MiRNA disease association prediction method based on global and local graph representation learning

Through multi-source information calculation and graph Transformer combined with adaptive graph convolutional neural network, the problem of insufficient feature oversmooth and long-distance dependency information capture in the GCN method is solved, the performance of miRNA disease association prediction is improved, and high-quality nodes are provided Characteristic representation and credible disease candidate miRNA reduce experimental costs and time.

CN118969089AActive Publication Date: 2024-11-15XINJIANG UNIVERSITY
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202411196676.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-29
Publication Date
2024-11-15
Estimated Expiration
2044-08-29

AI Technical Summary

Technical Problem

The existing GCN-based miRNA disease association prediction method cannot fully extract the local feature information of nodes, resulting in excessive smoothness of features, affecting prediction performance, and unable to effectively capture long-distance dependence information in the graph, limiting the quality and globality of node feature representation .

Method used

Multi-source information is used to calculate multiple similarity matrices of miRNA and disease, build a comprehensive similarity feature matrix and use an autoencoder to fuse it, and build a two-part graph of miRNA-disease attributes, combining graph Transformer and adaptive graph convolutional neural network. The global and local features of the node are extracted separately to avoid oversmoothing of features, and to predict features through fully connected neural networks.

Benefits of technology

Adaptive feature smoothing is achieved, local features of nodes are fully extracted, the capture ability of global features is improved, high-quality node feature representation is obtained, prediction performance is improved, credible disease candidate miRNA is provided, and experimental cost and time is reduced. .

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118969089A_ABST
    Figure CN118969089A_ABST
Patent Text Reader

Abstract

The invention provides a miRNA disease association prediction method based on global and local graph representation learning, and belongs to the technical field of miRNA disease association prediction. The miRNA disease association prediction method based on global and local graph representation learning comprises the following steps: calculating a plurality of similarity matrixes by using multi-source information of miRNA and diseases, fusing the similarity matrixes, and constructing a miRNA-disease attribute bipartite graph according to known association information and fused similarity characteristics; a graph Transform neural network is used to extract node global feature representation; extracting node local feature representation by using an adaptive graph convolutional neural network; fusing the global feature representation and the local feature representation, and predicting potential miRNA disease association by using a multi-layer perceptron classifier; according to the method, by fusing global and local feature representations of nodes, miRNA and diseases can be comprehensively and deeply depicted, high-precision prediction is realized, and credible candidate miRNA is provided for scientific research personnel to determine association between miRNA and diseases.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of bioinformatics, and in particular to the application of computer technology to miRNA disease association prediction, and specifically relates to a miRNA disease association prediction method based on global and local graph representation learning. Background Art

[0002] MiRNA is a non-coding single-stranded RNA that can regulate gene expression at the post-transcriptional level by base pairing with messenger RNA, leading to mRNA degradation or inhibiting its translation. MiRNA dysregulation or abnormal expression is closely related to the occurrence and development of diseases. Understanding the relationship between miRNA and disease helps to understand the pathological mechanism of the disease, assist in the diagnosis and treatment of the disease, and promote the development of new disease treatment plans. Using traditional biological experimental methods to determine the association between miRNA and disease is time-consuming, labor-intensive, and expensive, and cannot be widely promoted. Therefore, it is of great significance to use computational methods to predict potential miRNA-disease associations and provide researchers with disease candidate miRNAs.

[0003] Current computational methods mainly include similarity-based methods and machine learning-based methods. Similarity-based methods calculate the similarity between miRNA and disease in a variety of ways and use this to rank the association probabilities of miRNA-disease pairs. Limited by the learning ability of the methods used, the prediction performance of similarity-based methods is not high. Machine learning-based methods use neural networks to extract feature representations of miRNA and disease for subsequent predictions. With the rise of graph representation learning methods represented by Graph Convolutional Network (GCN) and the accumulation of a large amount of miRNA-disease association data, it is possible to model graph structure data, and many GCN-based miRNA-disease association prediction methods have been developed.

[0004] In existing GCN-based prediction methods, each node performs the same number of message transmissions and feature updates, a feature smoothing process. However, this operation cannot meet the needs of all nodes in the graph, making it impossible to fully extract the local feature information of the nodes. In addition, in order to aggregate the feature information of distant nodes, multiple GCN layers need to be stacked, which will cause excessive smoothing of node features, affecting the prediction performance of downstream classifiers and making it impossible to extract the global feature information of the nodes well. The inability to extract high-quality, comprehensive node feature representations limits the performance of existing prediction methods. Summary of the invention

[0005] In view of the shortcomings of the existing technical methods, the present invention discloses a miRNA-disease association prediction method based on global and local graph representation learning, which solves the problem that high-quality and comprehensive miRNA and disease node feature representations cannot be extracted, resulting in limited miRNA-disease association prediction effects.

