Model training method, drug and disease association prediction method and related device

Abstract

The invention provides a model training method, a drug and disease association prediction method and a related device. The method comprises the following steps: acquiring a medicine and disease original association diagram, a medicine and medicine original association diagram and a disease and disease original association diagram by the electronic equipment; decomposing and reconstructing the medicine and disease original association graph matrix into a medicine and medicine reconstruction association graph, a medicine and disease reconstruction association graph and a disease and disease reconstruction association graph; and carrying out comparative learning on the association graphs with the corresponding relationship through a graph neural network until the graph neural network meets the condition of serving as a drug and disease association prediction model. Due to the fact that the matrix decomposition is in a low-dimensional decomposition mode, potential features hidden in the data are effectively captured, missing associated information in the data is recovered, the problem of structural information loss caused by disturbance is solved, and therefore the prediction accuracy of the trained model is improved.

Description

Technical Field

[0001] The present application relates to the field of drug development, and more specifically, to a model training method, a drug-disease association prediction method, and related devices. Background Art

[0002] Traditional drug discovery methods are costly, time-consuming and laborious, as evidenced by the fact that many drug candidates that appear promising in early trials often fail in later stages due to toxicity or ineffectiveness. In contrast, with the advancement of drug-disease association prediction technology, approved drugs can be repurposed for new therapeutic applications beyond their original indications. This technology not only reduces risk, but also significantly reduces development costs and time. Therefore, drug-disease association prediction has great potential and offers many advantages, making it an indispensable method in drug discovery.

[0003] However, it has been found in practice that current drug-disease association prediction methods face many challenges. In particular, due to the sparsity of interaction information, random perturbations may lead to the loss of key structural information, thus affecting the quality of representation learning. Summary of the invention

[0004] In order to overcome at least one of the deficiencies in the prior art, the present application provides a model training method, a drug-disease association prediction method and related devices, specifically including:

[0005] In a first aspect, the present application provides a model training method, the method comprising:

[0006] Obtaining original drug-disease association graphs, drug-drug original association graphs, and disease-disease original association graphs;

[0007] Decomposing and reconstructing the original drug-disease association graph matrix into a drug-drug reconstructed association graph, a drug-disease reconstructed association graph, and a disease-disease reconstructed association graph;

[0008] The graph neural network is used to compare and learn the original drug-disease association graph with the reconstructed drug-disease association graph, the original drug-drug association graph with the reconstructed drug-drug association graph, and the original disease-disease association graph with the reconstructed disease-disease association graph, until the graph neural network meets the conditions for serving as a drug-disease association prediction model.

[0009] In combination with the optional implementation manner of the first aspect, the matrix decomposition and reconstruction of the original drug-disease association graph into a drug-drug reconstructed association graph, a drug-disease reconstructed association graph, and a disease-disease reconstructed association graph includes:

[0010] Generate a drug-disease original association matrix according to the drug-disease original association graph;

[0011] Performing probability matrix decomposition on the original association matrix of the drug and the disease to obtain a drug potential feature matrix and a disease potential feature matrix;

[0012] According to the drug potential feature matrix and the disease potential feature matrix, a drug-disease reconstructed association matrix, a drug-drug reconstructed association matrix and a disease-disease reconstructed association matrix are obtained;

[0013] According to the drug-disease reconstruction association matrix, the drug-drug reconstruction association matrix and the disease-disease reconstruction association matrix, the drug-drug reconstruction association graph, the drug-disease reconstruction association graph and the disease-disease reconstruction association graph are obtained.

[0014] In combination with the optional implementation manner of the first aspect, the comparative learning of the original drug-disease association graph and the reconstructed drug-disease association graph, the comparative learning of the original drug-drug association graph and the reconstructed drug-drug association graph, and the comparative learning of the original disease-disease association graph and the reconstructed disease-disease association graph through the graph neural network includes:

[0015] Extracting features from the original drug-disease association graph and the reconstructed drug-disease association graph through the graph neural network to obtain an embedded representation of the original drug-disease association graph and an embedded representation of the reconstructed drug-disease association graph;

[0016] Obtaining a contrast loss between drugs and diseases and a binary cross entropy loss according to the embedding representation of the original association graph between drugs and diseases and the embedding representation of the reconstructed association graph between drugs and diseases;

[0017] Performing feature extraction on the original drug-drug association graph and the reconstructed drug-drug association graph through the graph neural network to obtain an embedded representation of the original drug-drug association graph and an embedded representation of the reconstructed drug-drug association graph;

[0018] According to the embedding representation of the original association graph between the drugs and the embedding representation of the reconstructed association graph between the drugs and the drugs, a contrast loss between the drugs and the drugs and a binary cross entropy loss are obtained;

[0019] Performing feature extraction on the original disease-disease association graph and the reconstructed disease-disease association graph through the graph neural network to obtain an embedded representation of the original disease-disease association graph and an embedded representation of the reconstructed disease-disease association graph;

[0020] Obtaining a contrast loss between diseases and a binary cross entropy loss according to the embedding representation of the disease-disease original association graph and the embedding representation of the disease-disease reconstructed association graph;

[0021] The drug-disease contrast loss and binary cross entropy loss, the drug-drug contrast loss and binary cross entropy loss, and the disease-disease contrast loss and binary cross entropy loss are weighted to obtain a comprehensive loss;

[0022] The network parameters of the graph neural network are updated according to the comprehensive loss.

[0023] In combination with the optional implementation manner of the first aspect, obtaining the contrast loss between drugs and diseases and the binary cross entropy loss according to the embedding representation of the original association graph of drugs and diseases and the embedding representation of the reconstructed association graph of drugs and diseases includes:

[0024] Compare the embedding representation of the original association graph of drugs and diseases with the embedding representation of the reconstructed association graph of drugs and diseases from the row and column respectively to obtain the row comparison loss of drugs and diseases and the column comparison loss of drugs and diseases;

[0025] Weighting the row contrast loss between the drug and the disease and the column contrast loss between the drug and the disease to obtain the contrast loss between the drug and the disease;

[0026] Obtaining association scores between multiple drugs and multiple diseases according to the embedded representation of the original association graph between the drugs and the diseases and the embedded representation of the reconstructed association graph between the drugs and the diseases;

[0027] According to the association scores between the multiple drugs and the multiple diseases, the binary cross entropy loss between the drugs and the diseases is obtained.

