Multi-type feature fusion based on graph neural network for drug-drug interaction prediction

Area of Science:

• Pharmacology and Cheminformatics
• Artificial Intelligence in Drug Discovery

Background:

• Drug-drug interactions (DDIs) pose significant challenges in drug research and combination therapy.
• Accurate DDI prediction is crucial for patient safety and therapeutic efficacy.
• Existing deep learning methods for DDI prediction often lack robustness and scalability due to single-information utilization.

Purpose of the Study:

• To develop a robust and scalable model for predicting drug-drug interactions (DDIs).
• To effectively integrate multi-type drug features for improved DDI prediction accuracy.

Main Methods:

• Proposed a multi-type feature fusion based on graph neural network model (MFFGNN).
• Developed a novel feature extraction module to capture global molecular graph topology and local atomic features.
• Employed a gating mechanism within graph convolution layers to mitigate over-smoothing issues.

Main Results:

• MFFGNN effectively fuses topological, interaction, and chemical context information from molecular graphs, SMILES sequences, and drug-drug interaction networks.
• Extensive experiments demonstrated that MFFGNN outperforms state-of-the-art models in DDI prediction.
• Cross-dataset experiments confirmed the model's excellent generalization performance.

Conclusions:

• The MFFGNN model efficiently integrates diverse drug information sources for accurate DDI prediction.
• This multi-type feature fusion approach holds potential for discovering novel drug-drug interactions.
• The findings contribute to safer and more effective drug combination therapies.