Task-driven knowledge graph filtering improves prioritizing drugs for repurposing.

Area of Science:

• Bioinformatics
• Computational Biology
• Pharmacology

Background:

• Drug repurposing is crucial for cost-effective drug discovery.
• Biomedical knowledge graphs integrate diverse biological data for predicting new drug-disease connections.
• Existing knowledge graph embedding models may underfit specific tasks like drug repurposing due to optimizing over all relation types.

Purpose of the Study:

• To develop a method for filtering biomedical knowledge graphs to improve drug repurposing.
• To enhance the performance and computational efficiency of drug repurposing prediction models.
• To leverage domain knowledge in the form of metapaths for task-specific knowledge graph refinement.

Main Methods:

• Utilized metapaths to filter two large biomedical knowledge graphs: Hetionet and DRKG.
• Applied a task-driven filtering approach to remove irrelevant information.
• Evaluated the impact of filtering on drug repurposing prediction performance and computational efficiency.

Main Results:

• The proposed method significantly reduced the number of entities in the knowledge graphs (up to 60% on Hetionet, 26% on DRKG).
• Achieved substantial improvements in prediction performance (up to 40.8% on Hetionet, 14.2% on DRKG).
• Demonstrated improved prioritization of antiviral compounds for SARS-CoV-2 following filtering.

Conclusions:

• Biomedical knowledge graphs contain information detrimental to specific tasks like drug repurposing.
• Task-driven filtering effectively removes counterproductive facts, enhancing prediction performance.
• The filtering process leads to a more efficient learning process for drug repurposing models.