Combining High-Throughput Screening and In Silico Modeling to Derisk Novel Agrochemicals for Androgen Receptor...

Area of Science:

• Environmental chemistry
• Toxicology
• Computational chemistry

Background:

• Androgen receptor (AR) modulation is a key safety concern for environmental chemicals due to endocrine disruption.
• Existing public data for AR modulation lacks sufficient size and diversity for robust analysis.

Purpose of the Study:

• To conduct a large-scale high-throughput screening (HTS) of agrochemicals for AR binding.
• To develop and validate machine learning models for predicting AR binding activity.
• To create a comprehensive, publicly available dataset for AR modulation research.

Main Methods:

• Utilized a fluorescence polarization displacement assay for HTS of over 72,000 agrochemical compounds.
• Performed confirmatory dose-response testing to identify AR binders.
• Curated a dataset of 24,953 compounds with activity data and trained machine learning models (gradient-boosted trees) using molecular descriptors.

Main Results:

• Identified 4,183 AR binders (5.7% hit rate) with significant structural diversity.
• Developed machine learning models with a balanced accuracy of 0.77 and NPV of 0.98 for predicting AR binders.
• Achieved reasonable transferability of models to external datasets (balanced accuracy 0.66-0.72).

Conclusions:

• HTS combined with machine learning is effective for early chemical safety assessment.
• The study provides a valuable benchmark dataset to improve AR binding prediction.
• The developed models can aid in derisking large virtual chemical libraries for AR modulation potential.