Drug prioritization using the semantic properties of a knowledge graph

Area of Science:

• Biomedical Informatics
• Pharmacology
• Computational Biology

Background:

• Drug repurposing leverages existing drugs for new therapeutic indications.
• Biomedical knowledge is distributed across diverse literature and databases.
• Knowledge graphs integrate heterogeneous biological information to represent complex relationships.

Purpose of the Study:

• To develop a computational method for prioritizing drug repurposing candidates.
• To utilize semantic information from a comprehensive knowledge graph for feature extraction.
• To predict novel drug-disease associations for clinical development.

Main Methods:

• Extracted semantic features between drug and disease concepts from a knowledge graph integrating 200 sources.
• Trained a random forest classifier using the RepoDB database of drug-disease combinations.
• Evaluated classifier performance using 10-fold cross-validation, achieving a 92.2% AUC.

Main Results:

• Prioritized 21 preclinical drug repurposing candidates for Autosomal Dominant Polycystic Kidney Disease (ADPKD).
• Identified Mozavaptan, a vasopressin V2 receptor antagonist, as the top candidate for clinical approval.
• Mozavaptan belongs to the same drug class as tolvaptan, the sole approved ADPKD treatment.

Conclusions:

• Semantic properties within knowledge graphs are valuable for prioritizing drug repurposing.
• This approach can accelerate the identification of effective treatments for diseases like ADPKD.
• Computational methods enhance the efficiency of drug discovery and development pipelines.