A network similarity integration method for predicting microRNA-disease associations

Abstract

Increasing evidence has indicated that microRNAs (miRNAs) regulate gene expression at the post-transcriptional level. Aberrant miRNA expression has been associated with many types of human disease, including cancers. Their associations can be used to understand the pathogenesis of diseases. However, using experimental methods to identify the associations between diseases and miRNAs is time consuming and costly. Computational methods could find the most promising miRNA-disease associations in a short time, thereby significantly reducing experimental time and cost. This paper presents a network similarity integration method (NSIM) for predicting potential miRNA-disease associations, considering that diseases associated with highly related miRNAs are more similar (and vice versa). The NSIM is based on 5425 experimentally verified human miRNA-disease associations, which consist of 495 miRNAs and 381 diseases. The NSIM integrates the disease similarity network, miRNA similarity network, and known miRNA-disease association network on the basis of cousin similarity to predict novel miRNA-disease associations. We evaluate the NSIM using leave-one-out cross validation. The area under the curve of the method is 0.9475, indicating its outstanding performance. Case studies on prostate, breast, and colon neoplasms further proved the outstanding performance of the NSIM to predict not only disease-related miRNAs but also isolated diseases (diseases without any related miRNAs).