Classification of estrogenic compounds by coupling high content analysis and machine learning algorithms.

Area of Science:

• Endocrinology
• Toxicology
• Computational Biology

Background:

• Environmental toxicants, including endocrine-disrupting chemicals (EDCs), pose risks to human health.
• Estrogenic compounds interact with the estrogen receptor (ER), a key regulator in biological processes.
• Existing assays for estrogenic activity often lack the ability to capture multiple mechanistic endpoints at the single-cell level.

Purpose of the Study:

• To develop a machine learning pipeline for the rapid and accurate assessment of endocrine-disrupting potential.
• To classify compounds as estrogen receptor (ER) agonists or antagonists using high-content imaging data.
• To identify key features for predicting estrogenic activity.

Main Methods:

• Utilized a previously developed high-content imaging assay with a stable cell line expressing a green fluorescent protein (GFP)-ER fusion.
• Generated multiplex data from high-content analysis to capture multiple mechanistic ER endpoints.
• Trained and evaluated both logistic regression and Random Forest classifiers to predict ER activity.

Main Results:

• Achieved highly accurate and generalizable classification models for predicting ER agonists and antagonists.
• Identified informative features for classification through feature selection and data visualization.
• Demonstrated that machine learning models can effectively use minimal features for accurate prediction.

Conclusions:

• Machine learning pipelines can rapidly and sensitively assess the endocrine-disrupting potential of chemicals.
• High-content imaging data, when analyzed with machine learning, provides a powerful tool for mechanistic evaluation of estrogenic compounds.
• This data-driven approach enables efficient screening of numerous chemicals for ER activity.