Improving Docking Accuracy through Molecular Mechanics Generalized Born Optimization and Scoring.

Area of Science:

• Computational chemistry
• Structural biology
• Drug discovery

Background:

• Molecular docking is crucial in biopharmaceuticals for predicting binding modes and virtual screening.
• Current docking methods often fail to accurately reproduce known binding modes from cocrystal structures.
• Significant improvements in docking methodology are needed for reliable prediction of protein-ligand interactions.

Purpose of the Study:

• To develop and validate a hierarchical docking method for accurately reproducing known binding modes.
• To assess the performance of the hierarchical method against established industry-leading docking packages.
• To establish a new benchmark for accuracy in molecular docking using a standard dataset.

Main Methods:

• A hierarchical approach combining UCSF DOCK 4.0 for rigid docking of conformational ensembles.
• Optimization and scoring of docked poses using molecular mechanics and a standard MMGB energy function.
• Validation against a standard set of 79 Protein Data Bank (PDB) cocrystals.

Main Results:

• The hierarchical method achieved 90% accuracy, correctly identifying the binding mode in 71 out of 79 cases.
• Performance was comparable to industry-leading software like GOLD, Glide, and Surflex.
• Failures were observed in highly flexible, charged, and non-druglike molecules, indicating limitations for specific chemical spaces.

Conclusions:

• The developed hierarchical docking method demonstrates unprecedented accuracy on a highly benchmarked test set.
• The method utilizes physically based force fields without parameter fitting, enhancing its general applicability.
• This approach offers a significant advancement for reliable protein-ligand binding mode prediction in drug discovery.