Surflex-Dock: Docking benchmarks and real-world application.

Area of Science:

• Computational chemistry
• Drug discovery
• Structural biology

Background:

• Traditional molecular docking benchmarks focus on pose prediction accuracy using cognate ligands.
• Virtual screening performance is typically measured using large, diverse datasets of protein targets and ligands.
• Existing benchmarks may not fully capture the complexities of real-world drug discovery scenarios.

Purpose of the Study:

• To evaluate molecular docking performance on established benchmarks for both pose prediction and virtual screening.
• To investigate the impact of different docking protocols and data preparation methods on performance.
• To identify strategies for improving the accuracy and efficiency of molecular docking in drug discovery.

Main Methods:

• Pose prediction was assessed using the Astex Diverse set (85 protein-ligand complexes).
• Virtual screening performance was evaluated on the DUD set (40 protein targets).
• Cross-docking protocols, multiple screening methods (docking, 2D/3D similarity), and multiple protein conformations were tested.

Main Results:

• Re-prepared benchmark datasets yielded results consistent with previous Surflex-Dock reports.
• Minor protein coordinate changes significantly impacted pose prediction accuracy, highlighting limitations of cognate ligand re-docking.
• Cross-docking protocols substantially improved pose prediction performance.
• Combining multiple screening methods and using multiple protein conformations significantly enhanced virtual screening enrichment.

Conclusions:

• Cognate ligand re-docking is limited for assessing pose prediction accuracy due to sensitivity to minor structural variations.
• Cross-docking protocols offer improved pose prediction by accounting for protein flexibility.
• Ensemble-based virtual screening approaches, combining diverse methods and protein conformations, are crucial for robust drug discovery pipelines.