Predicting Drug-drug Interaction with Graph Mutual Interaction Attention Mechanism

Area of Science:

• Computational chemistry
• Artificial intelligence in drug discovery
• Bioinformatics

Background:

• Effective molecular representation is vital for AI-driven drug design and drug-drug interaction (DDI) prediction.
• Existing methods often overlook interaction information between molecular substructures and bond influences, leading to suboptimal drug representations.
• Key molecular substructures significantly impact DDI prediction outcomes.

Purpose of the Study:

• To propose a novel Graph learning framework of Mutual Interaction Attention mechanism (GMIA) for enhanced DDI prediction.
• To improve drug molecule representation by incorporating inter-molecular substructure interactions and bond information.
• To provide interpretability for DDI prediction by analyzing the significance of molecular substructures.

Main Methods:

• Developed a node-edge message communication encoder to aggregate atom node and incoming edge information for robust atom node representation.
• Designed a mutual interaction attention decoder to capture contextual interactions between molecular graphs of drug pairs.
• Implemented a co-attention matrix to analyze substructure significance and enhance model interpretability.

Main Results:

• GMIA achieved state-of-the-art performance on DDI prediction tasks, outperforming existing methods.
• The framework demonstrated superior results in terms of area under the precision-recall-curve (AUPR), area under the ROC curve (AUC), and F1 score across two datasets.
• Case studies confirmed GMIA's ability to identify key substructures contributing to potential DDIs.

Conclusions:

• The proposed GMIA framework offers an effective approach for DDI prediction by enhancing molecular representation through mutual interaction attention.
• GMIA provides significant improvements in prediction accuracy and offers valuable interpretability regarding key substructures.
• This work advances AI-driven drug discovery by providing a more comprehensive and interpretable model for predicting drug-drug interactions.