TrendProbe: Time profile analysis of emerging contaminants by LC-HRMS non-target screening and deep learning convolutional neural network

Abstract

Peak prioritization is one of the key steps in non-target screening of environmental samples to direct the identification efforts to relevant and important features. Occurrence of chemicals is sometimes a function of time and their presence in consecutive days (trend) reveals important aspects such as discharges from agricultural, industrial or domestic activities. This study presents a validated computational framework based on deep learning conventional neural network to classify trends of chemicals over 30 consecutive days of sampling in two sampling sites (upstream and downstream of a river). From trend analysis and factor analysis, the chemicals could be classified into periodic, spill, increasing, decreasing and false trend. The developed method was validated with list of 42 reference standards (target screening) and applied to samples. 25 compounds were selected by the deep learning and identified via non-target screening. Three classes of surfactants were identified for the first time in river water and two of them were never reported in the literature. Overall, 21 new homologous series of the newly identified surfactants were tentatively identified. The aquatic toxicity of the identified compounds was estimated by in silico tools and a few compounds along with their homologous series showed potential risk to aquatic environment.

Keywords: Deep learning; Emerging contaminants; High resolution mass spectrometry; Non-target trend analysis; Surfactants.