Prediction of Endocrine-Disrupting Chemicals Related to Estrogen, Androgen, and Thyroid Hormone (EAT) Modalities...

Area of Science:

• Toxicology and pharmacology
• Computational chemistry
• Genomics

Background:

• Endocrine-disrupting chemicals (EDCs) pose significant public health risks, linked to cancer, obesity, diabetes, and infertility.
• The endocrine system's estrogen (E), androgen (A), and thyroid hormone (T) pathways are critical homeostatic regulators.
• Accurate in silico tools are needed for rapid identification of EDCs and their mechanisms of action.

Purpose of the Study:

• To develop and validate Qualitative Gene expression Activity Relationship (QGexAR) models for predicting EDC-induced disruption of EAT modalities.
• To optimize prediction protocols using diverse feature selection and machine learning algorithms.
• To identify key genes involved in EDC perturbation mechanisms.

Main Methods:

• Utilized gene expression profiles from the LINCS database for MCF7 and A549 cell lines.
• Implemented and compared various classification algorithms including CATBoost, XGBoost, SVM, and deep learning.
• Employed consensus prediction strategies based on the best-performing models for each cell line and endpoint.

Main Results:

• Developed predictive models for estrogen receptor, androgen receptor binding, and thyroid hormone receptor antagonistic effects.
• Achieved a consensus balanced accuracy on the validation set ranging from 0.725 to 0.840.
• Identified important predictive features, highlighting known and suggesting novel genes in EAT perturbation.

Conclusions:

• QGexAR models provide an effective in silico approach for predicting endocrine disruption.
• The developed models contribute to the rapid toxicological assessment of chemicals.
• Feature importance analysis offers insights into the molecular mechanisms of endocrine disruption.