You are here:
Evaluation and Refinement of Quantitative Structure-Use Relationship (QSUR) Models for Chemical Function (ISES 2021)
Citation:
Hull, V., L. Eddy, K. Phillips, AND K. Isaacs. Evaluation and Refinement of Quantitative Structure-Use Relationship (QSUR) Models for Chemical Function (ISES 2021). International Society for Exposure Science (ISES) 2021 Virtual Meeting, Virtual, Virtual, August 30 - September 02, 2021. https://doi.org/10.23645/epacomptox.17909576
Impact/Purpose:
Poster presented to the International Society for Exposure Science (ISES) 2021 Virtual Meeting September 2021. The presentation will describe a project to update existing ORD machine learning models for chemical functional use. These models support a variety of Agency and external predictive models.
Description:
The U.S. Environmental Protection Agency’s (EPA’s) Office of Research and Development previously developed high-throughput quantitative structure-use relationship (QSUR) models that use the structure of chemicals to predict their functional use in consumer products and processes. These models can be used to fill gaps in exposure-relevant information for poorly characterized chemicals. Here, we evaluate these models based on industry-reported chemical functional uses from EPA and Health Canada and refine the models to improve performance and incorporate new harmonized functional use categories. The evaluation showed that the models were successful at predicting industry-reported chemical functional uses when the chemicals in question were within the models’ applicability domain (AD). However, many chemicals with reported industrial uses were outside the ADs of the QSUR models, as they were trained with data from primarily consumer product sources. To expand the model domain and refine model application, new data from the industrial sector were incorporated into an updated modeling framework that includes sector of use. The functional use dataset was augmented to include industry-specific functions from EPA’s Chemical Data Reporting (CDR), and the data were re-mapped to 107 internationally harmonized functional use categories developed by the Organisation for Economic Co-Operation and Development (OECD). New consensus QSURs were built for the OECD harmonized categories using a standardized workflow that included model validation. To address sector of use, a machine learning classifier model was built to predict consumer and/or industrial use. This model was applied to the augmented functional use data to assess the relevance of the model training data to each sector, with 75.7% of chemicals predicted to have industrial use, 60.4% having consumer use, and 55.5% having both. In the future, this model can be used to stratify QSUR training sets by sector. These refined QSUR models will inform existing models for predicting chemical release, consumer product composition, and exposure from chemical properties and structure, ultimately improving chemical prioritization workflows. This abstract does not reflect EPA policy.
URLs/Downloads:
DOI: Evaluation and Refinement of Quantitative Structure-Use Relationship (QSUR) Models for Chemical Function (ISES 2021)
HULLQSUR_ISES2021_FINAL.PDF (PDF, NA pp, 629.428 KB, about PDF)