Improving Screening Efficiency through Iterative Screening Using Docking and Conformal Prediction.

Area of Science:

• Computational chemistry
• Drug discovery
• Machine learning

Background:

• High-throughput screening (HTS) is a standard but costly drug discovery method.
• Current HTS methods demand significant time, effort, and resources.
• There is a need for more efficient screening strategies.

Purpose of the Study:

• To develop and validate an improved iterative screening process for drug discovery.
• To reduce the cost and time associated with identifying active compounds.
• To enhance the hit rate compared to traditional docking methods.

Main Methods:

• Employing an iterative screening process starting with molecular docking.
• Utilizing conformal prediction to classify remaining compounds based on initial screening results.
• Retrospective validation using 41 targets from the Directory of Useful Decoys, Enhanced (DUD-E) dataset.

Main Results:

• Successfully identified 57% of remaining active compounds while screening only 9.4% of the database.
• Achieved an overall hit rate of 7.6%, surpassing docking alone (5.2%).
• Identified 39.6% of active compounds within the top-scored docking compounds, outperforming docking-only screening (13.5%).

Conclusions:

• Iterative screening integrating molecular docking and conformal prediction offers enhanced efficiency in drug discovery.
• This method allows for the identification of active compounds with reduced screening volumes.
• Conformal predictors provide guidance for optimizing subsequent screening iterations.