Drug Repurposing using consilience of Knowledge Graph Completion methods

Area of Science:

• Biomedical informatics
• Computational pharmacology
• Drug discovery

Background:

• Link prediction in knowledge graphs is used for compound-disease associations.
• Existing methods, including knowledge graph embedding (KGE), lack sufficient evidence for drug-disease indications.
• Pathwalking algorithms improve confidence but have limitations in evidence and handling rare diseases.

Purpose of the Study:

• To evaluate seven link prediction methods for drug repurposing on a large biomedical knowledge graph.
• To combine path-based reasoning with KGE methods for more robust predictions.
• To identify potential drug repurposing indications with enhanced evidential support.

Main Methods:

• Evaluation of seven link prediction algorithms.
• Integration of path-based reasoning and KGE methods.
• Application to a comprehensive biomedical knowledge graph for drug repurposing.

Main Results:

• A novel approach combining path-based and KGE methods was developed.
• The integrated method provides stronger evidence for drug-disease indications.
• Potential drug repurposing candidates were identified.

Conclusions:

• Combining path-based and KGE methods enhances the reliability of drug repurposing predictions.
• The approach addresses limitations of individual methods, including evidence quality and rare disease prediction.
• The study demonstrates utility through a specific drug repurposing example.