[0028] In combination with the optional implementation manner of the first aspect, obtaining the binary cross entropy loss between the drugs and the diseases according to the association scores between the multiple drugs and the multiple diseases includes:

[0029] Determining association scores of known drug-disease pairs and predicted association scores of drug-disease pairs from the association scores between the multiple drugs and the multiple diseases;

[0030] The binary cross entropy loss between the drug and the disease is obtained according to the association score of the known drug-disease pair and the association score of the predicted drug-disease pair.

[0031] In combination with the optional implementation manner of the first aspect, obtaining association scores between multiple drugs and multiple diseases according to the embedded representation of the original association graph of drugs and diseases and the embedded representation of the reconstructed association graph of drugs and diseases includes:

[0032] fusing the embedding representation of the original association graph of drugs and diseases with the embedding representation of the same drug in the embedding representation of the reconstructed association graph of drugs and diseases to obtain a comprehensive embedding representation of multiple drugs;

[0033] fusing the embedding representation of the original association graph of drugs and diseases with the embedding representation of the same disease in the embedding representation of the reconstructed association graph of drugs and diseases to obtain a comprehensive embedding representation of multiple diseases;

[0034] According to the comprehensive embedding representation of multiple drugs and the comprehensive embedding representation of multiple diseases, the association scores between multiple drugs and multiple diseases are obtained.

[0035] In a second aspect, the present application also provides a method for predicting the association between a drug and a disease, the method comprising:

[0036] Obtain the original association diagram between the drug and the disease to be analyzed;

[0037] Decomposing and reconstructing the original association graph matrix of the drug to be analyzed and the disease into a reconstructed association graph of the drug to be analyzed and the disease;

[0038] The drug-disease association prediction model trained by the model training method is used to process the drug-drug reconstructed association graph to be analyzed and the drug-disease original association graph to be analyzed, so as to obtain drug-disease association prediction information.

[0039] In a third aspect, the present application also provides a model training device, the device comprising:

[0040] A graph decomposition module, used to obtain the original association graph between drugs and diseases, the original association graph between drugs and drugs, and the original association graph between diseases;

[0041] The graph decomposition module is further used to decompose and reconstruct the original drug-disease association graph matrix into a drug-drug reconstructed association graph, a drug-disease reconstructed association graph, and a disease-disease reconstructed association graph;

[0042] The model training module is used to compare and learn the original drug-disease association graph with the reconstructed drug-disease association graph, the original drug-drug association graph with the reconstructed drug-drug association graph, and the original disease-disease association graph with the reconstructed disease-disease association graph through a graph neural network, until the graph neural network meets the conditions of serving as a drug-disease association prediction model.

[0043] In conjunction with the optional implementation manner of the third aspect, the graph decomposition module is further specifically used for:

[0044] Generate a drug-disease original association matrix based on the drug-disease original association graph;

[0045] Performing probability matrix decomposition on the original association matrix of the drug and the disease to obtain a drug potential feature matrix and a disease potential feature matrix;

[0046] According to the drug potential feature matrix and the disease potential feature matrix, a drug-disease reconstructed association matrix, a drug-drug reconstructed association matrix and a disease-disease reconstructed association matrix are obtained;

[0047] According to the drug-disease reconstruction association matrix, the drug-drug reconstruction association matrix and the disease-disease reconstruction association matrix, the drug-drug reconstruction association graph, the drug-disease reconstruction association graph and the disease-disease reconstruction association graph are obtained.

[0048] In conjunction with the optional implementation manner of the third aspect, the model training module is further specifically used for:

[0049] Extracting features from the original drug-disease association graph and the reconstructed drug-disease association graph through the graph neural network to obtain an embedded representation of the original drug-disease association graph and an embedded representation of the reconstructed drug-disease association graph;

[0050] Obtaining a contrast loss between drugs and diseases and a binary cross entropy loss according to the embedding representation of the original association graph between drugs and diseases and the embedding representation of the reconstructed association graph between drugs and diseases;

[0051] Performing feature extraction on the original drug-drug association graph and the reconstructed drug-drug association graph through the graph neural network to obtain an embedded representation of the original drug-drug association graph and an embedded representation of the reconstructed drug-drug association graph;

[0052] According to the embedding representation of the original association graph between the drugs and the embedding representation of the reconstructed association graph between the drugs and the drugs, a contrast loss between the drugs and the drugs and a binary cross entropy loss are obtained;

[0053] Performing feature extraction on the original disease-disease association graph and the reconstructed disease-disease association graph through the graph neural network to obtain an embedded representation of the original disease-disease association graph and an embedded representation of the reconstructed disease-disease association graph;

[0054] Obtaining a contrast loss between diseases and a binary cross entropy loss according to the embedding representation of the disease-disease original association graph and the embedding representation of the disease-disease reconstructed association graph;

[0055] The drug-disease contrast loss and binary cross entropy loss, the drug-drug contrast loss and binary cross entropy loss, and the disease-disease contrast loss and binary cross entropy loss are weighted to obtain a comprehensive loss;

[0056] The network parameters of the graph neural network are updated according to the comprehensive loss.

[0057] In conjunction with the optional implementation manner of the third aspect, the model training module is further specifically used for:

[0058] Compare the embedding representation of the original association graph of drugs and diseases with the embedding representation of the reconstructed association graph of drugs and diseases from the row and column respectively to obtain the row comparison loss of drugs and diseases and the column comparison loss of drugs and diseases;

[0059] Weighting the row contrast loss between the drug and the disease and the column contrast loss between the drug and the disease to obtain the contrast loss between the drug and the disease;

[0060] Obtaining association scores between multiple drugs and multiple diseases according to the embedded representation of the original association graph between the drugs and the diseases and the embedded representation of the reconstructed association graph between the drugs and the diseases;

[0061] According to the association scores between the multiple drugs and the multiple diseases, the binary cross entropy loss between the drugs and the diseases is obtained.

[0062] In conjunction with the optional implementation manner of the third aspect, the model training module is further specifically used for:

[0063] Determining association scores of known drug-disease pairs and predicted association scores of drug-disease pairs from the association scores between the multiple drugs and the multiple diseases;

[0064] The binary cross entropy loss between the drug and the disease is obtained according to the association score of the known drug-disease pair and the association score of the predicted drug-disease pair.

