You are here:
Assessing Machine Learning Methods in the Identification and Quantification of Environmental Chemical-Key Event Pairs Associated with Adverse Health Outcomes
Citation:
Mortensen, H., T. Allen, J. Senn, G. Patlewicz, AND I. Shah. Assessing Machine Learning Methods in the Identification and Quantification of Environmental Chemical-Key Event Pairs Associated with Adverse Health Outcomes. Society of Toxicology 2021 Virtual Annual Meeting, RTP, NC, March 12 - 26, 2021. https://doi.org/10.23645/epacomptox.14466138
Impact/Purpose:
Poster presented to the Society of Toxicology annual meeting in March 2021. This product will focus on how molecular genetic susceptibility information related to specific adverse outcome can be utilized in existing studies of community-level risk. Using the EPA Adverse Outcome Pathway Database (AOP-DB) we will investigate the inclusion of population level data to produce indicator metrics of human genetic susceptibility at the AOP level. These indicator metrics will both support known, and identify novel, key events within an adverse outcome that exhibit levels of defined population variability. These data can be used to perform toxicodynamic predictions for a set of the most potent environmental chemicals that have gone under TSCA prioritization to identify in vitro points of departure and corresponding human oral equivalent values using medium throughput testing. Finally, this product will address how these data can be best visualized and communicated for use by the larger risk assessment community. This product provides regulatory scientists, students and researchers with the ability to effectively access and exploit the many in silico data streams to support different regulatory purposes and supports current Agency efforts to reduce mammal study requests by 30% by 2025, and completely eliminate all mammal study requests and funding by 2035.
Description:
The ability to link the chemistry of chemical toxicants to adverse outcomes at high levels of biological organization allows for greater understanding of the biological mechanisms responsible for toxicity. The adverse outcome pathway (AOP) construct is useful in establishing and documenting the biological mechanism(s) implicated in adverse health outcomes of toxicological concern. Though the AOP construct is chemically agnostic by definition, there is an interest in identifying chemical groups that are known to affect particular pathway function, specifically in the identification of molecular events that are both critical in the progression of the outcome and potential candidates for in vitro assay development. Modelling these molecular events and linking them to higher level key events allows for the effective combination of in vitro and in silico toxicology tools. Here we explore the US EPA Adverse Outcome Pathway Database (AOP-DB) to identify AOPs and associated key event gene identifiers, which include molecular identifying events (MIEs) and key events (KEs). This project links 79 machine learning neural network models constructed using publicly available data for the prediction of MIEs to 280 unique AOPs (1111 KEs and 193 MIEs) in the AOP-DB. Model performance has been evaluated and prediction accuracy is above 90% on test data and 75% on external validation data. This work is key to the development of next-generation risk assessment procedure, and new approach methodologies, which desire to move away from in vivo experiments. This abstract does not reflect EPA Policy.
URLs/Downloads:
DOI: Assessing Machine Learning Methods in the Identification and Quantification of Environmental Chemical-Key Event Pairs Associated with Adverse Health Outcomes
MORTENSEN_ALLEN_02_26_21_50TH_ORD_TEMPLATE_FINAL 2.PDF (PDF, NA pp, 1090.991 KB, about PDF)