Efficient modeling and active learning discovery of biological responses.

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• High throughput screening (HTS) and high content screening (HCS) assess compound effects on targets.
• Current screening methods underutilize data from prior targets and neglect off-target effects in drug development.
• Exhaustive screening of all compounds against all targets is prohibitively expensive.

Purpose of the Study:

• To develop a cost-effective solution for predicting compound-target interactions.
• To enhance the efficiency, reduce the cost, and shorten the timeline of drug development.
• To create predictive models for unmeasured compound-target combinations using probabilistic models and active learning.

Main Methods:

• Utilized probabilistic models to predict results for unmeasured compound-target combinations.
• Employed active learning algorithms to efficiently select experiments for model building.
• Determined optimal stopping points for experimentation to conserve resources.

Main Results:

• Developed powerful predictive models without requiring exhaustive experimentation.
• Demonstrated that active learning significantly accelerates model learning compared to random experiment selection.
• Validated approaches using both simulated and experimental data.

Conclusions:

• Probabilistic models and active learning offer a more effective, economical, and timely approach to drug discovery.
• These methods enable the prediction of compound effects, optimizing the selection of drug candidates.
• The proposed framework significantly improves the efficiency of building predictive models in screening campaigns.