[0065] In conjunction with the optional implementation manner of the third aspect, the model training module is further specifically used for:

[0066] fusing the embedding representation of the original association graph of drugs and diseases with the embedding representation of the same drug in the embedding representation of the reconstructed association graph of drugs and diseases to obtain a comprehensive embedding representation of multiple drugs;

[0067] fusing the embedding representation of the original association graph of drugs and diseases with the embedding representation of the same disease in the embedding representation of the reconstructed association graph of drugs and diseases to obtain a comprehensive embedding representation of multiple diseases;

[0068] According to the comprehensive embedding representation of multiple drugs and the comprehensive embedding representation of multiple diseases, the association scores between multiple drugs and multiple diseases are obtained.

[0069] In a fourth aspect, the present application also provides a storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the model training method or the drug-disease association prediction method.

[0070] In a fifth aspect, the present application also provides an electronic device, comprising a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the model training method or the drug-disease association prediction method is implemented.

[0071] Compared with the prior art, this application has the following beneficial effects:

[0072] The present application provides a model training method, a drug-disease association prediction method and related devices. Among them, the electronic device obtains the original drug-disease association graph, the original drug-drug association graph, and the original disease-disease association graph; the original drug-disease association graph matrix is ​​decomposed and reconstructed into a drug-drug reconstructed association graph, a drug-disease reconstructed association graph, and a disease-disease reconstructed association graph; the original drug-disease association graph and the drug-disease reconstructed association graph are compared and learned by a graph neural network, the original drug-disease association graph and the drug-disease reconstructed association graph are compared and learned, and the original disease-disease association graph and the disease-disease reconstructed association graph are compared and learned, until the graph neural network meets the conditions of being a drug-disease association prediction model. Since matrix decomposition effectively captures the potential features hidden in the data in a low-dimensional decomposition manner, and restores the missing association information in the data, it makes up for the problem of structural information loss caused by disturbance, thereby improving the prediction accuracy of the trained model. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0074] Figure 1 A schematic diagram of the scheme principle provided for the embodiment of the present application;

[0075] Figure 2 A flowchart of a model training method provided in an embodiment of the present application;

[0076] Figure 3 A comparison chart of the prediction effects of the models provided in the embodiments of the present application;

[0077] Figure 4A schematic diagram of the structure of a model training device provided in an embodiment of the present application;

[0078] Figure 5 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0079] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various different configurations.

[0080] Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the present application for which protection is sought, but merely represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0081] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0082] In the description of the present application, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inside", "outside", etc. indicate positions or positional relationships based on the positions or positional relationships shown in the accompanying drawings, or the positions or positional relationships in which the invented product is usually placed when in use. They are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific position, be constructed and operated in a specific position, and therefore cannot be understood as a limitation on the present application. In addition, the terms "first", "second", "third", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0083] In addition, the terms "comprises," "comprising," or any other variation thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0084] In addition, the terms "horizontal", "vertical", "overhanging" and the like do not mean that the components are required to be absolutely horizontal or overhanging, but can be slightly tilted. For example, "horizontal" only means that its direction is more horizontal than "vertical", and does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0085] In the description of this application, it should also be noted that, unless otherwise clearly specified and limited, the terms "set", "install", "connect", and "connect" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two elements. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0086] Based on the above statement, as introduced in the background technology, although drug-disease association prediction has great potential and provides many advantages, making it an indispensable method in drug discovery. However, due to the sparsity of interactive information, the current drug-disease association prediction methods may cause the loss of key structural information due to random perturbations, thus affecting the quality of representation learning.

[0087] In this regard, the research process found that graph neural networks (GNNs) can effectively process node and edge information in complex network structures by aggregating node neighbor information and learning high-quality node representations. Therefore, they can be used to capture implicit associations and interactions between drugs and diseases, thereby improving the accuracy and robustness of predictions. However, when graph neural networks learn high-quality drug and disease representations from limited interaction data, they are usually constrained by data sparsity.

[0088] During the research, it was further discovered that contrastive learning has been proven to be an effective way to deal with data sparsity by introducing the comparison of positive and negative samples. In the drug-disease association prediction task, contrastive learning can automatically mine potential association information and enhance the model's generalization ability for sparse data. Specifically, contrastive learning uses known drug-disease associations as positive samples and unknown or unlikely drug-disease pairs as negative samples. In this way, the model can better learn the differences and similarities between drugs and diseases, thereby improving prediction accuracy.

[0089] Based on the discovery of the above technical problems, the inventors, through creative work, proposed to apply contrastive learning to graph neural networks for drug-disease association prediction to improve the predictive performance of the model. However, in practice, due to the problem of sparsity of interactive information, even random perturbations may cause the loss of key structural information, thereby affecting the quality of representation learning. The so-called sparsity of interactive information means that there is very little association data between known drugs and diseases, and most of the data is unknown. It's like in a huge puzzle, only a few pieces of the puzzle are known, and most are blank. In this case, it is difficult to accurately predict which drugs may be effective for which diseases because there is not enough information to fill these gaps. This leads to difficulties in learning and predicting the model because it cannot accurately capture the potential connection between drugs and diseases.

[0090] It should be noted that the defects existing in the solutions in the above-mentioned prior art are the results obtained by the inventor after practice and careful research. Therefore, the discovery process of the above-mentioned problems and the solutions proposed in the embodiments of the present application for the above-mentioned problems below should be the contributions made by the inventor to the present application in the process of invention and creation, and should not be understood as technical contents known to technical personnel in this field.

[0091] In view of this, the present embodiment provides a model training method. In this method, the electronic device obtains the original drug-disease association graph, the original drug-drug association graph, and the original disease-disease association graph; the original drug-disease association graph matrix is ​​decomposed and reconstructed into a drug-drug reconstructed association graph, a drug-disease reconstructed association graph, and a disease-disease reconstructed association graph; the original drug-disease association graph and the drug-disease reconstructed association graph are compared and learned by the graph neural network, the original drug-drug association graph and the drug-drug reconstructed association graph are compared and learned, and the original disease-disease association graph and the disease-disease reconstructed association graph are compared and learned, until the graph neural network meets the conditions of being a drug-disease association prediction model. Since matrix decomposition effectively captures the potential features hidden in the data in a low-dimensional decomposition manner, and restores the missing association information in the data, it makes up for the problem of structural information loss caused by random disturbances, thereby improving the prediction accuracy of the trained model.

