Graph-based deep learning for drug-drug interaction prediction: a systematic review.

Area of Science:

• Computational chemistry
• Pharmacology
• Artificial intelligence in drug discovery

Background:

• Drug-drug interactions (DDIs) pose significant risks in drug development and patient safety.
• Computational methods are increasingly vital for predicting DDIs.
• Existing surveys lack a systematic organization based on graph-based deep learning paradigms.

Purpose of the Study:

• To provide a comprehensive review of graph-based deep learning methods for DDI prediction.
• To categorize existing DDI prediction models based on graph convolutional networks (GCNs), graph attention networks (GATs), and graph contrastive learning (GCL).
• To highlight the strengths and weaknesses of these approaches in handling molecular structures, heterogeneous interactions, and representation learning.

Main Methods:

• Systematic literature review focusing on graph-based deep learning techniques.
• Categorization of DDI prediction models according to GCN, GAT, and GCL paradigms.
• Analysis of modeling strategies, advantages, and limitations for each paradigm.

Main Results:

• Detailed review and categorization of graph-based DDI prediction methods.
• Identification of key challenges including multi-modal data integration and model interpretability.
• Discussion of the effectiveness of GCNs, GATs, and GCL in capturing complex molecular and interaction data.

Conclusions:

• Graph-based deep learning offers powerful tools for DDI prediction.
• Future research should focus on integrating diverse data sources and improving model interpretability for safer drug development.
• This review serves as a practical reference for developing advanced DDI prediction models.