MSHGCL: Multi-Scale Hierarchical Graph Contrastive Learning for Drug-Drug Interactions.

Area of Science:

• Pharmacology and Cheminformatics
• Computational Drug Discovery
• Artificial Intelligence in Medicine

Background:

• Drug-drug interactions (DDIs) significantly impact drug efficacy and safety, necessitating accurate prediction methods.
• Current graph neural network-based DDI prediction methods effectively extract drug substructures but overlook intra-drug substructure relationships.
• Understanding these intra-drug relationships is crucial for refining DDI prediction models.

Purpose of the Study:

• To propose MSHGCL, a novel multi-scale hierarchical graph contrastive learning framework for enhanced DDI prediction.
• To address the limitation of existing methods by incorporating the relationship between drug substructures from the same drug.
• To improve the accuracy and reliability of predicting potential adverse drug events.

Main Methods:

• Developed MSHGCL, a method utilizing multi-scale hierarchical graph contrastive learning.
• Implemented an intra-layer contrastive learning module to constrain same-scale drug substructure relationships.
• Incorporated an inter-layer contrastive learning module to constrain adjacent-layer drug substructure relationships.

Main Results:

• MSHGCL was evaluated on two real-world datasets.
• Experimental results demonstrated that MSHGCL outperforms state-of-the-art DDI prediction methods.
• The proposed method shows significant improvements in predicting drug-drug interactions.

Conclusions:

• The MSHGCL method effectively models relationships between drug substructures within the same drug.
• This approach enhances the performance of drug-drug interaction prediction.
• MSHGCL represents a significant advancement in computational approaches to DDI prediction.