SMALF

SMALF predicts unknown associations between microRNAs (miRNAs) and diseases to prioritize candidate miRNA–disease links for studying molecular disease mechanisms.

Key Features:

• Feature Learning with Stacked Autoencoder: Uses a stacked autoencoder to extract latent features for miRNAs and diseases from the original miRNA-disease association matrix.
• Integration of Similarity Measures: Integrates miRNA functional similarity and disease semantic similarity with the learned latent features to construct comprehensive feature vectors.
• Prediction Using XGBoost: Employs XGBoost gradient boosting to predict miRNA–disease associations from the integrated feature vectors.
• Model Validation and Performance: Assesses predictive performance via cross-validation and reports Area Under the Curve (AUC) metrics.
• Case-study Verification: Validates top predicted miRNAs for hepatocellular carcinoma, colon cancer, and breast cancer against MNDR v3.0 and miRCancer.

Scientific Applications:

• miRNA–disease association prediction: Prioritizes unknown miRNA–disease associations for downstream experimental validation.
• Disease-specific candidate discovery: Identifies candidate miRNAs associated with hepatocellular carcinoma, colon cancer, and breast cancer.
• Mechanistic inference and target prioritization: Supports inference of molecular disease mechanisms and prioritization of potential therapeutic miRNA targets.

Methodology:

Extracts latent features via a stacked autoencoder from the original miRNA-disease association matrix; integrates miRNA functional similarity and disease semantic similarity with latent features to form feature vectors; predicts associations using XGBoost; evaluates performance with cross-validation and AUC; verifies top predictions against MNDR v3.0 and miRCancer.