[0092] It is worth noting that the electronic device implementing the method may be, but is not limited to, a mobile terminal, a tablet computer, a laptop computer, a desktop computer, and a server, etc., as long as the computing power that can be provided is used for model training. When it is a server, the server may be a single server or a server group. The server group may be centralized or distributed (for example, the server may be a distributed system). In some embodiments, the server may be local or remote relative to the user terminal. In some embodiments, the server may be implemented on a cloud platform; as an example only, the cloud platform may include a private cloud, a public cloud, a hybrid cloud, a community cloud (Community Cloud), a distributed cloud, an inter-cloud (Inter-Cloud), a multi-cloud (Multi-Cloud), etc., or any combination thereof. In some embodiments, the server may be implemented on an electronic device having one or more components.

[0093] The model training method provided in this embodiment can be summarized as follows: Figure 1 As shown, the original drug-disease association graph is decomposed and reconstructed into a drug-disease reconstructed association graph, a drug-drug reconstructed association graph, and a disease-disease reconstructed association graph using a probabilistic matrix decomposition method with Gaussian noise; then, the drug-disease reconstructed association graph, the drug-drug reconstructed association graph, and the disease-disease reconstructed association graph are subjected to multi-view comparative learning with their respective initial association graphs; finally, a unified loss function is used to train and optimize these comparative views.

[0094] To make the solution provided by this embodiment clearer, the following uses a server as an electronic device to implement the method, and combines Figure 2 Each step of the method is described in detail. However, it should be understood that the operations of the flowchart may not be implemented in order, and steps without logical contextual relationships may be reversed in order or implemented simultaneously. In addition, those skilled in the art may add one or more other operations to the flowchart, or remove one or more operations from the flowchart under the guidance of the content of this application. Figure 2 As shown, the method includes:

[0095] S1, obtaining the original association graph between drugs and diseases, the original association graph between drugs and drugs, and the original association graph between diseases and diseases.

[0096] Among them, the original drug-disease association graph refers to a network representation of the known direct association between drugs and diseases obtained through actual observation or recording, in which nodes represent drugs or diseases, and edges represent the known associations between them, which are used as basic data for model training and verification. This kind of map can intuitively show which drugs have been proven to be effective for which diseases, and provides initial structured information for subsequent comparative learning and probability matrix decomposition and other technologies. The original drug-drug association graph and the original disease-disease association graph are similar, and this embodiment will not be repeated.

[0097] Continue to see Figure 2 , the model training method provided in this embodiment also includes:

[0098] S2, decompose and reconstruct the original drug-disease association graph matrix into a drug-drug reconstructed association graph, a drug-disease reconstructed association graph, and a disease-disease reconstructed association graph.

[0099] It should be noted that, unlike conventional matrix decomposition, this embodiment introduces Gaussian noise when decomposing the matrix in order to enhance the robustness of the model, prevent overfitting, and improve the generalization ability of the model when processing sparse or incomplete data. Therefore, step S2 can be implemented by the following optional implementation:

[0100] S2-1, generating a drug-disease original association matrix based on the drug-disease original association graph.

[0101] For example, assume that there are n drugs R = {r1, r2, r 3, …,r n} and m diseases D = {d1,d2,d 3, …,d m}, using a binary matrix Represents the original drug-disease association matrix, which is a binary matrix. Each row of the binary matrix represents a drug and each column represents a disease. The elements in the binary matrix are represented by Y ij ∈{0,1}, used to characterize the i and diseases j Is there a relationship between the two? When experimental evidence confirms that the drug i and diseases j When there is a correlation between ij =1; otherwise, Y ij =0.

[0102] Based on the introduction of the original association matrix of drugs and diseases in the above embodiment, step S2 further includes:

[0103] S2-2, perform probability matrix decomposition on the original association matrix between drugs and diseases to obtain drug potential feature matrix and disease potential feature matrix.

[0104] It should be understood that the matrix decomposition method needs to project the original drug-disease association matrix into the drug-disease potential feature matrix, and obtain the drug-disease reconstructed association matrix through their product. In this embodiment, the drug potential feature matrix is ​​represented as And the disease potential feature matrix is Then the drug-disease reconstructed association matrix can be expressed as Therefore, the goal of matrix decomposition is to find the accurate W and H to obtain the accurate reconstructed drug-disease association matrix.

[0105] As described in the above embodiment, Gaussian noise is introduced when decomposing the matrix, and the conditional distribution of the drug-disease interaction matrix Y can be represented by the following probability model:

[0106]

[0107] In the formula, Corresponds to represents the probability density function of the Gaussian normal distribution, I ij is the indicator function, when the drug r i and diseases j When relevant, I ij =1; otherwise, I ij =0; n,m represent the number of drugs and diseases respectively.

[0108] In order to further optimize the model, a prior distribution is added, that is, a zero-mean spherical Gaussian prior is introduced into the feature vectors of drugs and diseases:

[0109]

[0110]

[0111] Where I represents a k-dimensional unit diagonal matrix.

[0112] Based on the introduction of the above probability model, this implementation defines the corresponding objective function for matrix decomposition. That is, by taking the logarithm of the logarithmic posterior distribution and transforming it, the maximization problem is converted into a minimization problem while keeping the hyperparameters unchanged. The minimization process involves an objective function that includes squared error and quadratic regularization terms:

[0113]

[0114] In the formula, ||·|| FRepresents the Frobenius norm. Based on the above objective function, considering the similarity constraints between drugs and diseases, the drug latent feature matrix W and the disease latent feature matrix H are finally learned by minimizing the following loss function:

[0115]

[0116] In the formula, S r and S d They represent the drug original association matrix (also known as the drug similarity matrix) and the disease original association matrix (also known as the disease similarity matrix), δ W ,δ H ,δ1,δ2 respectively represent the variance of the data corresponding to the Frobenius norm. Using the above loss function, the drug potential feature matrix W and the disease potential feature matrix H are obtained by gradient descent iteration for a predetermined number of iterations. The expression for update iteration is:

[0117]

[0118]

[0119] Wherein, B is a binary matrix indicating the position of non-zero elements in the original drug-disease association matrix Y, and μ1~μ4 represent preset parameters respectively.

[0120] Based on the drug potential feature matrix and the disease potential feature matrix obtained by probability matrix decomposition in the above embodiment, step S2 further includes:

[0121] S2-3, based on the drug potential feature matrix and the disease potential feature matrix, obtain the drug-disease reconstructed association matrix, the drug-drug reconstructed association matrix and the disease-disease reconstructed association matrix.

