Prediction of Potential Small Molecule-Associated MicroRNAs Using Graphlet Interaction.

Area of Science:

• Biomedical Informatics
• Computational Biology
• Pharmacogenomics

Background:

• MicroRNAs (miRNAs) are increasingly recognized as therapeutic targets for various human diseases.
• Small molecules are emerging as potential agents to modulate miRNA activity, highlighting the need to understand their interactions.
• The relationship between small molecules and miRNAs is a developing field with significant therapeutic implications.

Purpose of the Study:

• To develop and evaluate a computational model, GISMMA, for predicting associations between small molecules and miRNAs.
• To leverage network-based approaches and graphlet interactions for enhanced prediction accuracy.
• To provide a tool for identifying potential small molecule-miRNA therapeutic strategies.

Main Methods:

• Constructed integrated networks incorporating small molecule similarity, miRNA similarity, and known associations.
• Utilized graphlet interaction patterns (28 isomers) to characterize complex relationships between molecules and miRNAs.
• Calculated association scores based on graphlet interaction counts within the integrated networks.
• Performed rigorous validation using global and local leave-one-out cross-validation (LOOCV) and five-fold cross-validation on two independent datasets.

Main Results:

• GISMMA demonstrated high predictive performance across multiple validation strategies, achieving AUCs up to 0.9505 (Dataset 1) and 0.8640 (Dataset 2).
• Case studies on known drugs (5-Fluorouracil, 17β-Estradiol, 5-Aza-2'-deoxycytidine) showed GISMMA accurately predicted a significant number of experimentally validated miRNA associations (25-30 out of top 50).
• The model's performance was consistently excellent, as evidenced by cross-validation results and successful validation in case studies.

Conclusions:

• GISMMA is a powerful and accurate computational tool for predicting small molecule-miRNA associations.
• The graphlet-based approach effectively captures complex biological relationships, offering a novel method for drug-target interaction prediction.
• This model holds promise for accelerating the discovery of novel miRNA-targeted therapies and understanding drug mechanisms.