Graph former-CL: A novel graph transformer with contrastive learning framework for enhanced drug-drug interaction...

Area of Science:

• Computational chemistry
• Artificial intelligence in medicine
• Pharmacology

Background:

• Drug-drug interactions (DDIs) cause over 125,000 deaths annually in the US.
• Existing computational methods struggle with complex molecular data and novel drug combinations.
• Traditional Graph Neural Networks (GNNs) and sequence-based models have limitations in capturing molecular structure and dependencies.

Purpose of the Study:

• To develop an advanced deep learning framework for accurate drug-drug interaction (DDI) prediction.
• To overcome limitations of current methods in capturing long-range molecular dependencies and generalizing to new drug pairs.
• To enhance computational approaches for predicting adverse drug events in clinical settings.

Main Methods:

• Proposed Graph Former-CL, integrating Graph Transformer architecture with contrastive learning.
• Utilized a hierarchical Graph Transformer with position-aware multi-head self-attention for molecular pattern recognition.
• Implemented a domain-specific contrastive learning module with molecular augmentation and a cross-modal fusion mechanism (SMILES and graph representations).
• Employed an adaptive pooling strategy for multi-scale molecular representation.

Main Results:

• Graph Former-CL achieved 98.2% accuracy on DrugBank and 89.4% on TWOSIDES, outperforming state-of-the-art methods (p < 0.001).
• Demonstrated strong generalization with 85.6% accuracy for novel drugs in inductive settings.
• Significantly improved prediction of complex drug-drug interactions.

Conclusions:

• Graph Former-CL offers a superior deep learning approach for DDI prediction.
• The framework's ability to generalize to novel drugs is crucial for clinical application and patient safety.
• This advancement holds significant potential for reducing DDI-related mortality and morbidity.