ABT-DDI: A Graph Transformer Model with Atomic-Bond Structure Awareness for Drug-Drug Interaction Prediction

Area of Science:

• Computational chemistry
• Pharmacology
• Bioinformatics

Background:

• Polypharmacy increases drug-drug interaction (DDI) risks.
• Current DDI prediction methods often neglect molecular 3D structures and atomic bonding relationships.

Purpose of the Study:

• To develop an advanced DDI prediction model, ABT-DDI.
• To enhance the accuracy of predicting drug-drug interaction events (DDIEs) by incorporating multimodal molecular information.

Main Methods:

• Utilized a graph transformer architecture for DDI prediction.
• Modeled spatial relationships (atom-atom, atom-bond, bond-bond) using a multiscale attention mechanism.
• Integrated molecular fingerprints with 3D spatial descriptors for comprehensive molecular representation.

Main Results:

• ABT-DDI demonstrated superior performance over existing state-of-the-art methods on benchmark datasets.
• The model effectively captured atomic and bonding interaction patterns.
• Achieved significant improvements across multiple performance metrics.

Conclusions:

• ABT-DDI offers a robust approach for predicting DDIs by leveraging detailed molecular structural and spatial information.
• The model has significant potential applications in drug development and polypharmacy risk management.