MSIGR-PLA: Integrating Multi-Scale Interaction and Global Representations for Protein-Ligand Affinity Prediction

Area of Science:

• Computational chemistry
• Structural biology
• Drug discovery

Background:

• Accurate protein-ligand affinity (PLA) prediction is vital for drug discovery.
• Current methods struggle with local interaction features and global representations, limiting predictive accuracy.

Purpose of the Study:

• To develop an integrative framework, MSIGR-PLA, for enhanced PLA prediction.
• To improve accuracy by combining multi-scale local interaction features with global protein-ligand representations.

Main Methods:

• MSIGR-PLA utilizes two feature encoders: a local encoder with a multi-scale dynamic interaction (MSDI) module (GCN, cross-attention, Graph Transformer) and a global encoder (ESM-2 for protein, CNN-Transformer for ligand sequence, GIN for ligand structure).

Main Results:

• MSIGR-PLA outperformed existing methods on four benchmark datasets, showing 3.7%-9.0% improvement in Pearson correlation coefficient (R).
• Ablation studies confirmed the effectiveness of key modules, and a case study highlighted the MSDI module's ability to identify key binding residues.

Conclusions:

• MSIGR-PLA offers a significant advancement in PLA prediction accuracy.
• The framework's ability to integrate multi-scale local and global features provides a more robust approach for drug discovery applications.