[0006] The scheme adopted by the present invention is as follows:

[0007] Multiple similarity matrices of miRNA and disease were calculated using multi-source information, and a comprehensive similarity feature matrix of nodes was constructed and fused using two autoencoders (AE) to obtain a fused similarity feature matrix of miRNA and disease.

[0008] According to the fusion similarity and known miRNA-disease association information, a miRNA-disease attribute bipartite graph is constructed with miRNA and disease as nodes, fusion similarity as the initial feature of the node, and known association as the edge. .

[0009] Computational Graph The degree of the nodes in the graph is calculated, and the degree fusion module is constructed and added to the Transformer framework to construct the graph Transformer neural network; the attributed bipartite graph Input into the network to extract the global feature embedding representation of the node.

[0010] According to the figure , an adaptive graph convolutional neural network is used to calculate the number of feature smoothing times required for each node in the graph, so as to fully extract the local feature embedding representation of the node and avoid the problem of feature over-smoothing.

[0011] A fully connected neural network is used to fuse the global and local features of nodes; the features of miRNA and disease are spliced ​​to construct a miRNA-disease pair feature representation; and a multi-layer perceptron classifier is used to predict potential associations.

[0012] The beneficial effects of the present invention include:

[0013] 1. Calculate the number of feature smoothing operations required for each node to adaptively aggregate the local features of the node and avoid the over-smoothing problem in the GCN feature extraction process;

[0014] 2. Use graph Transformer neural network to fully extract the global features of nodes, alleviating the problem that GCN cannot effectively capture long-distance dependent information in the graph;

[0015] 3. Fusion of global and local features to obtain high-quality, comprehensive miRNA and disease node feature representations to improve the performance of the prediction method;

[0016] 4. Provide researchers with reliable disease candidate miRNAs to reduce experimental time and cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a flow chart of the method of the present invention;

[0018] Figure 2 AUC curve comparison diagram of the method of the present invention and the neural network method in Figure 4;

[0019] Figure 3 4 is a comparison chart of the AUPR curves of the method of the present invention and the neural network method in Figure 4;

[0020] Figure 4 The figure is an experimental result diagram (ablation experiment result diagram) for verifying the effectiveness of each part of the method of the present invention. DETAILED DESCRIPTION

[0021] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0022] Example 1

[0023] The present invention provides a miRNA disease association prediction method based on global and local graph representation learning. Figure 1 As shown, the specific implementation of this method is as follows:

[0024] S1. Calculate multiple similarity matrices of miRNA and disease using multi-source information, construct a comprehensive similarity feature matrix of nodes and use two autoencoders (AE) to fuse them separately to obtain the fusion similarity feature matrix of miRNA and disease:

[0025] In this example, data was obtained from the Human MicroRNA Disease Database (HMDD v3.2), and 14,550 association information between 792 diseases and 917 miRNAs was obtained after processing. Record related information.

[0026] In this embodiment, the directed acyclic graph (DAG) structure of the disease is obtained from the medical thesaurus Mesh, which is used to calculate the semantic similarity matrix DSM of the disease:

[0027] For target diseases , any disease in its DAG right The semantic contribution of is:

[0028]

[0029] disease and The semantic similarity is:

[0030]

[0031] in and Diseases and A collection of nodes in a DAG.

[0032] Gaussian Interaction Spectrum Kernel Similarity Matrix of Diseases The calculation of is as follows:

[0033]

[0034]

[0035] in is the adjacency matrix No. List.

[0036] Gaussian interaction profile kernel similarity matrix of miRNA The calculation of is as follows:

[0037]

[0038]

[0039] In this example, the sequence information of miRNA was obtained from the miRBase database, and the similarity between two sequences was calculated using the Needleman-Wunsch algorithm to obtain the sequence similarity matrix of miRNA. The semantic similarity between the disease sets associated with two miRNAs is used to obtain the miRNA functional similarity matrix .

[0040] Comprehensive Disease Similarity Matrix and miRNA similarity matrix The formula is as follows:

[0041]

[0042]

[0043] Use two AEs to and Perform dimensionality reduction and fusion to obtain the fusion similarity matrix of the disease and miRNA fusion similarity matrix .

[0044] S2. Construct a miRNA-disease attribute bipartite graph based on the fused features and known association information:

[0045] In this example, miRNA and disease are used as nodes, the fusion similarity matrix is ​​used as the initial feature of the node, and the known association is used as the edge to construct the miRNA-disease attribute bipartite graph. ,picture The adjacency matrix of , the node feature matrix is ​​recorded as .

