KGiA: Drug repurposing through disease-aware knowledge graph augmentation.

Area of Science:

• Biomedical Informatics
• Computational Biology
• Drug Discovery

Background:

• Drug repurposing accelerates development by finding new uses for existing drugs.
• Biomedical knowledge graphs (KGs) aid drug repurposing but are limited by incompleteness.
• Existing KG methods struggle to generalize to new biomedical entities.

Purpose of the Study:

• To develop an inductive graph augmentation method (KGiA) for enhanced drug repurposing.
• To enable semi-inductive reasoning for generalizing to unseen biomedical entities.
• To improve the performance and generalizability of KG-based drug repurposing.

Main Methods:

• KGiA enhances knowledge graphs by incorporating counterfactual relationships derived from disease-specific topological patterns.
• The method was applied to a large-scale biomedical KG (1.6M triples, 100K entities, 30K diseases).
• Semi-inductive reasoning allows generalization to previously unseen biomedical entities.

Main Results:

• KGiA improved generalizability by up to 24x in Mean Reciprocal Rank (MRR) across five architectures.
• The proposed method outperformed state-of-the-art KG-based drug repurposing models by up to 32%.
• Case studies, including Alzheimer's Disease, demonstrated KGiA's potential for identifying novel repurposed drug candidates.

Conclusions:

• Augmenting knowledge graphs with counterfactual relationships improves KG-based drug repurposing.
• KGiA offers a promising approach to overcome KG incompleteness for drug discovery.
• The method enhances the ability to identify novel therapeutic applications for existing drugs.