SMMA-HNRL

SMMA-HNRL predicts potential associations between small molecules and microRNAs (miRNAs) by learning representations from a heterogeneous information network that integrates small molecules, miRNAs, and diseases to support therapeutic target discovery.

Key Features:

• Heterogeneous Network Construction: Constructs a heterogeneous information network with nodes representing small molecules (SMs), miRNAs, and diseases.
• Representation Learning: Applies HeGAN and HIN2Vec heterogeneous network representation learning algorithms to derive feature vectors for SM and miRNA nodes.
• Feature Integration and Classification: Merges feature vectors from HeGAN and HIN2Vec using a connect operation and classifies associations with LightGBM.
• Validation and Performance: Evaluates model performance using 10-fold cross-validation (AUC 0.9875) and independent validation datasets.

Scientific Applications:

• Therapeutic target prediction: Predicts miRNAs that could serve as targets for small molecule drugs, aiding investigation of interventions in complex diseases.
• Case studies and literature validation: Demonstrated predictions for 5-FU, cisplatin, and imatinib with literature validation counts of 35 out of the top 50 for 5-FU, 37 for cisplatin, and 22 for imatinib.

Methodology:

Constructs a heterogeneous network integrating small molecules, miRNAs, and diseases; applies HeGAN and HIN2Vec to obtain node embeddings; merges embeddings via a connect operation; uses LightGBM for classification; and evaluates using 10-fold cross-validation and independent validation datasets.