[0046] S3. Based on the attributed bipartite graph, the graph Transformer neural network is used to extract the global feature embedding of the nodes, which is implemented as follows:

[0047] S301, based on attributed bipartite graph The degree information of the nodes in the graph is used to build a degree fusion module to inject the inductive bias information of the graph;

[0048] S302. Build a Transformer-based node feature extraction framework, add the degree fusion module to the framework, and construct a graph Transformer neural network to extract the global features of the nodes.

[0049] The formula for extracting features from a multi-headed Transformer is as follows:

[0050]

[0051]

[0052] in According to the adjacency matrix The position encoding obtained by the Laplace transform matrix is is the parameter matrix. Use the multi-head attention mechanism to learn node features at different levels and concatenate and reduce the dimension of each head’s output to obtain .

[0053] In this embodiment, in order to utilize the structural information of the graph and construct a degree fusion module, the calculation formula is as follows:

[0054]

[0055] in For Node The degree of and are learnable parameters.

[0056] Use feedforward neural network and batch normalization layer to stabilize the training of the model. Finally, the global feature representation of the node is obtained. :

[0057]

[0058]

[0059] S4. According to the attributed bipartite graph, an adaptive graph convolutional neural network is used to calculate the required number of feature smoothing times for each node in the graph to fully extract the local feature embedding of the node. The implementation is as follows:

[0060] S401, respectively, attributed bipartite graphs For each node in , calculate the number of feature smoothing times required;

[0061] S402. According to the adaptive feature smoothing times, a simplified graph convolutional neural network is used to extract local features of nodes to avoid the problem of feature over-smoothing.

[0062] In this embodiment, the NDLS model is used to calculate the graph The number of adaptive feature smoothing times required for each node in is as follows:

[0063]

[0064] in is an arbitrarily small constant greater than 0. According to the number of feature smoothing, the activation function in GCN is deleted to extract the local features of the node :

[0065]

[0066] S5. Fusion of global and local features to construct miRNA-disease pair feature representation, and prediction of potential associations using a multi-layer perceptron classifier, which is implemented as follows:

[0067] S501, using a fully connected neural network to fuse the global features and local features of the node to obtain a final feature representation of the node;

[0068] S502, concatenating the miRNA features and the disease features to obtain a feature representation of the miRNA-disease pair;

[0069] S503. Using a multi-layer perceptron classifier, predict potential associations based on the characteristics of miRNA-disease pairs.

[0070] In this embodiment, the formula for using the fully connected neural network to fuse features is as follows:

[0071]

[0072] in is the parameter matrix, For deviation.

[0073] The miRNA features and disease features are concatenated to obtain the feature representation of the miRNA disease pair. This is then input into a multilayer perceptron (MLP) classifier for association prediction:

[0074]

[0075] In this embodiment, the cross entropy loss function is used to optimize the parameters of the neural network.

[0076] In this example, five-fold cross validation was used to test the performance of the method provided by the present invention and compared with four other graph neural network-based models. Figure 2 and Figure 3 As shown in the figure, the AUC and AUPR values ​​of our method (our) are higher than those of the four compared models.

[0077] In this example, the effectiveness of each part of the proposed method is verified through ablation experiments. The method of using only global features (our_global) and local features (our_local) is compared with the method of using Transformer to extract global features and using GCN to extract local features. Figure 4 As shown in the figure, both the global features and local features extracted by the proposed method are effective and necessary, and using the two features simultaneously can achieve better prediction performance.

[0078] In this example, in order to fully verify the use prospects of the proposed method in real application scenarios, breast tumors were selected for case study experiments. The potential miRNA-disease associations predicted by the method were verified by searching for relevant literature evidence in the PubMed database. As shown in Table 1, the top 30 potential associated miRNAs predicted by the method were verified.

[0079] Table 1 Results of breast tumor case studies

[0080] rank MiRNA Evidence(PMID) rank MiRNA Evidence(PMID) 1 hsa-mir-150 36340453 16 hsa-mir-181c 35611706 2 hsa-mir-142 36443711 17 hsa-mir-494 32372307 3 hsa-mir-15b 33519190 18 hsa-mir-99b 36381329 4 hsa-mir-106a 34837907 19 hsa-mir-30e 28288641 5 hsa-mir-192 31485620 20 hsa-mir-92b 29661250 6 hsa-mir-99a 33177704 21 hsa-mir-424 36305147 7 hsa-mir-378a 26255816 22 hsa-mir-491 33488122 8 hsa-mir-138 36340453 23 hsa-mir-184 34000513 9 hsa-mir-130a 35797350 24 hsa-mir-144 26124344 10 hsa-mir-212 33187481 25 hsa-mir-98 28232182 11 hsa-mir-196b 30416655 26 hsa-mir-483 34488224 12 hsa-mir-370 29698983 27 hsa-mir-542 34402683 13 hsa-mir-372 29456685 28 hsa-mir-130b 35191803 14 hsa-mir-208a 30911286 29 hsa-mir-134 34378285 15 hsa-mir-28 34593318 30 hsa-mir-181d 36340453

