You are here:
Evaluation of ecotoxicity using US EPA’s Cheminformatics Modules
Citation:
Martin, T. AND A. Williams. Evaluation of ecotoxicity using US EPA’s Cheminformatics Modules. SETAC Europe, Seville, SPAIN, May 05 - 09, 2024.
Impact/Purpose:
N/A
Description:
Datasets for acute aquatic toxicity, acute mammalian toxicity, and bioconcentration factor were compiled from publicly available data sources. Models were developed using QSAR methodologies such as random forest and XGBoost. Feature selection was employed to reduce the number of descriptors appearing in each model to a mechanistically defensible subset. Real time predictions will be available via the Predict module of US EPA’s Cheminformatics Modules. The Predict module utilizes a database centered approach to allow versioning of all aspects of model building. The Cheminformatics Modules enable users to search and compare a variety of chemical and hazard information to evaluate potential health effects of chemicals. The Predict module provides detailed model metadata via a QMRF (QSAR model reporting format) document and an Excel summary file. For each prediction, an HTML prediction report provides experimental and predicted values, an applicability domain analysis, and prediction results for similar chemicals from the training and test sets. The external prediction performance of each model will be compared to those from other EPA tools. The views expressed here are those of the authors and do not necessarily represent the views or the policies of the U.S. Environmental Protection Agency.