DeepGCL: Multi-View Graph Contrastive Learning for Enhanced Drug-Target Binding Affinity Prediction Through Protein...

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Drug-target binding affinity (DTA) prediction is crucial for drug discovery.
• Existing methods struggle to integrate local atomic interactions and global sequence contexts.
• Capturing cross-modal dependencies between protein pockets and drugs remains a challenge.

Purpose of the Study:

• To develop a novel multi-modal framework, DeepGCL, for accurate DTA prediction.
• To effectively model latent representations of pocket-drug interactions and molecular determinants.
• To address limitations in jointly modeling diverse molecular information.

Main Methods:

• DeepGCL integrates protein sequences, pocket-drug interaction graphs, and drug molecular graphs.
• It employs multi-view graph contrastive learning with augmented view generation.
• Sequence encoding is combined with graph contrastive learning for complementary representations.

Main Results:

• DeepGCL consistently outperforms state-of-the-art methods on multiple benchmarks.
• Ablation studies confirm significant contributions from protein feature and graph contrastive learning modules.
• The framework accurately characterizes spatial and chemical relationships between binding partners.

Conclusions:

• Multi-view learning effectively captures the multifaceted nature of drug-target interactions.
• Integrating diverse molecular representations is vital for accurate DTA prediction.
• DeepGCL offers a promising approach for advancing drug discovery pipelines.