A structure-guided approach for protein pocket modeling and affinity prediction

Area of Science:

• Computational chemistry
• Structural biology
• Drug discovery

Background:

• Binding affinity prediction commonly relies on molecular structure and activity data alone.
• Existing computational models may lack robustness when dealing with diverse ligand sets.

Purpose of the Study:

• To develop a hybrid structure-guided strategy for enhanced binding affinity prediction.
• To integrate protein structure information into computational models for improved structure-activity relationship (SAR) analysis.

Main Methods:

• A hybrid approach combining molecular similarity, docking, and multiple-instance learning.
• Development of the Surflex-QMOD method incorporating protein structural information.
• Construction of interpretable physical models of binding sites.

Main Results:

• Structure-guided models accurately predict binding affinities across a wide range of compounds.
• Improved representations of protein pockets and ligand binding modes were achieved.
• Significant performance gains in affinity and pose prediction, particularly for novel ligands.

Conclusions:

• Integrating protein structure information enhances the accuracy and robustness of binding affinity prediction models.
• The structure-guided Surflex-QMOD approach offers a powerful tool for drug discovery and development.
• This method provides more reliable insights into protein-ligand interactions.