You are here:
Predictive Models for Elimination Patterns and Rates for Untested PCB Congeners to Facilitate Toxicokinetic Modeling
Citation:
Pradeep, P., L. Carlson, G. Lehmann, P. Schlosser, R. Judson, AND G. Patlewicz. Predictive Models for Elimination Patterns and Rates for Untested PCB Congeners to Facilitate Toxicokinetic Modeling. 2020 International Workshop on QSAR in Environmental and Health Sciences (QSAR 2020), Durham, North Carolina, June 08 - 11, 2020.
Impact/Purpose:
Abstract submitted to the 2020 International Workshop on QSAR in Environmental and Health Sciences (QSAR 2020) June 2020
Description:
Humans are exposed to PCBs as complex mixtures including both labile and persistent congeners, all of which may contribute to toxicological outcomes. The 209 unique PCB congeners exhibit a wide range of toxicokinetic (TK) parameters. As an initial step to developing a comprehensive TK model that could be applied to PCB mixtures, a systematic approach was used to identify and extract reported half-lives of elimination for individual congeners from three scientific databases (PubMed, Web of Science, and Toxline). A summary of selected half-lives for each congener following oral exposure coupled with qualitative levels of confidence was compiled. Congeners were also classified by 3 elimination patterns identified by Tanabe et al (1981). The final dataset comprised elimination type designations for 76 congeners and oral half-life values (in years) for 31 congeners. Descriptive analytics results showed that the experimental half-life values were significantly different across different elimination types revealing that elimination type was a qualitative indicator of half-life. Subsequently, predictive models were developed for elimination types; the half-lives for the untested congeners were imputed based on their predicted elimination type. The model was developed using the random forest algorithm, position-specific chlorine substitution patterns on the biphenyl scaffold, ToxPrint chemical fingerprints and physicochemical properties. The 5-fold cross-validated model had an accuracy of 66.7% on the training set and 75% on the test set. These models will help infer half-life for untested congeners which is useful for route-to-route and interspecies extrapolation of toxicological effect levels. This abstract does not necessarily represent U.S. EPA policy.