An effective framework for predicting drug-drug interactions based on molecular substructures and knowledge graph...

Area of Science:

• Pharmacology
• Computational Chemistry
• Artificial Intelligence

Background:

• Drug-drug interactions (DDIs) are critical in drug research, influencing efficacy and causing adverse reactions.
• Current AI models for DDI prediction often neglect drug molecular structures or inter-drug relationships.
• Identifying potential DDIs is crucial for physicians, patients, and researchers.

Purpose of the Study:

• To develop an advanced AI framework for predicting drug-drug interactions (DDIs).
• To capture multimodal drug characteristics by integrating chemical structure and knowledge graph information.
• To improve the accuracy and scope of DDI prediction models.

Main Methods:

• Proposed MSKG-DDI, a two-component framework integrating drug chemical structure graphs and drug knowledge graphs.
• Utilized a multimodal fusion neural layer to combine drug representations.
• Evaluated performance on binary-class, multi-class, and multi-label DDI prediction tasks.

Main Results:

• MSKG-DDI significantly outperformed state-of-the-art models on real-world datasets.
• The framework demonstrated superior performance in both transductive and inductive settings.
• Ablation analysis confirmed the practical utility and effectiveness of the proposed components.

Conclusions:

• The MSKG-DDI framework offers a novel and effective approach to DDI prediction.
• Integrating multimodal drug information, including molecular structure and knowledge graphs, enhances prediction accuracy.
• This study highlights the importance of comprehensive feature representation in computational drug research.