You are here:
A quantitative structure-activity relationship (QSAR) model to estimate serum half-lives of per- and polyfluoroalkyl substances (PFAS) in multiple species
Citation:
Dawson, D., Chris Lau, P. Pradeep, R. Judson, R. Tornero-Velez, AND J. Wambaugh. A quantitative structure-activity relationship (QSAR) model to estimate serum half-lives of per- and polyfluoroalkyl substances (PFAS) in multiple species. International Society of Exposure Scientists, n/a, Virtual Meeting, September 20 - 24, 2020. https://doi.org/10.23645/epacomptox.13158107
Impact/Purpose:
Perfluoroalky substances(PFAS) are group of man-made chemicals that are facing increasing concern and scrutiny because they are commonly found in both environmental substrates and body tissues, and frequently have long residence times in both. To quantify exposure to these chemicals, their properties need to be described and included into toxicokinetic and toxicodynamic models. However, important properties of many of these chemicals are unknown (e.g.,in-vivo half life), and are not easily extrapolated from known chemicals using standard statistical techniques. In this work, we use machine learning to leverage information known about a subset of PFAS chemicals in 4 species to build models predicting in-vivo plasma half-life for PFAS chemicals. When applied to a set of PFAS chemicals curated by USEPA(>6600), we found that approximately 90% would have a half life exceeding 1 year in humans. This work is significant because it represents an important first step towards quantifying a critical toxicokinetic property of PFAS chemicals. It will be of interest to both regulators as well as regional partners tasked with assessing the risks posed by PFAS.
Description:
Perfluoroalkyl substances (PFAS) are a diverse group of man-made chemicals that are known to be widely distributed in the environment and commonly found in body tissues. The long in-vivo half lives (t1/2) and environmental persistence of some PFAS's are a factor in the increasing focus to this group of compounds. Understanding and modeling the toxicokinetics of PFAS's is of interest for both risk prioritization and risk assessment. However, in-vivo assays of PFAS's have demonstrated that t1/2 in plasma, a key toxicokinetic parameter, can vary widely between species, and between sexes of the same species for a given chemical. Thus read-across methods of predicting t1/2 for non-assayed PFAS chemicals are unreliable. In this study, we use a machine learning approach to construct a model of in-vivo PFAS t1/2. From a curated set of literature, we assembled a training set of in-vivo plasma t1/2 data for 11 PFAS chemicals across 4 species (human, monkey, rat, mouse). We also assembled a diverse predictor set that included kidney characteristics, fatty acid and albumin binding coefficients, critical micelle concentrations, similarity values with endogenous metabolites, and a suite of predicted physio-chemical characteristics from the EPA OpeRa Suite. We fit random forest classification and regression models (10x cross-validation, 10 reps apiece) using 29 (27 numeric, 2 categorical) predictors. The classification model used data points (66) binned t1/2 into fast (1 Day and 1 Year). The classification model had an average accuracy of 96.2% ± 0.018% SE, while the regression model had an average R2 = 0.81 ± 0.07 SE. For the classification model, the relative importance of variables dropped sharply after the first 3 variables, while variable relative importance dropped steadily for the regression model. When the classification model is applied to 6648 chemicals of the DSSTox PFAS list, approximately 90% are predicted fall into the very slow bin, with the remainder falling into the slow bin. This effort represents a first pass at a complex problem, the reliability of which can only be assessed against more data. In particular, regression model results suggest that the t1/2 of PFAS's with very rapid decay may be overestimated by the model. Overall, however, our results indicate that a large number of PFAS chemicals would be expected to persist for long periods of time within human tissues.
URLs/Downloads:
DOI: A quantitative structure-activity relationship (QSAR) model to estimate serum half-lives of per- and polyfluoroalkyl substances (PFAS) in multiple species
DAWSON_PFAS_QSAR_ISES_2020.PDF (PDF, NA pp, 394.263 KB, about PDF)