Generating explainable hypotheses for drug repurposing with graph neural networks.

Area of Science:

• Computational Biology
• Bioinformatics
• Artificial Intelligence in Medicine

Background:

• Biomedical knowledge discovery requires computational tools for pattern recognition in complex datasets.
• Generating explainable, evidence-based hypotheses for biological interactions is a significant challenge.
• Drug repurposing necessitates effective methods for identifying novel drug-disease relationships.

Purpose of the Study:

• To introduce XAIPath, an interpretable pipeline for uncovering and explaining novel drug-disease relationships.
• To leverage biomedical knowledge graphs and Graph Neural Networks (GNNs) for hypothesis generation.
• To facilitate applications such as drug repurposing through explainable AI.

Main Methods:

• XAIPath combines Graph Neural Network (GNN) predictions with a post-hoc interpretability layer.
• It extracts and compares simple paths connecting drug and disease nodes in a biomedical knowledge graph using MinHash similarity.
• K-means clustering groups similar paths to form interpretable clusters representing mechanistic hypotheses.

Main Results:

• XAIPath achieved strong predictive performance on the NeDRex knowledge graph for drug indication prediction.
• AUROC exceeded 95%, AUPRC surpassed 90%, and precision, sensitivity, and specificity were all above 85%.
• High-scoring predictions were literature-supported, and path clusters aligned with existing mechanistic annotations.

Conclusions:

• XAIPath provides a scalable and explainable method for identifying drug-disease associations.
• The pipeline facilitates hypothesis generation and biological validation in biomedical research.
• Explainable AI is valuable for advancing drug repurposing and understanding biological mechanisms.