace matrix of the graph G;

[0021] Use the hot kernel to convolve the feature vector of the gene node

[0022]

[0023] in, is the normalized adjacency matrix calculated based on the gene-drug association matrix A, in the form of D (g) and D (d) is the angle matrix, the adaptive parameter Used to reflect gene node g i The importance of is calculated as follows:

[0024]

[0025] Among them, σ is the softmax function, deg - (g i ) represents node g i The degree of ,∈ is a positive number;

[0026] Calculate gene node g i Adaptive aggregation weights at layer l

[0027]

[0028] According to a specific implementation of the embodiment of the present disclosure, the expression for updating the feature vectors of the gene node and the drug node is:

[0029]

[0030] Among them, N i and N j Respectively represent g i and d j The set of adjacent nodes of Indicates that symmetric normalization is performed.

[0031] According to a specific implementation of the embodiment of the present disclosure, the expression of the prediction score is:

[0032]

[0033] The expression of the BPR loss function is:

[0034]

[0035] Among them, O represents all known gene-drug association pairs O + and unknown associations -, σ is the Sigmoid function, and λ is the hyperparameter that controls L2 regularization.

[0036] The gene and drug association prediction scheme in the disclosed embodiment includes: step 1, using the construction module of the initial model, constructing the gene and drug association matrix according to the gene and drug association pairs in the database, randomly initializing the feature vector of each gene and drug, and obtaining the gene feature vector matrix matrix and the drug feature vector matrix; step 2, using the dynamic adjustment module of the initial model, based on the association matrix, the gene feature vector matrix matrix, the drug feature vector matrix and the graph convolution network framework, applying the heat kernel to simulate the diffusion process of genes and drugs, integrating the prior knowledge into the feature propagation process between nodes, and dynamically adjusting the adaptive aggregation weights of gene nodes and drug nodes; step 3, using the dynamic adjustment module of the initial model, based on the association matrix, the gene feature vector matrix matrix, the drug feature vector matrix and the graph convolution network framework, applying the heat kernel to simulate the diffusion process of genes and drugs, integrating the prior knowledge into the feature propagation process between nodes, and dynamically adjusting the adaptive aggregation weights of gene nodes and drug nodes. Use the update module of the initial model to update the feature vectors of gene nodes and drug nodes according to the adaptive aggregation weights and the information of the graph neural network and neighbor nodes; step 4, input the feature vectors of the updated nodes into the initial model, use the inner product-based calculation method to calculate the prediction scores of genes and drugs, calculate the BPR loss function according to the prediction scores and data labels, and update the parameters of the initial model through the back propagation of the BPR loss function; step 5, repeat steps 2 to 4 until the model is fitted to obtain the prediction model; step 6, input the feature vectors of the gene nodes and drug nodes in the updated unknown association pairs into the prediction model to perform the inner product operation to obtain the association score matrix of genes and drugs.

[0037] The beneficial effects of the embodiments of the present disclosure are as follows: through the scheme of the present disclosure, a gene-drug association matrix is ​​constructed through the gene-drug association pairs in the database, and the feature vectors of each gene and drug are initialized. In the information aggregation stage, we update the node feature vectors through enhanced node representation, use adaptive aggregation weights, introduce high-order information into the propagation process to further extract high-order relationships. Compared with the prior art, the innovative use of heat kernel simulation graph diffusion process effectively captures the variation of node types and local structures. In addition, this technology can quickly perform large-scale gene-drug association predictions, effectively overcoming the limitations of traditional biological experimental methods, such as high cost, long time and blindness, and improving prediction efficiency, accuracy and adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0039] Figure 1A schematic diagram of a method for predicting the association between genes and drugs provided in an embodiment of the present disclosure;

[0040] Figure 2 ROC curve comparing the performance of the present invention with other methods provided in the embodiments of the present disclosure;

[0041] Figure 3 The PR curve for comparing the performance of the present invention with other methods provided in the embodiments of the present disclosure. DETAILED DESCRIPTION

[0042] The embodiments of the present disclosure are described in detail below with reference to the accompanying drawings.

[0043] The following describes the embodiments of the present disclosure through specific examples, and those skilled in the art can easily understand other advantages and effects of the present disclosure from the contents disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all of the embodiments. The present disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present disclosure. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present disclosure.

[0044] It should be noted that various aspects of the embodiments within the scope of the appended claims are described below. It should be apparent that the aspects described herein may be embodied in a wide variety of forms, and any specific structure and / or function described herein is merely illustrative. Based on the present disclosure, it should be understood by those skilled in the art that an aspect described herein may be implemented independently of any other aspect, and two or more of these aspects may be combined in various ways. For example, any number of aspects described herein may be used to implement the device and / or practice the method. In addition, other structures and / or functionalities other than one or more of the aspects described herein may be used to implement this device and / or practice this method.

[0045] It should also be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present disclosure. The illustrations only show components related to the present disclosure rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the form, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0046] Additionally, in the following description, specific details are provided to facilitate a thorough understanding of the examples. However, it will be understood by those skilled in the art that the aspects described may be practiced without these specific details.

