Exploration of Chemical Space Covered by Nontarget Screening Based on the Prediction of Chemical Substances Amenable to LC-HRMS Analysis

Abstract

Nontarget screening (NTS) is a promising analytical technique for tracking emerging pollutants. However, the exact chemical space that can be covered by the method remains to be determined. A text-mining study in the literature noted that the number of compounds currently reported by NTS via liquid chromatography-high resolution mass spectrometry (LC-HRMS) was only about 2% of the approximate chemical space (i.e., NORMAN SusDat database). In view of the basic requirement on the presence of parent (MS¹) and daughter (MS²) ions at environmentally relevant concentrations for chemical identification, a binary classification model of artificial neural networks was developed based on the measured mass spectrum data of 1255 unique chemical substances. It was used to estimate the percentage of compounds amenable to LC-HRMS analysis from a broad range of candidates in chemical inventories. Molecular descriptors related to molecular size, branching, electronic states of atoms, and molecular charge distributions showed significant impacts on the sensitivity of the model. The predicted amenable compounds in the positive and negative modes of electrospray ionization accounted for about 41% and 23% of the approximate chemical space when the same database was used for comparison, suggesting a great potential for NTS within the LC-HRMS platform.