A case-based explainable graph neural network framework for mechanistic drug repositioning.

Area of Science:

• Computational Biology
• Pharmacology
• Artificial Intelligence

Background:

• Drug repositioning accelerates drug discovery by repurposing existing medications for new therapeutic uses.
• Graph Neural Networks (GNNs) show potential for predicting drug-disease associations but often lack interpretability.
• Explainability is crucial for validating predictions and understanding the biological mechanisms of drug action.

Purpose of the Study:

• To introduce Drug-Based Reasoning Explainer (DBR-X), an explainable GNN model for drug repositioning.
• To enhance the interpretability and faithfulness of GNN-based predictions for drug-disease associations.
• To provide multi-hop biological explanations for identified associations to aid clinical translation.

Main Methods:

• Developed DBR-X, an explainable GNN model integrating link prediction and path-identification modules.
• Benchmarked DBR-X against existing GNN link prediction frameworks for accuracy in identifying drug-disease associations.
• Assessed the biological quality of explanations using curated mechanisms, faithfulness studies (deletion/insertion), and stability analysis.

Main Results:

• DBR-X demonstrated superior performance in predicting known drug-disease associations compared to other GNN frameworks.
• Achieved higher accuracy across all evaluation metrics in identifying drug-disease links.
• Biological explanations generated by DBR-X were validated through multiple rigorous assessment approaches.

Conclusions:

• DBR-X advances the state-of-the-art in GNN-based drug repositioning by providing accurate and interpretable predictions.
• The model's ability to generate multi-hop explanations can accelerate the clinical application of computational drug discovery.
• DBR-X offers a valuable tool for understanding drug mechanisms and facilitating the development of new therapies.