MLGL-MP

MLGL-MP predicts metabolic pathways for drug-like compounds to support lead compound optimization in drug discovery.

Key Features:

• Multi-Label Graph Learning Framework: Predicts potential metabolic pathways for drug-like compounds using a multi-label graph learning framework.
• Pathway Interdependence: Integrates pathway interdependence by modeling interactions among metabolic pathways rather than treating them independently.
• Compound Encoder: Learns compound embeddings using graph neural networks.
• Pathway Encoder: Constructs a pathway dependence graph using re-trained word embeddings and pathway co-occurrences and learns pathway embeddings with graph convolutional networks.
• Multi-Label Predictor: Maps the compound embedding space into the pathway embedding space and measures proximity between spaces to determine pathway participation.
• Interpretability: Identifies compound substructures significantly associated with specific metabolic pathways to support interpretation of predictions.

Scientific Applications:

• Lead optimization in drug discovery: Predicts compound metabolism during lead optimization by identifying likely metabolic pathways.
• Metabolic pathway identification: Assigns metabolic pathways to drug-like compounds based on learned compound and pathway embeddings.
• Mechanistic insight: Provides interpretable associations between compound substructures and pathway participation to inform metabolic mechanism hypotheses.

Methodology:

Implements a multi-label graph learning framework with a compound encoder using graph neural networks, a pathway encoder that builds a pathway dependence graph from re-trained word embeddings and pathway co-occurrences and learns pathway embeddings via graph convolutional networks, and a multi-label predictor that maps compound embeddings into the pathway embedding space and uses proximity to assign pathway participation; interpretability is provided by identifying substructures associated with pathways.