Related Experiment Video
Updated: Aug 26, 2025
05:47
In Silico Modeling Method for Computational Aquatic Toxicology of Endocrine Disruptors: A Software-Based Approach Using QSAR Toolbox
Published on: August 28, 2019
14.1K
Novel machine learning models to predict endocrine disruption activity for high-throughput chemical screening
Sean P Collins1, Tara S Barton-Maclaren1
1Existing Substances Risk Assessment Bureau, Healthy Environments and Consumer Safety Branch, Health Canada Ottawa, Ottawa, ON, Canada.
Frontiers in Toxicology
|October 7, 2022
Summary
Computational toxicology aids endocrine disrupting chemical (EDC) assessment. New Random Forest models predict estrogen and androgen receptor activity, improving screening for thousands of untested chemicals with high accuracy.
Area of Science:
- Toxicology and Chemical Risk Assessment
- Computational Toxicology
- Endocrine Disruption
Background:
- Thousands of legacy chemicals lack toxicity testing for endocrine disrupting potential.
- Computational toxicology offers a crucial role in assessing these chemicals.
- US EPA programs CERAPP and CoMPARA aimed to predict estrogen and androgen activity.
Purpose of the Study:
- To develop and present Random Forest (RF) models for predicting estrogen and androgen receptor activity.
- To expand the applicability domain for endocrine disrupting activity prediction.
- To aid in the screening and prioritization of extensive chemical inventories.
Main Methods:
- Utilized large datasets from CERAPP and CoMPARA for estrogen and androgen activity.
- Developed Random Forest (RF) models using simple descriptors from open-source software.
- Trained RF models to conservatively predict activity, minimizing false negatives.
Main Results:
- Presented twelve binary and multi-class RF models for predicting estrogen and androgen receptor binding, agonism, and antagonism.
- RF models demonstrated high predictive capabilities, with some achieving 93% balanced accuracy and 89% coverage.
- These models showed superior performance compared to other in silico methods.
Conclusions:
- RF models effectively predict endocrine receptor activity, aiding in the identification of potential endocrine-disrupting substances.
- These models can be integrated into priority-setting workflows for chemical screening.
- The developed models support the selection of chemicals for further testing and assessment.
