Deciding when to stop: efficient experimentation to learn to predict drug-target interactions.

Area of Science:

• Computational chemistry
• Drug discovery and development
• Machine learning in pharmacology

Background:

• Active learning optimizes experimentation by selecting high-impact tests.
• It significantly reduces experiments for confident drug-target predictions.
• Reliable stopping criteria are essential for cost and time efficiency in active learning.

Purpose of the Study:

• To develop a method for evaluating active learning prediction quality.
• To establish reliable stopping criteria for active learning in drug discovery.
• To quantify potential savings in experimental efforts.

Main Methods:

• Simulated drug-target matrices were used to compute active learning traces.
• A regression model was developed to predict active learner accuracy.
• The model's performance on simulated data informed the design of stopping criteria.

Main Results:

• A regression model accurately predicted active learning performance on simulated data.
• Novel stopping criteria were designed based on this model.
• Application of these criteria to real-world drug effect data yielded up to 40% savings in experiments.

Conclusions:

• Active learning's accuracy can be reliably predicted using simulated data.
• The developed stopping criteria significantly reduce the number of experiments needed for accurate drug-target predictions.
• This approach offers substantial time and cost savings in drug discovery pipelines.