You are here:
Statistical Evaluation of Quantitative Non-Targeted Analysis Methods Using ENTACT Data
Citation:
Groff, L., A. Kruve, C. Lowe, J. Minucci, D. Kapraun, Tom Purucker, J. Grossman, J. McCord, K. Phillips, E. Ulrich, AND J. Sobus. Statistical Evaluation of Quantitative Non-Targeted Analysis Methods Using ENTACT Data. American Chemical Society (ACS) Fall 2021 National Meeting, Virtual, NC, August 22 - 26, 2021. https://doi.org/10.23645/epacomptox.17430578
Impact/Purpose:
Presentation to the American Chemical Society (ACS) Fall 2021 National Meeting August 2021. Non-targeted analysis (NTA) strives to detect and identify a wide range of chemicals without preconceived target list or standards. Such efforts can be applied to and use a variety of exposure science efforts, including basic research into the technique (ENTACT), informatics tools (CompTox Chemicals Dashboard), complex mixtures (UVCBs) and pairing exposure to effects (EDA). Approximately 30 laboratories worldwide are characterizing their non-targeted analysis approaches. Because the instrumental and data processing methods vary significantly, so do performance metrics. Physicochemical properties are statistically different between instrumental method choices (gas vs. liquid chromatography, electrospray vs. atmospheric pressure chemical ionization) showing the techniques cover different portions of chemical space. Current and future NTA research areas are also linked to larger scientific efforts within the exposure community at EPA.
Description:
For decades, targeted analysis methods have been the gold standard for chemical quantification in environmental studies. These methods, while highly accurate, have limited practical use when chemical identities are unknown, or when numbers of chemicals per sample approach thousands. Further, targeted analyses require prior knowledge of the chemicals of interest, and readily obtainable analytical standards. Conversely, non-targeted analysis (NTA) methods aim to identify and quantify thousands of chemicals in any sample of interest without prior knowledge. While NTA chemical identification methods have rapidly matured, ongoing efforts toward NTA quantification have lagged, and generally do not consider statistical confidence about individual predictions. For NTA concentration estimates to be utilized in regulatory contexts, statistically defensible methods of NTA chemical quantitation must be developed. Furthermore, since bounded concentration estimates are generally specific to prepared sample extracts (in solvent), extrapolation methods are needed to extend quantitative predictions to the original sampled media, and eventually, to hazard metrics that can support risk characterization. Here, we detail the mathematical basis of traditional calibration methods used in targeted analyses, and how these methods can be extended to support quantitative estimates for all chemicals observed in NTA experiments. A default one-dimensional method is detailed for bounding concentration estimates based on an observed (and transformed) distribution of instrument response factors. A more complex method is then described, based on modeled ionization efficiencies calculated from machine-learning models using physicochemical descriptors. The uncertainty in both quantitative NTA methods is compared to the uncertainty observed when performing automated NTA experiments with traditional approaches (using calibration curves). We show that, in lieu of analytical standards, statistically defensible quantitative NTA is feasible using either method given a base set of known response factors. The views expressed are those of the author(s) and do not necessarily reflect the views or policies of the US EPA.
URLs/Downloads:
DOI: Statistical Evaluation of Quantitative Non-Targeted Analysis Methods Using ENTACT Data
LCGROFF_QUANTNTA1B_ACSFALL2021_V5.PDF (PDF, NA pp, 5305.596 KB, about PDF)