Predicting miRNA-Disease Associations via Meta-Path Embedding

MicroRNAs (miRNAs) play a critical role in various biological processes, making the accurate identification of miRNA-disease associations essential for understanding human diseases. Existing methods for predicting these associations often lack interpretability. In this study, we introduce MNEMDA, a novel computational method that leverages meta-path-based network embedding (metapath2vec) to extract features from a heterogeneous network of miRNA-disease interactions. Using the XGBoost classifier, MNEMDA predicts potential miRNA-disease associations. Our approach achieves an overall AUC of 0.959 and an overall AUPR of 0.952, outperforming several state-of-the-art models with an average AUC score of 0.958 across 15 common diseases. Case studies on renal cell carcinoma and breast neoplasms confirm MNEMDA's efficacy, with 93% and 97% of the top 30 predicted miRNAs, respectively, validated by recent experimental reports. Survival analysis also verifies the effectiveness and practical value of MNEMDA. Additionally, MNEMDA maintains excellent performance on other datasets, demonstrating its generalization ability. Moreover, MNEMDA shows insensitivity to input data noise, providing a robust and interpretable tool for predicting miRNA-disease associations.