Degree-Based Similarity Indexes for Identifying Potential miRNA-Disease Associations

Abstract

Identifying disease-associated miRNAs is helpful to explore the pathogenesis of diseases. However, without foreknowledge of the experimentally valid disease-associated miRNAs information, the development of promising and affordable approaches for effective treatment of human diseases is challenging. In this study, we develop DCNMDA and DJMDA, a degree-based similarity indexes methodology for identifying potential miRNAs-disease associations. We solely focused on the similarity and the degree between nodes without adopting negative samples or other external prior information beyond the miRNA-disease associations bipartite network. Trained on HMDD v2.0 and HMDD v3.0, DCNMDA achieved the highest AUCs (0.9237 and 0.9432, respectively) based on the 5-fold cross-validation and outperformed the published state-of-the-art methodologies. Moreover, case studies about breast neoplasms, lung neoplasms, and ovarian neoplasms further evaluate the reliability of the models. As a result, biological experiments can correspondingly verify 28 out of top-30 DJMDA-predicted MDAs and 29 out of top-30 DCNMDA-predicted MDAs. In summary, DCNMDA and DJMDA offer a powerful degree-based similarity index approach for identifying potential miRNAs-disease associations with superior performance.

Keywords

• miRNA-disease associations
• degree-based similarity indexes
• common neighbor and Jaccard
• area under curve (AUC)
• area under the precise recall curve (AUPR)