Maximizing gain in high-throughput screening using conformal prediction.

Area of Science:

• Computational Chemistry
• Drug Discovery
• Bioinformatics

Background:

• Traditional high-throughput screening is often inefficient.
• Iterative screening methods learn from data to improve efficiency.
• Evaluating screening campaigns involves balancing costs and potential gains.

Purpose of the Study:

• To develop a novel approach for maximizing gain in iterative screening.
• To couple conformal prediction with a gain-cost function for optimized compound selection.
• To accurately predict the most effective screening strategies and confidence levels.

Main Methods:

• Implemented a conformal predictor integrated with a gain-cost function.
• Evaluated model predictions on training data to forecast performance on test data.
• Tested the approach on 12 PubChem bioactivity datasets, using 20% of data for training.

Main Results:

• The approach accurately identified settings for maximum gain in 8-10 out of 12 datasets.
• It successfully predicted optimal strategies: screen predicted actives, screen all data, or screen no further compounds.
• When screening predicted actives, the method indicated the optimal confidence level to maximize gain.

Conclusions:

• The developed method enhances decision-making in iterative screening campaigns.
• It facilitates resource allocation by predicting the likely outcomes and maximizing value.
• This approach optimizes screening efficiency and the discovery of active compounds.