[0122] In this regard, continuing with the above-mentioned drug potential feature matrix W and disease potential feature matrix H as an example, multiply the drug potential feature matrix W by its own transpose to obtain the drug-drug reconstruction association matrix Multiply the disease potential feature matrix H by its own transpose to obtain the disease-disease reconstruction association matrix The expression is as follows:

[0123]

[0124] S2-4, according to the drug-disease reconstructed association matrix, the drug-drug reconstructed association matrix and the disease-disease reconstructed association matrix, obtain the drug-drug reconstructed association graph, the drug-disease reconstructed association graph and the disease-disease reconstructed association graph.

[0125] Thus, through the above implementation, the drug-disease reconstructed association matrix can be obtained: Drug-drug reconstructed association matrix Disease-disease reconstructed association matrix Then, using these reconstructed association matrices, we constructed the drug-disease reconstructed association graphs. Drug and drug reconstruction association diagram and disease-disease reconstruction association graph This provides a data basis for further analysis and prediction. It should be understood that the above probability matrix decomposition method can effectively capture the complex relationship between drugs and diseases and reveal their internal similarity structure, thus providing a new perspective and scientific basis for drug-disease association prediction and disease treatment.

[0126] Based on the reconstructed association matrices decomposed in the above embodiment, see Figure 2 , the model training method provided in this embodiment also includes:

[0127] S3, through the graph neural network, comparative learning is performed on the original drug-disease association graph and the reconstructed drug-disease association graph, the original drug-drug association graph and the reconstructed drug-drug association graph, and the original disease-disease association graph and the reconstructed disease-disease association graph, until the graph neural network meets the conditions for serving as a drug-disease association prediction model.

[0128] In this regard, this embodiment provides the following optional implementations of step S3:

[0129] S3-1, feature extraction is performed on the original drug-disease association graph and the reconstructed drug-disease association graph through a graph neural network to obtain an embedded representation of the original drug-disease association graph and an embedded representation of the reconstructed drug-disease association graph.

[0130] Among them, the Graph Neural Network (GNN) is a neural network model that learns graph structure data. The model captures the structural information and attribute information of the graph by learning the relationship between nodes and the characteristics of the nodes themselves. In the graph neural network, each node updates its own features based on the characteristics of its neighboring nodes and the information of the edges. Therefore, the input of the graph neural network is a graph with node features and edge relationships, and the output is a graph after feature transformation, in which the features of the nodes have changed, but the structure of the graph (that is, the connection relationship between the nodes) usually remains unchanged.

[0131] For example, for the original association graph between drugs and diseases, the graph neural network of this embodiment can use two layers of graph convolution (GCN) to aggregate the neighbor node information of each node, update the representation of each node, and then use it as the input of the next layer. Taking the first layer of graph convolution as an example, the aggregation process of the first layer is as follows:

[0132]

[0133] In the formula, and Represents drug r i and the first layer embedding of disease dj, Ω(·) represents the activation function, p(·) represents the removal of edges in the initial association matrix Y between drugs and diseases to alleviate the overfitting problem, Represents the embedded representation of the disease in the input drug-disease original association graph, Represents the embedding representation of the drug in the input original drug-disease association graph.

[0134] Then, after aggregation and summation at all layers, the final embedding of drugs and diseases on the initial drug-disease association graph G is expressed as:

[0135]

[0136] In the formula, and Represents drug r i and diseases j The final embedding representation of .

[0137] Similarly, for the drug-disease reconstructed association graph, the graph neural network is also used to propagate messages on the drug-disease reconstructed association graph, which is specifically expressed as follows:

[0138]

[0139] In the formula, and Represents drug r i and diseases j Reconstructing the drug-disease association graph Finally, the drug-disease reconstructed association graph The final embedding representation of drugs and diseases on is:

[0140]

[0141] in, and Represents drug r i and diseases j In the reconstruction The aggregate embedding representation on .

[0142] Based on the embedding representation of the original drug-disease association graph and the embedding representation of the reconstructed drug-disease association graph, step S3 further includes:

[0143] S3-2, based on the embedding representation of the original drug-disease association graph and the embedding representation of the reconstructed drug-disease association graph, the drug-disease contrast loss and the binary cross entropy loss are obtained.

[0144] Among them, the above-mentioned drug-disease contrast loss is obtained through contrast learning. Conventional graph contrast learning usually aggregates the features of each node (for example, summation, average, maximum value, etc.) and then compares them. Since the embedded representation of the original drug-disease association graph and the embedded representation of the reconstructed drug-disease association graph are respectively a vector matrix, this embodiment is different from this in that the embedded representation of the original drug-disease association graph and the embedded representation of the reconstructed drug-disease association graph are compared from rows and columns respectively to obtain the contrast loss. On this basis, in order to enhance the generalization ability of the model, binary cross entropy loss is also introduced. Therefore, step S3-2 includes the following optional implementations:

[0145] S3-2-1, compare the embedding representation of the original drug-disease association graph with the embedding representation of the reconstructed drug-disease association graph from the rows and columns respectively, and obtain the row comparison loss of drugs and diseases and the column comparison loss of drugs and diseases.

[0146] S3-2-2, weighting the row comparison loss between drugs and diseases and the column comparison loss between drugs and diseases to obtain the comparison loss between drugs and diseases.

[0147] S3-2-3, based on the embedding representation of the original association graph between drugs and diseases and the embedding representation of the reconstructed association graph between drugs and diseases, the association scores between multiple drugs and multiple diseases are obtained.

[0148] In this embodiment, for the association scores between the above-mentioned multiple drugs and multiple diseases, the server can fuse the embedded representation of the original drug-disease association graph with the embedded representation of the same drug in the embedded representation of the reconstructed drug-disease association graph to obtain a comprehensive embedded representation of the multiple drugs; fuse the embedded representation of the original drug-disease association graph with the embedded representation of the same disease in the embedded representation of the reconstructed drug-disease association graph to obtain a comprehensive embedded representation of the multiple diseases; and obtain the association scores between the multiple drugs and the multiple diseases based on the comprehensive embedded representation of the multiple drugs and the comprehensive embedded representation of the multiple diseases.

[0149] For example, in order to obtain the best prediction effect, the two embedding representations of the same drug are fused, and the two embedding representations of the same disease are fused. The expression is as follows:

[0150]

[0151]

[0152] In the formula, Q r Includes various drugs i The comprehensive embedding representation of Q d Includes various diseases j The comprehensive embedding representation of drug r is shown in Figure 2. λ1 and λ2 are hyperparameters that adjust the importance of the two view embedding representations. Based on the above comprehensive embedding representation, drug r i With disease j The correlation score between them can be calculated by the following expression:

[0153]

[0154] The association scores between the various drugs and the various diseases obtained in the above embodiment can be further calculated to obtain the binary cross entropy loss between the drugs and the diseases:

[0155] S3-2-4, based on the association scores between multiple drugs and multiple diseases, the binary cross entropy loss between drugs and diseases is obtained.