[0047] The present disclosure provides a gene and drug association prediction method, which can be applied to the drug application process in medical scenarios.

[0048] See also Figure 1 , is a flow chart of a gene and drug association prediction method provided by an embodiment of the present disclosure. Figure 1 As shown, the method mainly comprises the following steps:

[0049] Step 1, using the construction module of the initial model, construct the gene-drug association matrix according to the gene-drug association pairs in the database, randomly initialize the feature vectors of each gene and drug, and obtain the feature vector matrix of the gene and the feature vector matrix of the drug;

[0050] In the specific implementation, we can first define the gene set G in the database = {g1, g2, ..., g m} and the drug set D = {d1,d2,...,d n}, where m and n represent the total number of genes and drugs, respectively. We construct an association matrix In gene g i and drugs j When there is a known relationship between i,j =1, otherwise 0. For each gene and each drug, randomly initialize the feature vector of each gene and drug, and integrate these vectors into the feature vector matrix of the gene and the drug feature vector matrix

[0051] Step 2: Using the dynamic adjustment module of the initial model, based on the association matrix, the gene feature vector matrix, the drug feature vector matrix and the graph convolutional network framework, the heat kernel is used to simulate the diffusion process of genes and drugs, and the prior knowledge is integrated into the feature propagation process between nodes to dynamically adjust the adaptive aggregation weights of gene nodes and drug nodes.

[0052] In the specific implementation, a heat kernel is used to simulate the graph diffusion process, combining prior knowledge with feature propagation between nodes. This process is described by the following differential equation:

[0053]

[0054] in, and They represent the node g at time t. i and d jN(i) is the characteristic of g i The equation shows that the node features are updated according to the difference of the neighbor node features. By solving this equation, we get:

[0055]

[0056] Here H t =e -(I-K)t is the heat kernel, I is the identity matrix, and K is the Laplacian matrix of the graph G. The operation of the heat kernel allows convolution of node features:

[0057]

[0058] in, is the normalized adjacency matrix calculated based on the gene-drug association matrix A, in the form of D (g) and D (d) is the angle matrix, the adaptive parameter Used to reflect gene node g i The importance of is calculated as follows:

[0059]

[0060] Here σ is the softmax function, deg - (g i ) represents node g i The degree of ∈ is a small positive number, set to 10 -7 . Combined with the formula, node g i Adaptive aggregation weights at layer l Expressed as:

[0061]

[0062] Step 3: Using the update module of the initial model, the feature vectors of the gene nodes and drug nodes are updated using the graph neural network and the information of neighbor nodes according to the adaptive aggregation weights;

[0063] In the specific implementation, in order to fully mine the high-level node information, we adopt an adaptive method to fuse the enhanced node representation within each layer. Specifically, the node g i and d j The feature vector update mechanism is as follows:

[0064] For gene node g i :

[0065]

[0066] For drug node dj :

[0067]

[0068] Here, N i and N j Respectively represent g i and d j The set of adjacent nodes of Used to perform symmetric normalization. Through this adaptive aggregation mechanism, the model can effectively integrate the features of neighboring nodes at different levels, thereby improving the ability to learn complex relationships.

[0069] Step 4: Input the updated node feature vector into the initial model, calculate the prediction scores of genes and drugs using the inner product-based calculation method, calculate the BPR loss function based on the prediction score and data label, and update the parameters of the initial model through the back propagation of the BPR loss function;

[0070] In practice, in order to predict the potential association between genes and drugs, we calculate the degree of association between them by integrating multi-level feature vector information. The specific operations are as follows:

[0071]

[0072] The aggregated feature vector is used to calculate the gene g i and drugs j The correlation score between:

[0073]

[0074] Based on this calculation, the model is optimized by Bayesian Personalized Ranking (BPR), and its loss function is defined as:

[0075]

[0076] Here O includes all known gene-drug association pairs O + and unknown associations - , σ is the Sigmoid function, and λ is the hyperparameter that controls L2 regularization to balance the fit and complexity of the model and ensure its generalization ability.

[0077] Update the gene embedding matrix Z through back propagation of the BPR loss function (g) and the drug embedding matrix Z (d), which is a low-dimensional representation vector for each gene and drug. Through the back propagation of BPR, the model adjusts these embedding vectors according to the loss gradient, so that the final calculated gene-drug association score is more accurate. In this way, not only is the adaptive propagation of features in the network guaranteed, but also the performance of the model is optimized by focusing on the prediction accuracy of the actual association between genes and drugs.

[0078] Step 5, repeat steps 2 to 4 until the model is fitted and a prediction model is obtained;

[0079] Step 6: Input the updated feature vectors of the gene node and the drug node in the unknown association pair into the prediction model to perform an inner product operation to obtain the gene-drug association score matrix.

[0080] In specific implementation, an inner product operation is performed on the feature vectors of the gene node and the drug node in the final unknown association pair obtained in step 4 to obtain a gene-drug association score matrix. The higher the score, the higher the potential association strength between the gene and the drug.

