A Literature-Based Knowledge Graph Embedding Method for Identifying Drug Repurposing Opportunities in Rare Diseases

Area of Science:

• Biomedical Informatics
• Drug Discovery
• Computational Biology

Background:

• Rare diseases affect millions, often with limited treatment options due to small market sizes and high R&D costs.
• Drug repurposing, using existing FDA-approved drugs for new indications, presents a viable alternative to traditional drug development.
• Systematic hypothesis generation requires integrating diverse biomedical information from pharmacology, genetics, and pathology.

Purpose of the Study:

• To develop a systematic method for generating drug repurposing hypotheses.
• To leverage a biomedical knowledge graph for comprehensive data integration.
• To validate the generated hypotheses using external data and network analysis.

Main Methods:

• Utilized the Global Network of Biomedical Relationships (GNBR), a knowledge graph of drugs, diseases, and genes.
• Applied a knowledge graph embedding method incorporating uncertainty in literature-derived relationships.
• Employed link prediction to identify potential drug repurposing candidates.

Main Results:

• Achieved high performance (AUROC = 0.89) on a gold-standard dataset of known drug indications.
• Generated novel drug repurposing hypotheses validated through external literature and protein interaction networks.
• Demonstrated the model's capability to provide explanations for its predictions.

Conclusions:

• The knowledge graph embedding approach effectively generates and validates drug repurposing hypotheses.
• This method accelerates the identification of potential treatments for rare diseases.
• The model's explainability enhances trust and facilitates clinical translation.