NARRMDA: negative-aware and rating-based recommendation algorithm for miRNA-disease association prediction

An increasing amount of evidence indicates that microRNAs (miRNAs) are closely related to many important biological processes and play a significant role in various human diseases. More and more researchers have begun to seek effective methods to predict potential miRNA-disease associations. However, reliable computational methods to predict potential disease-related miRNAs are lacking. In this study, we developed a new miRNA-disease association prediction model called Negative-Aware and rating-based Recommendation algorithm for miRNA-Disease Association prediction (NARRMDA) based on the known miRNA-disease associations in the HMDD database, miRNA functional similarity, disease semantic similarity and Gaussian interaction profile kernel similarity. NARRMDA combined a rating-based recommendation algorithm and a negative-aware algorithm to score and rank miRNAs without known associations with investigated diseases. Furthermore, we used leave-one-out cross validation to evaluate the accuracy of NARRMDA and compared NARRMDA with four previous classical prediction models (RLSMDA, HDMP, RWRMDA and MCMDA). As it turned out, NARRMDA and the other four prediction models achieved AUCs of 0.8053, 0.6953, 0.7702, 0.7891 and 0.7718, respectively, which proved that NARRMDA has superior performance of prediction accuracy. Furthermore, we verified the prediction results associated with colon neoplasms, esophageal neoplasms, lymphoma and breast neoplasms by two different validation schemas. In these case studies, 92%, 84%, 92%, and 100% of the top 50 potential miRNAs for these four diseases were confirmed by experimental discoveries, respectively. These results further show that NARRMDA has reliable performance of prediction ability.