Study On MiRNA And Disease Associations Prediction Based On Graph Representation Learning

MiRNAs(MicroRNAs)are a class of non-coding RNA molecules that bind to target gene m RNA to inhibit its translation or degradation.Mi RNAs play important roles in many biological processes,including cell proliferation,differentiation,and cell death.More and more studies have shown that miRNAs are closely related to diseases,such as tumors,cardiovascular diseases,neurological diseases,etc.Therefore,the identification of potential miRNA associated with disease has practical application value and wide application prospect in the field of disease diagnosis and treatment.Due to the long cycle and high cost of traditional experimental verification methods,an efficient method is urgently needed to reduce the research cost.With the rapid development of computer technology,computing methods have been widely used,among which the method of using graph to represent learning has attracted much attention in recent years.Based on previous studies,this thesis uses two graph representation learning methods to predict the association between miRNAs and diseases,which are described as follows:A novel miRNA-disease association prediction model based on jump knowledge network architecture and graph attention model JKNMDA is proposed.This method effectively alleviates the defects caused by the fixed aggregation layers or deepening of the traditional messaging methods in neural networks.In the specific process,the heterogeneity graph is constructed by using the known similarity information and association information of miRNA and disease,and then GAT model is used to aggregate the information of each neighbor layer.The aggregated information is fused by jumping knowledge network to obtain the final embedded representation of nodes.Finally,the embedded miRNA and the node vector of the disease are spliced together as side information into the full connection layer to obtain the final prediction probability.Experimental results show that the model has good predictive performance under the five-fold cross validation,and the AUC values of HMDD v2.0 and HMDD v3.0 datasets are 93.30% and 94.54%,respectively.In the disease case studies,45,46 and 45 of the top 50 miRNAs predicted from the model were confirmed by the db DEMC and miR2 Disease datasets for lymphoma,esophageal and prostate tumors,respectively.A graph attention network model based on structured interaction was proposed to predict miRNA-disease association(SIGATMDA).This method expands the aggregation mode of the attention mechanism of the original graph and makes full use of the topological features of the graph.The specific calculation process is as follows:firstly,the feature similarity score between the two nodes processed by the graph attention mechanism is obtained.Then,the structured fingerprint interaction score of the two nodes is obtained by combining the restart random walk method and the Jaccard similarity calculation,and finally the joint attention score is obtained.Then the original node features are updated according to the joint attention score of the neighbor nodes in the neighborhood.The average AUC values of the SIGATMDA model in the HMDD v2.0 and HMDD v3.0 data sets were 93.71% and 94.72%,respectively.In case studies of esophageal,colon,and prostate tumors,46,46 and 45 of the top 50 disease-related miRNAs predicted were confirmed by db DEMC and miR2 Disease datasets,respectively.