[0156] Based on the above implementation, the above contrast loss and binary cross entropy loss are described in detail below in combination with specific expressions. Since the contrast loss and binary cross entropy loss between drugs and diseases, the contrast loss and binary cross entropy loss between drugs, and the contrast loss and binary cross entropy loss between diseases are all calculated in the same way. That is, when calculating the contrast loss, there are three contrast views. For the sake of description, in this embodiment, the original association graph G between drugs and drugs is used. r , Disease and disease original association graph G d And the row vector and column vector in the final embedding representation of the original drug-disease association graph G are denoted as Z (r) and Z (C) .

[0157] Similarly, reconstruct the association graph between drugs Disease-disease reconstructed association graph And the drug-disease reconstructed association map The row vector and column vector in the final embedding representation are denoted as G (r) and G (C) .

[0158] Then, the total loss in the three contrasting views is calculated through a consistent optimization scheme. The goal of contrastive learning is to align node embeddings in different views, that is, different perspectives of the same node are used as positive examples, and different perspectives of different nodes are used as negative examples. To this end, the InfoNCE contrastive loss can be used to enhance the alignment of positive examples while reducing the impact of negative examples, thereby obtaining the row contrastive loss represented by the row embeddings in the three contrasting views:

[0159]

[0160] In the formula, s(·) represents the cosine similarity function, and τ represents the temperature coefficient. Similarly, the column contrast loss represented by column embedding is obtained:

[0161]

[0162] Finally, the above row contrast loss and column contrast loss are weighted to obtain the contrast loss of each contrast view:

[0163] l cl =β1*l row +β2*l col

[0164] In this way, the contrast loss of each contrast view can be calculated by the above expression.

[0165] For the above binary cross entropy loss, the server can determine the association scores of known drug-disease pairs and the association scores of predicted drug-disease pairs from the association scores between multiple drugs and multiple diseases; according to the association scores of known drug-disease pairs and the association scores of predicted drug-disease pairs, the binary cross entropy loss between drugs and diseases is obtained. In this regard, the calculation expression of binary cross entropy loss is as follows:

[0166]

[0167] In the formula, l bce represents the binary cross entropy loss between drugs and diseases, represents the association score between the ith drug and the jth drug, represents the set of known drug-disease pairs, represents the set of predicted drug-disease pairs, α represents the preset balance factor, and m and n represent the number of drugs and diseases, respectively.

[0168] It should be noted that, since the calculation method of the binary cross entropy loss between drugs and diseases is consistent, when calculating the drug-drug relationship, the above (i, j) represents a drug pair. represents the set of all known drug association pairs, It represents the set of predicted drug association pairs. When calculating diseases and diseases, the above (i, j) represents a disease pair. represents the set of all known disease association pairs, represents the set of predicted disease association pairs.

[0169] Therefore, the above embodiment takes the comparison view of drugs and diseases as an example to illustrate the comparison loss between drugs and diseases and the binary cross entropy loss. The comparison loss between drugs and the binary cross entropy loss and the comparison loss between diseases and the binary cross entropy loss can be obtained in the same calculation method; therefore, step S3 also includes:

[0170] S3-3, feature extraction is performed on the original drug-drug association graph and the reconstructed drug-drug association graph through a graph neural network to obtain an embedded representation of the original drug-drug association graph and an embedded representation of the reconstructed drug-drug association graph.

[0171] S3-4, based on the embedding representation of the original drug-drug association graph and the embedding representation of the reconstructed drug-drug association graph, the drug-drug contrast loss and the binary cross entropy loss are obtained.

[0172] 3-5. Feature extraction is performed on the original disease-disease association graph and the reconstructed disease-disease association graph through graph neural network to obtain the embedded representation of the original disease-disease association graph and the embedded representation of the reconstructed disease-disease association graph.

[0173] S3-6, based on the embedding representation of the original disease-disease association graph and the embedding representation of the reconstructed disease-disease association graph, the disease-disease contrast loss and the binary cross entropy loss are obtained.

[0174] S3-7, the drug-disease contrast loss and binary cross entropy loss, drug-drug contrast loss and binary cross entropy loss, disease-disease contrast loss and binary cross entropy loss are weighted to obtain a comprehensive loss.

[0175] S3-8, update the network parameters of the graph neural network according to the comprehensive loss.

[0176] In summary, this embodiment aims at the problem that random disturbance may cause loss of key structural information in contrastive learning, thereby affecting the quality of representation learning. Probabilistic matrix decomposition is used to reconstruct the structure graph, and then multi-view contrastive learning is implemented. Among them, probabilistic matrix decomposition effectively captures the potential features hidden in the data through low-dimensional decomposition, restores the missing associated information in the data, and thus makes up for the problem of structural information loss caused by random disturbance. In this process, Gaussian noise is also introduced, which has the advantage of enhancing the robustness of the model, preventing the occurrence of overfitting, and helping the model to focus more accurately on the core features of the data during the contrastive learning process. In addition, Gaussian noise can simulate the uncertainty in the data, enhance the generalization ability of the model when processing sparse or incomplete data, and further improve the prediction effect.

[0177] In addition, in order to solve the problem of data sparsity, this embodiment constructs a multi-view comparative learning framework. This framework makes full use of multiple network information such as drug-disease, drug-drug, disease-disease, etc., mines data from different perspectives, and can identify and compare changes and stability in the network. The multi-view comparative learning strategy reveals potential key information markers and therapeutic targets, enabling the model to make more accurate association predictions when faced with complex drug-disease relationships.

[0178] Finally, in the graph convolutional network (GCN) layer, the representation of the node is continuously updated by aggregating the information of neighboring nodes, thereby capturing the local connection pattern and its dynamic changes in the network. This mechanism is crucial for understanding the complex interaction mechanism between drugs and diseases, and can reveal the potential relationship between drugs and diseases more deeply, further enhancing the accuracy and reliability of predictions.

