A Deep Metric Learning Based Method for Predicting MiRNA-Disease Associations

MicroRNAs (miRNAs) play crucial roles in various human diseases. Identifying miRNA-disease associations (MDAs) can supply us with an understanding for prevention, diagnosis, and treatment of diseases. However, predicting MDAs through traditional biomedical experiments is time-consuming and expensive. So, in this study, we propose a computational method (named TRNMDA) for MDA prediction based on the optimization of similarity weights using the FKL (Fast kernel learning) model to enhance the quality of dataset and utilize two triplet networks in learning feature vectors. From that, it can enhance the performance of the predictive model. We have demonstrated the effectiveness of our proposed method by conducting 5-fold Cross-Validation (5-fold-CV) of TRNMDA and eight compared methods. Experiment results also demonstrate that TRNMDA achieves the highest AUC and AUPR values ( \(0.9666\) and \(0.7305\) (on HMDD v2.0), \(0.9637\) and \(0.7157\) (on HMDD v3.2), respectively). To further illustrate the prediction ability of TRNMDA on real applications, we also conduct a case study in Breast Neoplasms disease and validation result show that all of the top 50 predicted miRNAs are confirmed for this disease. Results show that TRNMDA is an efficient computational model for predicting MDAs. The source code in this work is available at https://github.com/npxquynhdhsp/TRNMDA .