R2-DDI: relation-aware feature refinement for drug-drug interaction prediction

Area of Science:

• Pharmacology
• Computational Biology
• Artificial Intelligence

Background:

• Drug-drug interactions (DDIs) pose significant risks in combination therapy.
• Machine learning and deep learning have advanced DDI prediction, but often overlook interaction types.
• Accurate DDI prediction is vital for patient safety and drug discovery.

Purpose of the Study:

• To develop a novel framework, R$^2$-DDI, for improved DDI prediction.
• To integrate relation-aware features into drug representation learning.
• To enhance the generalization ability of DDI prediction models.

Main Methods:

• Proposed the R$^2$-DDI framework incorporating a relation-aware feature refinement module.
• Integrated relation features into drug representations and refined them within the framework.
• Employed consistency training to regularize multi-branch predictions for better generalization.

Main Results:

• The R$^2$-DDI approach significantly improved DDI prediction performance across multiple datasets.
• The relation-aware feature refinement enhanced drug representation learning.
• The method demonstrated superior generalization ability compared to existing approaches.

Conclusions:

• The R$^2$-DDI framework effectively improves DDI prediction accuracy and generalization.
• Incorporating relation types into feature learning is crucial for robust DDI prediction models.
• This work offers a valuable tool for safer drug discovery and clinical practice.