[0179] This embodiment also provides a method for predicting the association between drugs and diseases. In this method, the electronic device obtains the original association graph between drugs and diseases to be analyzed; decomposes the matrix of the original association graph between drugs and diseases to be analyzed into a reconstructed association graph between drugs and drugs to be analyzed; and processes the reconstructed association graph between drugs and drugs to be analyzed and the original association graph between drugs and diseases to be analyzed with the drug-disease association prediction model trained by the above-mentioned model training method to obtain drug-disease association prediction information. In this way, the potential relationship between drugs and diseases is revealed.

[0180] In this regard, this embodiment uses the above-trained drug-disease association prediction model as CDPMF-DDA, and compares it with five existing models (LBMFF, SCPMFDD, SCPMF, MKGCN, MNGACDA) on three datasets: Fdataset, Cdataset, and LRSSL. The performance of each model is as follows: Figure 3As shown in the figure, the area under the ROC curve (AUROC) is used to measure the overall performance of the model. Its value range is between 0 and 1. The closer to 1, the better the model performance. The boldface represents the optimal value; the area under the precision-recall curve (AUPR): represents the average accuracy of the model at different thresholds. AUPR is also between 0 and 1. The larger the value, the better the model performs on positive samples. Therefore, it is not difficult to see that CDPMF-DDA has a good performance on each data set.

[0181] Based on the same inventive concept as the model training method provided in this embodiment, this embodiment also provides a model training device, which includes at least one software function module that can be stored in a memory or fixed in an electronic device in the form of software. The processor in the electronic device is used to execute the executable module stored in the memory. For example, the software function module and computer program included in the device. Please refer to Figure 4 , functionally speaking, the device may include:

[0182] A graph decomposition module 11, used to obtain an original drug-disease association graph, an original drug-drug association graph, and an original disease-disease association graph;

[0183] The graph decomposition module 11 is also used to decompose and reconstruct the original drug-disease association graph matrix into a drug-drug reconstructed association graph, a drug-disease reconstructed association graph, and a disease-disease reconstructed association graph;

[0184] The model training module 12 is used to compare and learn the original drug-disease association graph and the reconstructed drug-disease association graph, the original drug-drug association graph and the reconstructed drug-drug association graph, and the original disease-disease association graph and the reconstructed disease-disease association graph through a graph neural network, until the graph neural network meets the conditions of serving as a drug-disease association prediction model.

[0185] In this embodiment, the graph decomposition module 11 is used to implement Figure 2 In steps S1 and S2, the model training module 12 is used to implement Figure 2 Therefore, for the detailed description of each module, please refer to the specific factual method of the corresponding step. In addition, it is worth noting that since it has the same invention wording as the above-mentioned model training method, the model training device can also implement other steps or sub-steps of the method through the above-mentioned modules, which will not be repeated in this embodiment.

[0186] In addition, the functional modules in the various embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.

[0187] It should also be understood that if the above implementation is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application.

[0188] Therefore, this embodiment also provides a storage medium, which is a computer-readable storage medium. The storage medium stores a computer program, and when the computer program is executed by the processor, the model training method or the drug-disease association prediction method provided in this embodiment is implemented. Among them, the storage medium can be a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a disk or an optical disk, etc., which can store program codes.

[0189] This embodiment also provides an electronic device for implementing the above-mentioned drug-disease association prediction method or model training method. Figure 5 As shown, the electronic device may include a processor 22 and a memory 21. In addition, the memory 21 stores a computer program, and the processor implements the drug-disease association prediction method or model training method provided in this embodiment by reading and executing the computer program corresponding to the above implementation in the memory 21.

[0190] Continue to see Figure 5 The electronic device further includes a communication unit 23. The memory 21, the processor 22 and the communication unit 23 are electrically connected to each other directly or indirectly through a system bus 24 to achieve data transmission or interaction.

[0191] The memory 21 may be an information recording device based on any electronic, magnetic, optical or other physical principle, used to record execution instructions, data, etc. In some embodiments, the memory 21 may be, but is not limited to, a volatile memory, a non-volatile memory, a storage drive, etc.

[0192] In some embodiments, the volatile memory may be a random access memory (RAM); in some embodiments, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable read-only memory (EEPROM), a flash memory, etc.; in some embodiments, the storage drive may be a disk drive, a solid-state drive, any type of storage disk (such as a CD, a DVD, etc.), or a similar storage medium, or a combination thereof, etc.

[0193] The communication unit 23 is used to send and receive data through a network. In some embodiments, the network may include a wired network, a wireless network, a fiber optic network, a telecommunication network, an intranet, the Internet, a local area network (LAN), a wide area network (WAN), a wireless local area network (WLAN), a metropolitan area network (MAN), a wide area network (WAN), a public switched telephone network (PSTN), a Bluetooth network, a ZigBee network, or a near field communication (NFC) network, or any combination thereof. In some embodiments, the network may include one or more network access points. For example, the network may include a wired or wireless network access point, such as a base station and / or a network switching node, through which one or more components of the service request processing system may be connected to the network to exchange data and / or information.

[0194] The processor 22 may be an integrated circuit chip having a signal processing capability, and the processor may include one or more processing cores (e.g., a single-core processor or a multi-core processor). By way of example only, the processor may include a central processing unit (CPU), an application specific integrated circuit (ASIC), an application specific instruction set processor (ASIP), a graphics processing unit (GPU), a physical processing unit (PPU), a digital signal processor (DSP), a field programmable gate array (FPGA), a programmable logic device (PLD), a controller, a microcontroller unit, a reduced instruction set computer (RISC), or a microprocessor, or any combination thereof.

[0195] Understandably, Figure 5 The structure shown is for illustration only. The electronic device may also have Figure 5 More or fewer components than shown, or with Figure 5 Different configurations are shown. Figure 5 The components shown may be implemented in hardware, software or a combination thereof.

[0196] It should be understood that the apparatus and method disclosed in the above-mentioned embodiments can also be implemented in other ways. The apparatus embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the apparatus, methods and computer program products according to the multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or the flowchart, and the combination of boxes in the block diagram and / or the flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0197] The above are only various implementations of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A model training method, characterized in that: The method comprises: Obtaining original drug-disease association graphs, drug-drug original association graphs, and disease-disease original association graphs; Decomposing and reconstructing the original drug-disease association graph matrix into a drug-drug reconstructed association graph, a drug-disease reconstructed association graph, and a disease-disease reconstructed association graph; The graph neural network is used to compare and learn the original drug-disease association graph with the reconstructed drug-disease association graph, the original drug-drug association graph with the reconstructed drug-drug association graph, and the original disease-disease association graph with the reconstructed disease-disease association graph, until the graph neural network meets the conditions for serving as a drug-disease association prediction model.

