MLGL-MP: a Multi-Label Graph Learning framework enhanced by pathway interdependence for Metabolic Pathway prediction.

Area of Science:

• Computational chemistry
• Bioinformatics
• Machine learning

Background:

• Accurate prediction of drug metabolism is vital for lead compound optimization in drug discovery.
• Existing machine learning methods for metabolic pathway prediction often overlook pathway interdependencies and lack interpretability.
• Understanding why compounds enter specific metabolic pathways is crucial for elucidating drug behavior.

Purpose of the Study:

• To develop a novel Multi-Label Graph Learning framework for Metabolic Pathway prediction (MLGL-MP) that accounts for pathway interdependence.
• To improve the accuracy and interpretability of predicting metabolic pathways for drug-like compounds.
• To provide insights into the structural features of compounds associated with specific metabolic pathways.

Main Methods:

• Utilized graph neural networks (GNNs) for a compound encoder to learn compound embeddings.
• Constructed a pathway dependence graph using word embeddings and pathway co-occurrences, feeding into a pathway encoder with graph convolutional networks (GCNs).
• Implemented a multi-label predictor that adapts compound embeddings to the pathway embedding space to predict pathway participation.

Main Results:

• MLGL-MP demonstrated superior performance compared to state-of-the-art methods on KEGG pathways.
• Ablation studies confirmed the significant contributions of pathway dependence, the embedding space adapter, and the pre-training strategy.
• A case study showcased the model's interpretability by identifying key compound substructures linked to metabolic pathways.

Conclusions:

• MLGL-MP effectively predicts metabolic pathways by incorporating pathway interdependence and offers interpretability.
• The framework has the potential to significantly advance metabolic pathway prediction in drug discovery pipelines.
• The developed model and associated data are publicly available to facilitate further research.