A foundation model for clinician-centered drug repurposing.

Area of Science:

• Artificial Intelligence in Medicine
• Computational Drug Discovery
• Graph Neural Networks

Background:

• Current AI models for drug repurposing are limited, focusing only on diseases with existing treatments.
• Drug repurposing offers a cost-effective approach to discovering new therapies.

Purpose of the Study:

• To introduce TXGNN, a graph foundation model for zero-shot drug repurposing.
• To identify potential drug candidates for diseases lacking treatment options.
• To improve the accuracy and interpretability of drug repurposing predictions.

Main Methods:

• TXGNN was trained on a comprehensive medical knowledge graph.
• It employs a graph neural network and metric-learning module for ranking drug indications and contraindications.
• An Explainer module provides multi-hop medical knowledge paths for rationale transparency.

Main Results:

• TXGNN demonstrated a 49.2% improvement in indication prediction accuracy and a 35.1% improvement in contraindication prediction accuracy.
• Predictions align with real-world off-label prescriptions in a large healthcare system.
• Human evaluation confirmed the model's accuracy and interpretability.

Conclusions:

• TXGNN advances zero-shot drug repurposing by enabling predictions for diseases with limited or no existing drugs.
• The model's interpretable explanations facilitate expert investigation and validation.
• TXGNN shows significant potential for accelerating drug discovery and expanding therapeutic options.