Knowledge Graphs for Indication Expansion: An Explainable Target-Disease Prediction Method.

Area of Science:

• Computational drug discovery
• Bioinformatics
• Pharmacology

Background:

• Indication expansion accelerates drug development by identifying new uses for existing targets.
• Semantic knowledge graphs (KGs) are increasingly used for target repositioning in computational drug discovery.
• Existing KGs often lack comprehensive data integration, limiting their effectiveness.

Purpose of the Study:

• To assess the suitability of a novel KG for explainable target-disease link prediction in indication expansion.
• To evaluate the predictive performance of shortest path and embedding-based methods using the KG.
• To investigate the impact of tissue expression data on prediction accuracy.

Main Methods:

• Constructed a KG integrating human-curated literature and gene expression data.
• Applied shortest path (with tissue information) and embedding-based prediction methods to the KG.
• Evaluated prediction accuracy using post-2010 gene-disease links and compared against random baselines.
• Assessed the contribution of tissue expression data by removing it from the KG.

Main Results:

• Shortest path methods significantly outperformed random baselines.
• Embedding-based methods achieved higher accuracy than shortest path predictions.
• Excluding tissue expression data substantially degraded prediction quality, particularly for embedding methods.
• The glass-box approach demonstrated interpretability for candidate target-disease predictions.

Conclusions:

• The developed KG is effective for explainable target-disease link prediction in indication expansion.
• Tissue expression data is a critical component for improving prediction accuracy.
• The KG's glass-box nature enhances the interpretability of drug repositioning predictions.