A fused deep learning approach to transform drug repositioning

Area of Science:

• Pharmacology
• Computational Biology
• Artificial Intelligence

Background:

• Drug repositioning accelerates therapeutic development and reduces costs by finding new uses for existing drugs.
• Current methods struggle with diverse data, cold start problems, and attribute representation.
• There is a need for advanced computational frameworks to enhance drug repositioning efficiency.

Purpose of the Study:

• To introduce a Unified Knowledge-Enhanced deep learning framework for Drug Repositioning (UKEDR).
• To address key challenges in drug repositioning, including cold start issues and data integration.
• To improve the accuracy and efficiency of identifying new therapeutic applications for existing drugs.

Main Methods:

• Developed UKEDR, integrating knowledge graph embedding, pre-training, and recommendation systems.
• Employed a semantic similarity-driven embedding approach to mitigate the cold start problem.
• Evaluated UKEDR against classical machine learning, network-based, and deep learning baselines.

Main Results:

• UKEDR outperformed various baseline methods in drug repositioning tasks.
• Demonstrated superior performance in cold start scenarios, handling unseen nodes and new compounds effectively.
• Showcased robustness on imbalanced datasets and strong generalization in drug- and disease-specific cold-start situations.

Conclusions:

• UKEDR offers a powerful and versatile framework for drug repositioning.
• The semantic similarity approach significantly improves performance in cold start scenarios.
• UKEDR shows strong potential for real-world applications in accelerating drug discovery.