Binding mode information improves fragment docking.

Area of Science:

• Computational chemistry
• Structural biology
• Drug discovery

Background:

• Molecular docking is crucial for predicting ligand-protein interactions in drug discovery.
• Current docking programs struggle to accurately rank correct binding poses, especially for small molecules like fragments.
• Fragments, due to weak interactions, are particularly sensitive to scoring inaccuracies.

Purpose of the Study:

• To evaluate the effectiveness of binding mode information for fragment pose prediction.
• To compare the performance of interaction fingerprints, 3D-matching of interaction patterns, and 3D-matching of shapes.
• To identify optimal strategies for improving fragment docking accuracy.

Main Methods:

• Utilized a test set of 586 high-quality fragment/protein complexes from the Protein Data Bank.
• Generated and evaluated docking poses using three distinct post-processing approaches.
• Compared the accuracy of these methods against native scoring functions.

Main Results:

• The best binding mode-based approach demonstrated twice the accuracy of native scoring functions.
• Post-processing methods showed diminishing returns for smaller fragment molecules.
• Both fragment and drug-like molecules served as effective references for pose validation.

Conclusions:

• Binding mode information significantly enhances fragment pose prediction accuracy in molecular docking.
• Interaction pattern matching and shape matching are promising strategies for fragment-based drug discovery.
• Reference structures, including both fragments and known drugs, are valuable for validating docking predictions.