Practical outcomes of applying ensemble machine learning classifiers to High-Throughput Screening (HTS) data analysis...

Area of Science:

• Computational chemistry
• Cheminformatics
• Machine learning in drug discovery

Background:

• Machine learning methods are widely used for analyzing biological screening data in drug discovery.
• Traditional model evaluation using random holdout sets often leads to overly optimistic performance assessments.

Purpose of the Study:

• To develop and evaluate ensemble-based decision tree models for drug discovery screening.
• To investigate the practical impact of these models on high-throughput screening (HTS) campaigns.
• To provide a more realistic assessment of predictive model performance.

Main Methods:

• Development of ensemble-based decision tree models.
• Application of models to vendor screening data.
• Acquisition and screening of forecasted compounds.
• Evaluation of model impact on hit rates in HTS.

Main Results:

• Ensemble-based decision tree models were developed and refined through various stages.
• Models were applied to vendor offerings, and forecasted compounds were screened.
• Well-developed models demonstrated a significant increase in observed hit rates during HTS campaigns.

Conclusions:

• Ensemble-based decision tree models offer a more robust approach to analyzing drug discovery screening data.
• These models can substantially enhance the efficiency and success of HTS campaigns.
• The findings challenge the reliability of traditional random holdout validation methods.