Science Inventory

Developing a Framework to Assess and Optimize Transcriptomic Dose-Response Models (SOT 2023)

Citation:

Fredenburg, Jake, F. Harris, D. Haggard, J. Harrill, AND L. Everett. Developing a Framework to Assess and Optimize Transcriptomic Dose-Response Models (SOT 2023). Society of Toxicology 62nd Annual Meeting and ToxExpo 2023, Nashville, TN, March 19 - 23, 2023. https://doi.org/10.23645/epacomptox.22714039

Impact/Purpose:

This poster will be presented to the Society for Toxicology meeting in March 2023. It is intended to communicate recent findings on best practices for transcriptomic dose-response modeling.

Description:

Recent technological advancements have led to the development of new high-throughput profiling methods, such as transcriptomics, that can be used to rapidly screen environmental chemicals for potential hazards. It is now feasible to profile all protein-coding genes across thousands of samples, allowing for broad evaluation of many target pathways and mechanisms of action simultaneously in a single experiment. US EPA has developed a robust and reliable workflow to rapidly screen chemical effects in vitro using High-throughput Transcriptomics (HTTr). However, the scale and complexity of the resulting data is much greater than previous high-throughput screening (HTS) assays (e.g. ToxCast), and a variety of analysis approaches are available to derive key outputs of interest, such as estimating an overall transcriptomic point-of-departure (tPOD). The accuracy and reproducibility of tPOD estimates may depend on a broad array of analysis parameters that have not yet been fully explored. To date, there is limited consensus on best practices for analysis of transcriptomic concentration-response data, and insufficient understanding of how specific analysis choices impact the primary estimates of interest. While it is unlikely that there is a single “one size fits all” approach suitable for the increasing variety of transcriptomic studies, technologies, and applications, there remains a major unmet need for a standardized framework to evaluate and tailor analysis methods for specific use cases. Here we propose a framework to evaluate the reliability and reproducibility of transcriptomic concentration-response modeling methods, including 1) the false positive rate of “hit calls” for concentration-responsive genes, 2) the accuracy of concentration-response models, and 3) the reproducibility of tPODs. To assess false positive rate, we propose a method to simulate inactive chemical profiles by randomly sampling HTTr profiles of cells treated with relatively low concentrations of chemicals within larger screening studies. To assess the accuracy of concentration-response models, we generated “synthetic” transcriptomic concentration-response data by precisely adjusting the relative dilutions of two standard reference mixtures. Within this data set, we have simulated multiple effect sizes and curve shapes matching commonly used parametric models. Finally, to assess the reproducibility of tPOD values, we have repeatedly tested several standard reference chemicals throughout our HTTr screening studies, resulting in dozens of replicate experiments that can be analyzed independently and compared based on the final tPOD estimates. We demonstrate how this framework can be used to optimize and compare existing methods such as BMDExpress, which has numerous tunable parameters. Our preliminary results demonstrate that 1) current methods have acceptable false positive rates (≤ 1%) at the level of identifying individual concentration-responsive genes, 2) BMDExpress can accurately model the majority of concentration-responsive curves in our synthetic data, and 3) that tPOD estimates based on the 25th lowest gene-level BMD are highly reproducible with standard deviations below 2-fold based on our reference chemical data. These results can inform best practices and further optimize methods for transcriptomic concentration-response modeling, which will help build confidence in these assays for use in regulatory decision-making. This abstract does not necessarily reflect US EPA policy. Company or product names do not constitute endorsement by US EPA.