Research On Mirna-disease Association Prediction Algorithm Based On Network Topology Technology And Multi-source Information

MicroRNAs(miRNAs)are short non-coding RNA molecules regulating the expression of other genes in biological processes and forming eukaryotic cell-dependent gene expression programs.Exploring miRNA-disease associations not only can decipher pathogenic mechanisms but also provide treatment solutions for diseases.As it is inefficient to identify undiscovered relationships between diseases and miRNAs using biotechnology,explosion of computational methods have been advanced.However,the prediction accuracy of existing models is hampered by the sparsity of known association network,single-category feature,insufficient use of similarity information and lack of efficient learning graph structure information and node attributes,which is hard to model the complicated relationships between diseases and miRNAs.Therefore,this paper proposes two novel miRNA-disease association prediction methods based on attention mechanism.The main contents are as follows:(1)In order to solve the problem of insufficient use of similarity information in current models and no efficient learning of graph structure information and node attributes,this study proposes a miRNA-disease association prediction method GATEMDA based on stacked denoising autoencoder and graph attention auto-encoder.Firstly,the model incorporates Gaussian kernel similarity for miRNA similarity network and disease similarity network respectively.Secondly,the stacked denoising autoencoder is used to learn the similarity network for getting dense low-dimensional features,then the graph attention autoencoder is used to reconstruct the node attributes and graph structures of the miRNA-disease association network to obtain the association network features of each node.Finally,random forest algorithm is used to fuse the similarity network features and association network features of miRNA and disease,in order to infer the association pairs between miRNAs and diseases.This study verify that the proposed method has higher classification performance than the exist binary network method through the five-fold cross-validation method,and prove the stability and reliability of the prediction results in practical application by case study.(2)In order to solve the problems that current model is limited by sparse association network and single feature type,we advance a new computational framework(GATMDA)to discover unknown miRNA-disease associations based on graph attention network with multisource information.In our method,the linear features of diseases and miRNAs are constructed by disease-lnc RNA correlation profiles and miRNA-lnc RNA correlation profiles,respectively.Then,the graph attention network is employed to extract the non-linear features of diseases and miRNAs by aggregating information of each neighbor with different weights.Finally,the random forest algorithm is applied to infer the disease-miRNA correlation pairs through fusing linear and non-linear features of diseases and miRNAs.As a result,GATMDA achieves impressive performance: an average AUC of 0.9566 with five-fold cross validation,which is superior to other previous models.In addition,case studies of breast cancer,colon cancer,and lymphoma could verify the model’s great independent predictive performance.