DrugRep-KG: Toward Learning a Unified Latent Space for Drug Repurposing Using Knowledge Graphs

Area of Science:

• Computational pharmacology and bioinformatics.
• Drug discovery and development.
• Artificial intelligence in medicine.

Background:

• Drug repurposing (DR) identifies new uses for existing drugs, but current computational methods struggle with data representation and imbalanced datasets.
• Existing DR approaches often fail to integrate diverse features into a unified latent space, limiting prediction accuracy.
• The scarcity of known drug-disease associations compared to unknown ones creates significant negative data sampling challenges.

Purpose of the Study:

• To propose DrugRep-KG, a knowledge graph embedding approach to enhance drug repurposing prediction.
• To address data representation and negative data sampling issues in computational drug repurposing.
• To identify novel therapeutic applications for existing drugs, including potential treatments for COVID-19 and dermatological conditions.

Main Methods:

• Employed a knowledge graph embedding technique to represent drugs and diseases in a unified latent space.
• Developed a refined negative data sampling strategy by selecting unknown associations linked to adverse drug reactions.
• Evaluated DrugRep-KG's performance using AUC-ROC and AUC-PR metrics and tested its predictive power on specific diseases.

Main Results:

• Achieved superior performance with AUC-ROC of 90.83% and AUC-PR of 90.10%, surpassing previous drug repurposing methods.
• Successfully predicted known effective treatments for contact dermatitis and atopic eczema, including novel suggestions like fluorometholone for contact dermatitis.
• Identified potential drug candidates for COVID-19, aligning with existing databases and experimental evidence.

Conclusions:

• DrugRep-KG effectively addresses key challenges in computational drug repurposing through knowledge graph embeddings and improved negative sampling.
• The method demonstrates high accuracy in predicting drug-disease associations and discovering potential treatments for various conditions.
• DrugRep-KG offers a promising framework for accelerating drug discovery and repurposing, with potential applications in infectious and dermatological diseases.