A Network Embedding-Based Method for Predicting miRNA-Disease Associations by Integrating Multiple Information

Abstract

MicroRNAs (miRNAs) play important roles in various human complex diseases. Therefore, identifying miRNA-disease associations is deeply significant for pathological progress, diagnosis, and treatment of complex diseases. However, considering the expensive and time-consuming of traditional biological experiments, more and more attentions have been paid on developing computational methods for predicting miRNA-disease associations (MDAs). In this paper, we propose a novel network embedding-based method for predicting miRNA-disease associations by integrating multiple information. Firstly, we constructed a multi-molecular associations network by integrating five known molecules and the associations among them. Then, the behavior features of miRNAs and diseases are extracted by the network embedding model Laplacian Eigenmaps. Finally, Random Forest classifier is trained to predict associations between miRNAs and diseases. As a result, the proposed method achieved outstanding performance on the HMDD V3.0 dataset by using five-fold cross validation, whose average AUC could be reached 0.9317. The promising results demonstrate that the proposed model is a reliable model for the prediction of potential miRNA-disease associations.