Teaching old docks new tricks with machine learning enhanced ensemble docking.

Area of Science:

• Computational chemistry
• Drug discovery
• Machine learning

Background:

• Virtual screening (VS) predicts protein-small molecule binding.
• Ensemble VS accounts for protein flexibility by using multiple protein conformations.
• Ranking compounds in ensemble VS requires system-specific decisions for score interpretation.

Purpose of the Study:

• Introduce Ensemble Optimizer (EnOpt), a machine learning tool.
• Improve accuracy and interpretability of ensemble VS.
• Provide a broadly adoptable, open-source solution.

Main Methods:

• Ensemble Optimizer (EnOpt) utilizes machine learning algorithms.
• EnOpt optimizes the selection of protein conformations and score mapping.
• Benchmark VS studies were performed to evaluate EnOpt's performance.

Main Results:

• EnOpt enhances the accuracy of ranking active compounds over inactive or decoy molecules.
• The tool demonstrates improved performance compared to traditional ensemble VS methods across various systems.
• EnOpt provides a more interpretable approach to analyzing compound scores.

Conclusions:

• EnOpt effectively addresses challenges in ensemble virtual screening.
• The tool offers a more accurate and interpretable method for prioritizing drug candidates.
• EnOpt is freely available, promoting wider adoption in computational drug discovery.