DDI-AttendNet: cross attention with structured graph learning for inter-drug connectivity analysis

Area of Science:

• Computational drug discovery
• cheminformatics
• Bioinformatics

Background:

• Accurate characterization of inter-drug connectivity is crucial for drug discovery, synergistic effects, and polypharmacy.
• Traditional methods face limitations in scalability, interpretability, and capturing complex chemical interactions.

Purpose of the Study:

• Introduce DDI-AttendNet, a novel cross-attention architecture for drug-drug interaction (DDI) prediction.
• Address limitations of existing computational approaches in modeling complex drug relationships.

Main Methods:

• Utilize dual graph encoders for intra-drug atomic interactions and inter-drug relational graphs.
• Employ a cross-attention module to align and contextualize relevant substructures across drug pairs.
• Evaluate DDI-AttendNet on large-scale DDI benchmark datasets.

Main Results:

• DDI-AttendNet significantly outperforms state-of-the-art baselines, improving AUC and precision-recall metrics by 5%-10%.
• Attention weight visualization enhances model interpretability by linking predictions to chemically meaningful features.

Conclusions:

• DDI-AttendNet effectively models complex drug interaction structures.
• The model has the potential to accelerate safer and more efficient data-driven drug discovery pipelines.