HNMDA: heterogeneous network-based miRNA-disease association prediction.

Recently, accumulating evidences have shown that microRNAs (miRNAs) could play key roles in the development and progression of multiple important human diseases. Nonetheless, due to the shortcoming of being expensive and time-consuming existing in experimental approaches, computational methods are needed for the prediction of potential miRNA-disease associations. In our study, we proposed a computational model named Heterogeneous Network-based MiRNA-Disease Association prediction (HNMDA) for the latent miRNA-disease association prediction by integrating known miRNA-disease associations, miRNA functional similarity, disease semantic similarity and Gaussian interaction profile kernel similarity. The Gaussian interaction profile kernel similarity can make up for the shortages of the traditional similarity calculation methods. Furthermore, we applied a heterogeneous network-based method, in which we first implemented a network diffusion algorithm of random walk with restart, and then we applied a method to find the optimal projection from miRNA space to disease space, which enabled the prediction of new miRNA-disease associations that are not experimentally confirmed so far. In the cross-validation, HNMDA obtained the AUC of 0.8394, which achieved improvement compared with previous methods. In the case studies of breast neoplasms, esophageal neoplasms and kidney neoplasms based on known miRNA-disease associations in the HMDD V2.0 database, there were 82, 76 and 84% of top 50 predicted related miRNAs that were confirmed to have associations with these three diseases, respectively. In the further case studies for new diseases without any known related miRNAs and the case using HMDD V1.0 database as known associations, there were also high ratio of the predicted miRNAs confirmed by experimental reports.