You are here:
Leveraging Publically Available Chemical Functional Use Data in Support of Exposure Prediction
Citation:
Isaacs, K., K. Phillips, P. Egeghy, K. Dionisio, AND J. Wambaugh. Leveraging Publically Available Chemical Functional Use Data in Support of Exposure Prediction. 2016 Annual International Society of Exposure Science Meeting, Utrecht, NETHERLANDS, October 09 - 13, 2016.
Impact/Purpose:
The U.S. EPA Exposure Forecasting (ExpoCast) project aims to provide rapid screening-level exposure predictions for thousands of chemicals, most of which lack detailed exposure data. The functional use data and models will be used to refine and/or parameterize heuristic-based and mechanistic models of human exposures in ExpoCast.
Description:
The U.S. EPA Exposure Forecasting (ExpoCast) project aims to provide rapid screening-level exposure predictions for thousands of chemicals, most of which lack detailed exposure data. Chemical functional use - the role a chemical plays in processes or products (e.g. solvent, antimicrobial, plasticizer) - may be a useful heuristic for predicting exposure potential in that it reflects both the compound’s likely physical properties as well as the product formulations, consumer articles, or industrial processes in which it may be used. Functional use information is also critical in alternatives assessment, in which safer chemicals that can perform a particular role in products are identified. Here, data on chemical functional use for more than 14,000 chemicals were collected from publically available government, manufacturer, and industry sources. A new standardized Functional Use (FUse) database was created by harmonizing 240 function categories across sources. The FUse database was used to build machine-learning classifier models for function and consumer product weight fraction using either predicted physical-chemical properties or chemical structural descriptors. Validated models could built for 49 functions and weight fraction. The final models were applied to a library of 10,000 mostly data-poor chemicals, including those being tested using high-throughput methods in the Toxicology in the 21st Century (Tox21) program. In addition, the predictions generated by the classification models were used to screen the chemical library for potential alternatives on the basis of an average in-vitro bioactivity metric generated from a suite of 16 Tox21 assays. Function could be predicted with high probability (>80%) for 2,332 chemicals; of these chemicals, 1034 had a lower bioactivity metric than at least one known chemical with that function. The functional use data and models will be used to refine and/or parameterize heuristic-based and mechanistic models of human exposures in ExpoCast. This abstract does not necessarily reflect U.S. EPA policy.
