GCMDR

GCMDR predicts associations between microRNAs (miRNAs) and drug resistance using graph convolutional methods to model molecular interactions and inform mechanisms of resistance.

Key Features:

• Graph Convolution Technique: Employs graph convolution to construct a three-layer latent factor model that integrates high-dimensional attributes of miRNAs and drugs within an end-to-end learning framework.
• Bipartite Attributed Graph Formulation: Formulates miRNA–drug resistance association prediction as a link prediction task on a bipartite attributed graph modeling interactions among miRNAs, genes, and drugs.
• Latent Factor Model: Learns latent factors that capture intricate patterns relating miRNA regulation of genes and mechanisms of drug efficacy and resistance.
• Graph Embedding Features: Learns graph embedding features for both miRNAs and drugs to incorporate rich contextual information for prediction.
• Integration of Diverse Data Sources: Integrates miRNA expression profiles, drug substructure fingerprints, gene ontology, and disease ontology to enrich predictions with biological context.

Scientific Applications:

• Predicting Drug Resistance Mechanisms: Identifies associations between specific miRNAs and drug resistance to aid understanding of molecular resistance mechanisms.
• Enhancing miRNA Therapeutics: Highlights miRNA targets and associations relevant for developing miRNA-based therapeutic strategies to overcome drug resistance.

Methodology:

Formulates the task as link prediction on a bipartite graph with nodes representing miRNAs and drugs, applies graph convolutional networks to learn three-layer latent representations in an end-to-end framework, and validates performance using 2-, 5-, and 10-fold cross-validation.