2. The model training method according to claim 1, characterized in that: The step of decomposing and reconstructing the original drug-disease association graph matrix into a drug-drug reconstructed association graph, a drug-disease reconstructed association graph, and a disease-disease reconstructed association graph comprises: Generate a drug-disease original association matrix based on the drug-disease original association graph; Performing probability matrix decomposition on the original association matrix of the drug and the disease to obtain a drug potential feature matrix and a disease potential feature matrix; According to the drug potential feature matrix and the disease potential feature matrix, a drug-disease reconstructed association matrix, a drug-drug reconstructed association matrix and a disease-disease reconstructed association matrix are obtained; According to the drug-disease reconstruction association matrix, the drug-drug reconstruction association matrix and the disease-disease reconstruction association matrix, the drug-drug reconstruction association graph, the drug-disease reconstruction association graph and the disease-disease reconstruction association graph are obtained.

3. The model training method according to claim 1, characterized in that: The comparative learning of the original drug-disease association graph and the reconstructed drug-disease association graph, the comparative learning of the original drug-drug association graph and the reconstructed drug-drug association graph, and the comparative learning of the original disease-disease association graph and the reconstructed disease-disease association graph by using a graph neural network includes: Extracting features from the original drug-disease association graph and the reconstructed drug-disease association graph through the graph neural network to obtain an embedded representation of the original drug-disease association graph and an embedded representation of the reconstructed drug-disease association graph; Obtaining a contrast loss between drugs and diseases and a binary cross entropy loss according to the embedding representation of the original association graph between drugs and diseases and the embedding representation of the reconstructed association graph between drugs and diseases; Performing feature extraction on the original drug-drug association graph and the reconstructed drug-drug association graph through the graph neural network to obtain an embedded representation of the original drug-drug association graph and an embedded representation of the reconstructed drug-drug association graph; According to the embedding representation of the original association graph between the drugs and the embedding representation of the reconstructed association graph between the drugs and the drugs, a contrast loss between the drugs and the drugs and a binary cross entropy loss are obtained; Performing feature extraction on the original disease-disease association graph and the reconstructed disease-disease association graph through the graph neural network to obtain an embedded representation of the original disease-disease association graph and an embedded representation of the reconstructed disease-disease association graph; Obtaining a contrast loss between diseases and a binary cross entropy loss according to the embedding representation of the disease-disease original association graph and the embedding representation of the disease-disease reconstructed association graph; The drug-disease contrast loss and binary cross entropy loss, the drug-drug contrast loss and binary cross entropy loss, and the disease-disease contrast loss and binary cross entropy loss are weighted to obtain a comprehensive loss; The network parameters of the graph neural network are updated according to the comprehensive loss.

4. The model training method according to claim 3, characterized in that: The method of obtaining the contrast loss between drugs and diseases and the binary cross entropy loss based on the embedding representation of the original association graph between drugs and diseases and the embedding representation of the reconstructed association graph between drugs and diseases comprises: Compare the embedding representation of the original association graph of drugs and diseases with the embedding representation of the reconstructed association graph of drugs and diseases from the row and column respectively to obtain the row comparison loss of drugs and diseases and the column comparison loss of drugs and diseases; Weighting the row contrast loss between the drug and the disease and the column contrast loss between the drug and the disease to obtain the contrast loss between the drug and the disease; Obtaining association scores between multiple drugs and multiple diseases according to the embedded representation of the original association graph between the drugs and the diseases and the embedded representation of the reconstructed association graph between the drugs and the diseases; According to the association scores between the multiple drugs and the multiple diseases, the binary cross entropy loss between the drugs and the diseases is obtained.

5. The model training method according to claim 4, characterized in that: The binary cross entropy loss between the drugs and the diseases is obtained according to the association scores between the multiple drugs and the multiple diseases, including: Determining association scores of known drug-disease pairs and predicted association scores of drug-disease pairs from the association scores between the multiple drugs and the multiple diseases; The binary cross entropy loss between the drug and the disease is obtained according to the association score of the known drug-disease pair and the association score of the predicted drug-disease pair.

6. The model training method according to claim 4, characterized in that: The step of obtaining association scores between multiple drugs and multiple diseases based on the embedding representation of the original association graph between the drugs and the diseases and the embedding representation of the reconstructed association graph between the drugs and the diseases comprises: fusing the embedding representation of the original association graph of drugs and diseases with the embedding representation of the same drug in the embedding representation of the reconstructed association graph of drugs and diseases to obtain a comprehensive embedding representation of multiple drugs; fusing the embedding representation of the original association graph of drugs and diseases with the embedding representation of the same disease in the embedding representation of the reconstructed association graph of drugs and diseases to obtain a comprehensive embedding representation of multiple diseases; According to the comprehensive embedding representation of multiple drugs and the comprehensive embedding representation of multiple diseases, the association scores between multiple drugs and multiple diseases are obtained.

7. A method for predicting the association between drugs and diseases, characterized in that: The method comprises: Obtain the original association diagram between the drug and the disease to be analyzed; Decomposing and reconstructing the original association graph matrix of the drug to be analyzed and the disease into a reconstructed association graph of the drug to be analyzed and the disease; The drug-drug reconstructed association graph to be analyzed and the drug-disease original association graph to be analyzed are processed by a drug-disease association prediction model trained by the model training method according to any one of claims 1 to 6 to obtain drug-disease association prediction information.

8. A model training device, characterized in that: The device comprises: A graph decomposition module, used to obtain the original association graph between drugs and diseases, the original association graph between drugs and drugs, and the original association graph between diseases; The graph decomposition module is further used to decompose and reconstruct the original drug-disease association graph matrix into a drug-drug reconstructed association graph, a drug-disease reconstructed association graph, and a disease-disease reconstructed association graph; The model training module is used to compare and learn the original drug-disease association graph with the reconstructed drug-disease association graph, the original drug-drug association graph with the reconstructed drug-drug association graph, and the original disease-disease association graph with the reconstructed disease-disease association graph through a graph neural network, until the graph neural network meets the conditions of serving as a drug-disease association prediction model.

9. A storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by the processor, it implements the model training method described in any one of claims 1 to 6 or the drug-disease association prediction method described in claim 7.

10. An electronic device, characterized in that: The electronic device includes a processor and a memory, the memory stores a computer program, and when the computer program is executed by the processor, it implements the model training method described in any one of claims 1 to 6 or the drug-disease association prediction method described in claim 7.

Images