Office of Research and Development Publications

Consensus models to predict endocrine disruption for all human-exposure chemicals (AAAS Annual Meeting)

Citation:

Mansouri, K. AND R. Judson. Consensus models to predict endocrine disruption for all human-exposure chemicals (AAAS Annual Meeting). Presented at AAAS annual meeting, Boston, MA, February 16 - 20, 2017. https://doi.org/10.23645/epacomptox.5057698

Impact/Purpose:

Slide Presentation to be presented at the AAAS annual meeting. Collaborative Estrogen Receptor Activity Prediction Project (CERAPP) was intended to be a demonstration of the use of predictive computational models on HTS data including ToxCast and Tox21 assays to prioritize a large chemical universe of 32464 unique structures for one specific molecular target – the estrogen receptor. CERAPP combined multiple computational models for prediction of estrogen receptor activity, and used the predicted results to build a unique consensus model.

Description:

Humans are potentially exposed to tens of thousands of man-made chemicals in the environment. It is well known that some environmental chemicals mimic natural hormones and thus have the potential to be endocrine disruptors. Most of these environmental chemicals have never been tested for their ability to disrupt the endocrine system, in particular, their ability to interact with the estrogen receptor. EPA needs tools to prioritize thousands of chemicals, for instance in the Endocrine Disruptor Screening Program (EDSP). Collaborative Estrogen Receptor Activity Prediction Project (CERAPP) was intended to be a demonstration of the use of predictive computational models on HTS data including ToxCast and Tox21 assays to prioritize a large chemical universe of 32464 unique structures for one specific molecular target – the estrogen receptor. CERAPP combined multiple computational models for prediction of estrogen receptor activity, and used the predicted results to build a unique consensus model. Models were developed in collaboration between 17 groups in the U.S. and Europe and applied to predict the common set of chemicals. Structure-based techniques such as docking and several QSAR modeling approaches were employed, mostly using a common training set of 1677 compounds provided by U.S. EPA, to build a total of 42 classification models and 8 regression models for binding, agonist and antagonist activity. All predictions were evaluated on ToxCast data and on an external validation set collected from the literature. In order to overcome the limitations of single models, a consensus was built weighting models based on their prediction accuracy scores (including sensitivity and specificity against training and external sets). Individual model scores ranged from 0.69 to 0.85, showing high prediction reliabilities. The final consensus predicted 4001 chemicals as actives to be considered as high priority for further testing and 6742 as suspicious chemicals. The same approach is now being applied on a larger scale project to predict the potential androgen receptor (AR) activity of chemicals. This project called CoMPARA (Collaborative Modeling Project for Androgen Receptor Activity) is a collaboration between 35 international groups working on a common set of ~55k chemicals.This abstract does not necessarily reflect U.S. EPA policy