Biomedical knowledge graph learning for drug repurposing by extending guilt-by-association to multiple layers

Area of Science:

• Computational biology
• Bioinformatics
• Pharmacology

Background:

• Computational drug repurposing uses biomedical knowledge graphs to find new drug indications.
• Learning on these graphs is hindered by gene dominance and sparse drug/disease entities, leading to poor representations.

Purpose of the Study:

• To develop an effective computational drug repurposing method by addressing limitations in biomedical knowledge graph learning.
• To improve the accuracy of predicting drug-disease associations.

Main Methods:

• Proposed a "semantic multi-layer guilt-by-association" approach.
• Developed DREAMwalk, a model employing semantic information-guided random walks to generate node sequences.
• Generated unified embeddings for drugs and diseases.

Main Results:

• Achieved up to 16.8% improvement in drug-disease association prediction accuracy compared to state-of-the-art models.
• Demonstrated a harmonious alignment between biological and semantic contexts within the embedding space.
• Successfully applied the approach to drug repurposing case studies for breast carcinoma and Alzheimer's disease.

Conclusions:

• The multi-layer guilt-by-association perspective offers a powerful strategy for drug repurposing on biomedical knowledge graphs.
• DREAMwalk effectively maps drugs and diseases, improving prediction accuracy and revealing biological insights.