Research On MiRNA-Disease Association Types Predicting Algorithm Based On Similarity Networks

MicroRNAs(MiRNAs)are a class of endogenous non-coding RNAs.Recently,increasing evidence suggests that the mutation and dysregulation of miRNAs may lead to a variety of diseases,which indicates that miRNAs can serve as efficient biomarkers for disease detection and prognosis.In the early stage,disease-associated miRNAs with highly accurate can be obtained by biological experiment,which faces severe challenges,such as long expensive equipment and experimental periods.Therefore,identifying regulatory relationship between miRNAs and diseases by computational methods has become a research hotspot in recent years,which is crucial for predicting miRNA-disease associations.In this paper,a high quality similarity network was constructed based on multiple biological data,and two computational models are proposed to predict miRNA-disease association types,which is network-based label propagation algorithm to predict miRNA-disease association types(NLPMMDA)and similarity-based tensor decomposition algorithm to predict miRNA-disease association types(STDMMDA).NLPMMDA and STDMMDA are both integrating disease similarity and miRNA similarity.In addition,NLPMMDA method takes full use of similarity information to construct the heterogeneous network,then the network-based label propagation algorithm was implementing on heterogeneous network based on mutual information of nodes,which can predict the label information of a large number of unlabeled data by label information of a small number of labeled data.Besides,STDMMDA method takes use of similarity information and collaborative filtering ideas to update three dimensional tensor of association types,and the association scores of miRNA-disease interaction under four types are obtained,then similarity-based tensor decomposition was implementing on three dimensional tensor to fill data and obtain underlying information.The methods proposed in this paper integrate disease similarity and miRNA similarity data,which can better excavate potential information.Besides,NLPMMDA method is a semi-supervised machine learning model,which does not require verified negative miRNA-disease associations.And STDMMDA method uses tensor decomposition model to fill data,which can avoid errors caused by negative samples.In addition,STDMMDA method can be applied to isolated diseases.Leave-one-out cross validation was implemented and the value of micro-average AUC of NLPMMDA and STDMMDA method is 0.9736 and 0.8628 respectively,which is much better than advanced method of 0.8606.In addition,the case studies demonstrated the reliable and effective performance of NLPMMDA and STDMMDA method.