Binary Classification of the Endocrine Disrupting Chemicals by Artificial Neural Networks

Area of Science:

• Computational toxicology
• Endocrinology
• Machine learning applications in chemical safety

Background:

• Endocrine-disrupting chemicals (EDCs) pose risks to human and wildlife health by interfering with hormone systems.
• Evaluating the vast number of existing chemicals for endocrine disruption potential using traditional in vitro or in vivo methods is challenging and time-consuming.

Purpose of the Study:

• To develop and validate a machine learning framework for predicting estrogen receptor activity of chemical compounds.
• To model the separation between estrogen receptor agonists and antagonists using high-content image analysis and artificial neural networks.

Main Methods:

• Integration of high-content/high-throughput image analysis with artificial neural networks (ANNs).
• Preprocessing of experimental data including cleaning, scaling, and feature engineering, focusing on the middle 50% of receptor-DNA binding values.
• Application of Principal Component Analysis (PCA) to minimize experimental noise and enhance feature representation for classification.

Main Results:

• The developed ANN-based framework achieved high accuracy in classifying estrogen receptor agonists and antagonists.
• The model demonstrated 98.41% accuracy in distinguishing between benchmark agonist and antagonist chemicals.

Conclusions:

• Machine learning, particularly ANNs combined with image analysis, offers a powerful and rapid approach for assessing chemical toxicity and endocrine-disrupting potential.
• This data-driven framework provides a promising tool for the comprehensive evaluation of chemical safety and the identification of potential endocrine disruptors.