Polypharmacy side-effect prediction with enhanced interpretability based on graph feature attention network.

Area of Science:

• Pharmacology and Toxicology
• Computational Biology
• Drug Discovery

Background:

• Polypharmacy, the concurrent use of multiple drugs, presents significant challenges in new drug development due to complex drug-drug interactions and their associated side effects.
• Existing Graph Neural Network (GNN) models show promise in predicting these interactions but often lack interpretability for domain experts.

Purpose of the Study:

• To develop an interpretable prediction model for polypharmacy side effects.
• To address the limitations of current GNN models in providing understandable predictions for biomedical and pharmaceutical professionals.

Main Methods:

• Introduction of a novel Graph Feature Attention Network (GFAN) model designed for interpretable polypharmacy side effect prediction.
• Formulation of a node classification problem using line graph concepts to simulate polypharmacy scenarios.
• Emphasis on differential target gene highlighting within the GFAN model.

Main Results:

• Experimental validation on benchmark datasets confirmed the interpretability of the GFAN model.
• GFAN demonstrated competitive predictive performance compared to existing graph attention networks.
• Case studies showed GFAN's sensitivity in identifying crucial target genes for specific polypharmacy side effect predictions.

Conclusions:

• The GFAN model offers a significant advancement in the interpretable prediction of polypharmacy side effects.
• This approach enhances understanding of complex drug-drug interactions, supporting safer and more effective drug development.
• GFAN's ability to identify key genes provides valuable insights for pharmaceutical research.