Inferring novel lncRNA–disease associations based on a random walk model of a lncRNA functional similarity network

Accumulating evidence demonstrates that long non-coding RNAs (lncRNAs) play important roles in the development and progression of complex human diseases, and predicting novel human lncRNA-disease associations is a challenging and urgently needed task, especially at a time when increasing amounts of lncRNA-related biological data are available. In this study, we proposed a global network-based computational framework, RWRlncD, to infer potential human lncRNA-disease associations by implementing the random walk with restart method on a lncRNA functional similarity network. The performance of RWRlncD was evaluated by experimentally verified lncRNA-disease associations, based on leave-one-out cross-validation. We achieved an area under the ROC curve of 0.822, demonstrating the excellent performance of RWRlncD. Significantly, the performance of RWRlncD is robust to different parameter selections. Predictively highly-ranked lncRNA-disease associations in case studies of prostate cancer and Alzheimer's disease were manually confirmed by literature mining, providing evidence of the good performance and potential value of the RWRlncD method in predicting lncRNA-disease associations.