Predicting candidate disease-related lncRNAs based on network random walk

With a large number of long non-coding RNAs (lncRNAs) identified over the past decades, more and more evidences have indicated that mutations and dysregulations of lncRNAs have a close relationship with many complex human diseases. Therefore, identification of potential disease related lncRNAs is an effective means for improving the quality of disease diagnostics and treatments. In this article, we proposed a computational model for potential disease related lncRNAs identification based on multiple biological datasets. Adopting the recommendation strategy of Collaborative Filtering, we calculated functional associations between lncRNAs with different data sources as dimensions. Subsequently, a disease associated lncRNA network was built with functional similarities between lncRNAs as weight. And potential disease related lncRNAs could be identified based on scores ranking derived by random walking with restart (RWR). Then we extracted train set and test sets from two different versions of disease-IncRNA datasets and assessed the performance of this method on 54 diseases. As a result, its average AUC (area under the receiver operating characteristic curve) reached 78.08%. Sufficient validations show that this method has a good performance for identifying potential disease-IncRNA pairs.