[0081] The gene and drug association prediction method provided in this embodiment constructs a gene-drug association matrix through the gene-drug association pairs in the database, and initializes the feature vector of each gene and drug. In the information aggregation stage, we update the node feature vector through enhanced node representation, use adaptive aggregation weights, introduce high-order information into the propagation process to further extract high-order relationships. Compared with the prior art, the innovative use of heat kernel simulation graph diffusion process effectively captures the variation of node types and local structures. In addition, this technology can quickly perform large-scale gene-drug association predictions, effectively overcoming the limitations of traditional biological experimental methods, such as high cost, long time and blindness, and improving prediction efficiency, accuracy and adaptability.

[0082] The method of the present disclosure will be described below in conjunction with a specific embodiment. In order to verify the effectiveness of the adaptive graph diffusion mechanism of ADiffGDA, a gene and drug association prediction method of the embodiment of the present disclosure, in information aggregation, we used an alternative node aggregation method for 5-fold cross validation. The experimental results are shown in Table 1. These alternative methods are named ADiffGDA-GCN and ADiffGDA-GAT, respectively. ADiffGDA-GCN means stopping using the adaptive graph diffusion mechanism of ADiffGDA for information aggregation and using the standard LightGCN instead. ADiffGDA-GAT means that we replace the adaptive graph diffusion mechanism of ADiffGDA with the graph attention mechanism during the information aggregation process. The experimental results show that compared with GCN and GAT, the introduction of the adaptive graph diffusion mechanism in the information aggregation process helps to improve the representation ability of GNN on sparse graphs.

[0083] Table 1

[0084]

[0085] In order to further evaluate the performance of the present invention, the present invention is also compared with other state-of-the-art methods, such as Figure 2 and Figure 3 The verification results show that the performance of the proposed method is better than other currently most advanced methods.

[0086] It should be understood that various parts of the present disclosure may be implemented in hardware, software, firmware, or a combination thereof.

[0087] The above is only a specific implementation of the present disclosure, but the protection scope of the present disclosure is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present disclosure should be included in the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure should be based on the protection scope of the claims.

Claims

1. A gene and drug association prediction method, characterized in that: include: Step 1, using the construction module of the initial model, construct the gene-drug association matrix according to the gene-drug association pairs in the database, randomly initialize the feature vectors of each gene and drug, and obtain the feature vector matrix of the gene and the feature vector matrix of the drug; Step 2: Using the dynamic adjustment module of the initial model, based on the association matrix, the gene feature vector matrix, the drug feature vector matrix and the graph convolutional network framework, the heat kernel is used to simulate the diffusion process of genes and drugs, and the prior knowledge is integrated into the feature propagation process between nodes to dynamically adjust the adaptive aggregation weights of gene nodes and drug nodes. Step 3: Using the update module of the initial model, the feature vectors of the gene nodes and drug nodes are updated using the graph neural network and the information of neighbor nodes according to the adaptive aggregation weights; Step 4: Input the updated node feature vector into the initial model, calculate the prediction scores of genes and drugs using the inner product-based calculation method, calculate the BPR loss function based on the prediction score and data label, and update the parameters of the initial model through the back propagation of the BPR loss function; Step 5, repeat steps 2 to 4 until the model is fitted and a prediction model is obtained; Step 6: Input the updated feature vectors of the gene node and the drug node in the unknown association pair into the prediction model to perform an inner product operation to obtain the gene-drug association score matrix.

2. The method according to claim 1, characterized in that , the step 1 specifically includes: Define the gene set G in the database as {g1, g2, ..., g m } and drug set D = {d1,d2,...,d n }, where m and n represent the number of genes and drugs, respectively, to construct the association matrix If the gene g i With drug d j If there is a known relationship between i,j =1, otherwise 0, for each gene node g i and drug node d j Assign d-dimensional feature vector z(g) i and z(d) j , forming the gene feature vector matrix and the drug eigenvector matrix 3. The method according to claim 2, characterized in that , the step 2 specifically includes: The heat kernel simulates the diffusion process of genes and drugs, and integrates prior knowledge into the feature propagation process between nodes, expressed as a differential equation in, and Represents the gene node g at time t i and drug node d j The characteristics of g, N(i) is i The set of neighbor nodes of Solve the differential equation to get the node g after diffusion time t i Features: Among them, H t =e -(I-K)t , H t represents the heat kernel, I is the identity matrix, and K is the Laplace matrix of the graph G; Use the hot kernel to convolve the feature vector of the gene node in, is the normalized adjacency matrix calculated based on the gene-drug association matrix A, in the form of D (g) and D (d) is the angle matrix, the adaptive parameter Used to reflect gene node g i The importance of is calculated as follows: Among them, σ is the softmax function, deg - (g i ) represents node g i The degree of ,∈ is a positive number; Calculate gene node g i Adaptive aggregation weights at layer l 4. The method according to claim 3, characterized in that , the expression of updating the feature vector of gene node and drug node is Among them, N i and N j Respectively represent g i and d j The set of adjacent nodes of Indicates that symmetric normalization is performed.

5. The method according to claim 4, characterized in that , the expression of the prediction score is The expression of the BPR loss function is: Among them, O represents all known gene-drug association pairs O + and unknown associations - , σ is the Sigmoid function, and λ is the hyperparameter that controls L2 regularization.

Images