Docking-based classification models for exploratory toxicology studies on high-quality estrogenic experimental data.

Area of Science:

• Computational toxicology
• Cheminformatics
• Structure-based drug design

Background:

• Animal testing faces ethical and practical limitations, driving the need for computational alternatives.
• High-throughput screening of chemical libraries requires efficient and reliable predictive methods.

Purpose of the Study:

• To develop and validate computational models for predicting the estrogenic potential of chemicals.
• To assess the adaptability of structure-based methods for exploratory toxicology.

Main Methods:

• Derived 24 docking-based classification models.
• Utilized a large chemical collection from the US Environmental Protection Agency.
• Validated model performance using AUC, EF1%, and likelihood ratios with reference compounds.

Main Results:

• Successfully predicted estrogenic potential for a large chemical dataset.
• Achieved reliable performance metrics, including AUC and EF1%.
• Demonstrated successful external validation on known estrogenic compounds and a large chemical set (>32,000).

Conclusions:

• Structure-based computational methods are effective for exploratory toxicology.
• These methods can be adapted from drug discovery for toxicological assessments.
• The developed models provide a reliable tool for screening chemical libraries.