Cobdock: an accurate and practical machine learning-based consensus blind docking method.

Area of Science:

• Computational chemistry
• Structural biology
• Drug discovery

Background:

• Predicting protein-small molecule interactions is vital for drug discovery.
• Blind docking methods sample the entire protein surface but often lack accuracy due to large search spaces.
• Existing cavity detection-guided methods improve accuracy but depend heavily on the performance of a single detection tool.

Purpose of the Study:

• To develop a novel blind docking method that overcomes the limitations of single-tool dependency.
• To enhance both binding site identification and pose prediction accuracy in drug discovery.
• To improve the reliability and accuracy of blind docking through consensus-based machine learning.

Main Methods:

• Developed Consensus Blind Dock (CoBDock), a parallel docking approach.
• Integrated docking and cavity detection results using machine learning algorithms.
• Evaluated CoBDock on diverse datasets including PDBBind 2020, ADS, MTi, DUD-E, and CASF-2016.

Main Results:

• CoBDock demonstrated superior performance in binding site identification compared to state-of-the-art cavity detection tools.
• The method achieved higher binding mode prediction accuracy than existing blind docking approaches.
• Experimental results validated CoBDock's effectiveness across multiple benchmark datasets.

Conclusions:

• CoBDock offers a more accurate and reliable solution for blind docking in drug discovery.
• The consensus-based machine learning approach effectively integrates diverse computational predictions.
• This method advances the field by reducing reliance on single-tool performance for cavity detection.