Molecular-Driven Multi-View Hypergraph Contrastive Learning for Drug-Drug Interaction Prediction.

Area of Science:

• Computational chemistry
• Bioinformatics
• Pharmacology

Background:

• Drug combinations can cause adverse reactions, necessitating accurate drug-drug interaction (DDI) prediction.
• Existing DDI prediction methods often focus on superficial molecular features, neglecting crucial internal structural information.

Purpose of the Study:

• To propose Mol-HCL, a multi-view hypergraph contrastive learning framework for enhanced DDI prediction.
• To leverage internal molecular structure information for more accurate DDI forecasting.

Main Methods:

• Developed a multi-view hypergraph contrastive learning framework (Mol-HCL) with molecular, structural, and semantic views.
• Incorporated hypernodes and hyperchains to capture complex intra- and inter-molecular relationships.
• Utilized contrastive learning between structural/semantic hypergraphs and the molecular view to refine drug representations.

Main Results:

• Mol-HCL demonstrated significant improvements over existing methods in DDI prediction tasks.
• The framework effectively captures both intra-molecular and inter-molecular information for DDI analysis.
• Experimental validation on two real-world datasets confirmed the efficacy of the proposed approach.

Conclusions:

• Mol-HCL offers a powerful new approach for DDI prediction by analyzing internal molecular structures.
• The multi-view hypergraph contrastive learning strategy enhances the accuracy and robustness of DDI prediction.
• This framework provides valuable insights into potential drug combination risks.