Claims

1. A miRNA-disease association prediction method based on global and local graph representation learning, It is characterized in that The miRNA-disease association prediction method based on global and local graph representation learning comprises the following steps: S1. Calculate multiple similarity matrices of diseases and miRNAs using multi-source information, construct a comprehensive similarity feature matrix of nodes and perform fusion; S2, constructing a miRNA-disease attribute bipartite graph based on the fusion features and known miRNA-disease associations; S3. Based on the attributed bipartite graph, the graph Transformer neural network is used to extract the global feature embedding of the nodes; S4. Based on the attributed bipartite graph, an adaptive graph convolutional neural network is used to calculate the number of feature smoothing times required for each node in the graph, so as to fully extract the local feature embedding of the node and avoid the problem of feature over-smoothing; S5. Fusion of global and local features to construct miRNA-disease pair feature representation, and use of multi-layer perceptron classifier to predict potential associations.

2. The miRNA-disease association prediction method based on global and local graph representation learning according to claim 1, It is characterized in that The comprehensive similarity feature matrix of the nodes in step S1 includes the comprehensive similarity matrix of the disease and the comprehensive similarity matrix of miRNA, and the corresponding expressions are as follows: Comprehensive similarity matrix of diseases: Comprehensive similarity matrix of miRNAs: in represents the semantic similarity matrix calculated based on the disease-based directed acyclic graph DAG, represents the Gaussian interaction spectrum kernel similarity matrix calculated based on the disease association information representation vector; represents the miRNA sequence similarity matrix calculated based on the Needleman-Wunsch algorithm, represents the miRNA functional similarity matrix calculated based on the associated disease groups, Represents the Gaussian interaction spectrum kernel similarity matrix calculated based on the association information representation vector of miRNA; after obtaining the comprehensive similarity matrices of miRNA and disease respectively, two autoencoders (AE) are used to map the features of miRNA and disease to the same low-dimensional space respectively to achieve dimensionality reduction and fusion of features.

3. The miRNA-disease association prediction method based on global and local graph representation learning according to claim 1, It is characterized in that The step S2 comprises: constructing a miRNA-disease attribute bipartite graph by taking miRNA and disease as nodes, the features fused by the autoencoder as the initial features of the nodes, and the associations between known miRNA and diseases recorded in the public database as edges. ;picture The adjacency matrix of , the node feature matrix is ​​recorded as .

4. The miRNA-disease association prediction method based on global and local graph representation learning according to claim 1, It is characterized in that The step S3 comprises the following steps: S301, based on attributed bipartite graph The degree information of the nodes in the graph is used to build a degree fusion module to inject the inductive bias information of the graph; S302. Build a Transformer-based node feature extraction framework, add the degree fusion module to the framework, and construct a graph Transformer neural network to extract the global features of the nodes.

5. The miRNA-disease association prediction method based on global and local graph representation learning according to claim 1, It is characterized in that The step S4 comprises the following steps: S401, respectively, attributed bipartite graphs For each node in , calculate the number of feature smoothing times required; S402. According to the number of adaptive feature smoothing times, a graph convolutional neural network is used to extract local features of the nodes to avoid over-smoothing of the features.

6. The miRNA-disease association prediction method based on global and local graph representation learning according to claim 1, It is characterized in that The step S5 comprises the following steps: S501, using a fully connected neural network to fuse the global features and local features of the node to obtain a final feature representation of the node; S502, concatenating the miRNA features and the disease features to obtain a feature representation of the miRNA-disease pair; S503. Using a multi-layer perceptron classifier, predict potential associations based on the characteristics of miRNA-disease pairs.

Citation Information

Patent Citations

  • MiRNA-disease association prediction method and device based on graph neural network fusion multi-view information

    CN112784913A

  • MiRNA-disease association prediction method and system based on similarity constraint

    CN113539479A

  • MiRNA-mRNA target prediction method based on sequence statistical characterization learning

    CN114664376A

  • CircRNA-disease incidence relation prediction method based on graph convolutional neural network and node2vec

    CN114999635A