Rapid Screening of Consumer Products by GCxGC-HRT and Machine Learning Assisted Data Processing.

ABSTRACT

This work demonstrates high-throughput screening of personal care products to provide an overview of potential exposure. Sixty-seven products from five categories (body/fragrance oil, cleaning product, hair care, hand/body wash, lotion, sunscreen) were rapidly extracted and then analyzed using suspect screening by two-dimensional gas chromatography (GCxGC) high-resolution mass spectrometry (GCxGC-HRT). Initial peak finding and integration were performed using commercial software, followed by batch processing using the machine learning program Highlight. Highlight automatically performs background subtraction, chromatographic alignment, signal quality review, multidilution aggregation, peak grouping, and iterative integration. This data set resulted in 2,195 compound groups and 43,713 individual detections. Compounds of concern (101) were downselected and classified as mild irritants (29%), environmental toxicants/severe irritants (51%) and endocrine disrupting chemicals/carcinogens (20%). High risk compounds such as phthalates, parabens, and avobenzone were detected in 46 out of the 67 products (69%), and only 5 out of the 67 products (7%) listed these compounds on their ingredient labels. The Highlight results for the compounds of concern were compared to commercial software results (ChromaTOF) and 5.3% of the individual detections were discerned only by Highlight, demonstrating the strength of the iterative algorithm to effectively discover low-level signatures. Highlight provides a significant labor advantage, requiring only 2.6% of the time estimated for a largely manual workflow using commercial software. In order to address significant time needed for postprocessing assignment of identification confidence, a new machine-learning-based algorithm was developed to assess the quality of assigned library matches, and a balanced accuracy of 79% was achieved.