Identifying Small Molecule-miRNA Associations Based on Credible Negative Sample Selection and Random Walk.

Area of Science:

• Biomedical Informatics
• Computational Biology
• Pharmacogenomics

Background:

• MicroRNAs (miRNAs) are emerging as critical targets for small molecule drugs.
• Identifying small molecule-miRNA associations (SMiRs) is vital for developing novel human disease therapies.
• Experimental validation of SMiRs is resource-intensive, necessitating efficient computational approaches.

Purpose of the Study:

• To develop and validate a novel computational model, RWNS, for predicting SMiR associations.
• To integrate diverse biological information and advanced network analysis for enhanced prediction accuracy.
• To provide a powerful tool for discovering potential therapeutic interventions through SMiR identification.

Main Methods:

• Designed RWNS, a framework integrating biological data, negative sample selection, and random walk on a triple-layer heterogeneous network.
• Employed similarity computation, rigorous negative sample selection, and random walk on a small molecule-disease-miRNA network.
• Validated performance using leave-one-out cross-validation (LOOCV) and 5-fold cross-validation, comparing with state-of-the-art methods.

Main Results:

• RWNS achieved high AUC values (e.g., 0.9829 LOOCV, 0.9916 5-fold CV on SM2miR1; 0.8938 LOOCV, 0.9899 5-fold CV on SM2miR2).
• The model successfully predicted experimentally validated SMiR associations among top candidates (9, 17, 37 for top 10, 20, 50).
• Identified potential associations, including enoxacin with mir-21 and decitabine with mir-155.

Conclusions:

• RWNS demonstrates superior performance in predicting SMiR associations compared to existing methods.
• The model offers a robust and efficient platform for identifying novel SMiR links.
• RWNS holds significant potential as a tool for accelerating drug discovery and therapeutic development.