Predicting drug-target binding affinity with cross-scale graph contrastive learning.

Area of Science:

• Computational chemistry
• Bioinformatics
• Machine learning

Background:

• Accurate drug-target binding affinity prediction is crucial for drug discovery and repurposing.
• Existing computational methods often analyze molecular structures or network interactions separately, limiting comprehensive feature capture.
• Integrating molecular-scale and network-scale information can enhance prediction quality.

Purpose of the Study:

• To introduce CSCo-DTA, a novel cross-scale graph contrastive learning approach for drug-target binding affinity prediction.
• To effectively combine molecular and network features for robust drug-target interaction representation.
• To improve the accuracy and reliability of computational drug discovery models.

Main Methods:

• Developed a novel cross-scale graph contrastive learning framework (CSCo-DTA).
• Integrated molecular structure features with drug-target bipartite network information.
• Employed contrastive learning to capture both local and global perspectives from multi-scale data.

Main Results:

• CSCo-DTA significantly outperformed existing state-of-the-art methods on two benchmark datasets.
• Ablation studies confirmed the effectiveness of multi-scale features and cross-scale contrastive learning.
• Successfully predicted novel potential targets for Erlotinib, validated by molecular docking analysis.

Conclusions:

• CSCo-DTA provides a powerful new approach for drug-target binding affinity prediction by integrating diverse data scales.
• The model's ability to leverage both molecular and network information enhances prediction performance.
• This method holds promise for accelerating drug discovery and repurposing efforts.