Integrated Knowledge Graph and Drug Molecular Graph Fusion via Adversarial Networks for Drug-Drug Interaction...

Area of Science:

• Computational pharmacology
• Bioinformatics
• Drug discovery

Background:

• Co-administering drugs can improve treatment efficacy but increases the risk of adverse drug interactions (DDIs).
• Existing DDI prediction methods struggle to integrate knowledge graph biomedical information with drug properties effectively.

Purpose of the Study:

• To develop a novel end-to-end framework for predicting drug-drug interactions (DDIs).
• To enhance the accuracy and robustness of DDI prediction by effectively integrating molecular structure and knowledge graph information.

Main Methods:

• Utilized a message-passing neural network for molecular structure information.
• Employed a knowledge-aware graph attention network for knowledge graph drug representation.
• Implemented dual generative adversarial networks for adversarial training to capture interrelations between molecular and semantic information.

Main Results:

• Outperformed state-of-the-art algorithms in binary classification tasks, showing improvements in accuracy, AUC, AUPR, and F1 score.
• Achieved significant improvements in multiclass classification metrics, including accuracy, macro precision, macro recall, and macro F1.
• Ablation studies confirmed the method's effectiveness and robustness in DDI prediction.

Conclusions:

• The proposed generative adversarial network-based framework effectively integrates diverse drug information for accurate DDI prediction.
• This approach offers a robust solution for identifying potential drug-drug interactions, advancing computational pharmacology and patient safety.