You are here:
Identifying Chemical Signatures of Manufacturing and Recycling in Household Products
Citation:
Isaacs, K. AND J. Sobus. Identifying Chemical Signatures of Manufacturing and Recycling in Household Products. 2019 SOT Annual Meeting, Baltimore, MD, March 09 - 14, 2019.
Impact/Purpose:
This presentation will describe the challenges inherent in understanding high-resolution mass spectrometry (HRMS) data generated in non-targeted (NTA) analyses of the chemical constituents of household products. The talk is in a Workshop session titled "Order from Chaos: Pattern Recognition in Challenging Human Health Datasets." The talk describes EPA informatics and data mining approaches for identifying patterns in NTA data potentially associated with exposure sources.
Description:
This presentation will describe the challenges inherent in understanding high-resolution mass spectrometry (HRMS) data generated in non-targeted (NTA) analyses of the chemical constituents of household products. These methods are attractive in that they can capture information on thousands of molecular features (and thus chemicals) in a single analytical run without any a priori knowledge of the compounds present. These studies are being used by EPA to rapidly screen consumer products articles of commerce for chemical content. These NTA screens generate large numbers of chemical features that are understood in part through comparison against large, newly compiled databases of chemical ingredients and weight fractions in household products. However, not all chemicals can be identified in these databases. Consumer product ingredient databases have therefore served as training sets for machine learning tools capable of predicting the functional use of chemicals in consumer products based upon structure alone. These tools prioritize chemicals for further study and identify novel compounds. Across many media samples of a given type, analysis of these features can uncover co-occurring groups of chemicals (“signatures”) that may be indicative of unique exposure sources. In consumer products, signatures could be associated with intentional (functional) addition, manufacturing process, or contamination. While unsupervised clustering and co-clustering methods are useful in elucidating local patterns, there are challenges in applying them to HRMS data. True variability in chemical content across samples, reproducibility in analytical results, and uncertainly in mapping of molecular features to chemicals all contribute to noise that imparts a large degree of complexity (and thus computational burden) to the problem. Where sample metadata are available, supervised or semi-supervised approaches can be used to further focus the analysis. Case studies of unsupervised and supervised approaches will be presented and challenges discussed. Development of standard methods and tools for identifying chemical signatures can improve identification, prioritization, and mitigation of chemicals in consumer products
