MiRNA-Disease Association Prediction Based on Stacked Autoencoders and Variant Triplet Networks

MicroRNAs (miRNAs) have a significant role in the prevention, diagnosis, and treatment of complex diseases. However, identifying miRNA-disease associations (MDAs) through traditional experiments is costly and time-consuming. Recent studies have reported numerous validated MDAs, providing a foundation for discovering new MDAs using computational methods. In this study, we propose a computational method named SAETNMDA to predict MDAs, which is applied by Fast kernel learning (FKL) and variant triplet networks. First, we integrate miRNA and disease similarities into two integrated kernels through the FKL model which enriches the biological information. Second, feature representations are obtained from integrated similarities and two triplet networks with Stacked Autoencoders (SAEs). This facilitates the identification similarities or differences between associations in which miRNA-disease pairs are mapped to nearby locations in the embedding space, while unassociated ones are mapped to distant locations. Third, we utilize XGBoost to obtain the MDAs’ predictive scores from features obtained from these two triplet networks. We conduct 5-fold Cross-Validation (5-fold-CV) to evaluate SAETNMDA’s performance and compare it with other methods. SAETNMDA achieves the highest  AUC and AUPR values (0.9419 and 0.4749 (HMDD v2.0), 0.9496 and 0.5355 (HMDD v3.2), respectively). We also validate its performance on an independent dataset and de novo miRNAs. Results show SAETNMDA obtains the highest AUC and AUPR in all validations. Case studies further demonstrate its prediction ability for three diseases, with the top 50 predicted miRNAs confirmed for each. These results highlight SAETNMDA as an efficient model for predicting MDAs. The source code in this work is available at https://github.com/npxquynhdhsp/SAETNMDA.