Screening and identification of unknown chemical contaminants in food based on liquid chromatography–high-resolution mass spectrometry and machine learning

Abstract

Unknown or unexpected chemical contaminants and/or their transformation products in food that may be harmful to humans need to be discovered for comprehensive safety evaluation. Liquid chromatography–high-resolution mass spectrometry (LC-HRMS) is a powerful tool for detecting chemical contaminants in food samples. However, identifying all of peaks in LC-HRMS is not possible, but if class information is known in advance, further identification will become easier. In this work, a novel MS2 spectra classification-driven screening strategy was constructed based on LC-HRMS and machine learning. First, the classification model was developed based on machine learning algorithm using class information and experimental MS2 data of chemical contaminants and other non-contaminants. By using the developed artificial neural network classification model, in total 32 classes of pesticides, veterinary drugs and mycotoxins were classified with good prediction accuracy and low false-positive rate. Based on the classification model, a screening procedure was developed in which the classes of unknown features in LC-HRMS were first predicted through the classification model, and then their structures were identified under the guidance of class information. Finally, the developed strategy was tentatively applied to the analysis of pork and aquatic products, and 8 chemical contaminants and 11 transformation products belonging to 8 classes were found. This strategy enables screening of unknown chemical contaminants and transformation products in complex food matrices.