Two-Stage Inference for LncRNA-Disease Associations Based on Diverse Heterogeneous Information Sources

Abstract

Long non-coding RNAs (lncRNAs) exert impacts on multiple fundamental and important biological processes. Many lncRNAs have been functionally associated with cancers. Utilizing experimental and bioinformatics approaches to identify and annotate lncRNAs with cancer-associated roles is laborious and time-consuming. Therefore, more and more researchers have focused on computational methods as an alternative candidate to find out unknown associations between lncRNAs and diseases, in particular, cancers. In this study, under the situation that there were few known lncRNA-disease associations out of huge unknown associations, we explored a novel two-stage prediction model (namely DRW-BNSP) for inferring lncRNA-disease associations: In the first stage, we designed a Dual Random Walk (DRW) model to obtain the primary prediction scores by walking on two combined similarity networks which were reconstructed; In the second stage, we used a Bipartite Network Space Projection (BNSP) model to make the primary prediction scores to be more accurate furtherly. Compared with other state-of-the-art methods in similar type, our DRW-BNSP could not only function on new lncRNAs and isolated diseases, but also achieve higher AUC value of 0.9344 and 0.9432 on the first dataset (namely Dataset1) and second dataset (namely Dataset2) built by us. Furthermore, case study further confirmed the predictive dependability of our DRW-BNSP for inferring potential lncRNA-disease associations.