Non-targeted screening with LC/HRMS is a go-to approach to discover relevant contaminants in environmental water samples that contain an abundance of chemicals. The rapidly increasing popularity of non-targeted LC/ HRMS screening has initiated development of a diverse set of methods for assessing the concentration and toxicity of the detected chemicals. This review aims to benchmark the trends in the environmental NTS literature with particular focus on (1) methods used for the quantification of tentatively identified chemicals that lack analytical standards, (2) methods for assessing the toxicity of detected chemicals, and (3) methods combining the former into a risk evaluation. Here we provide a scientometric review of these strategies based on the Web of Science referenced papers published between 2019 and 2022. General trends show that quantification and toxicity assessments are widely employed in NTS, reaching 66 % and 45 % over four years, respectively. Simultaneously, only 13 % of the papers covered here combine these results into a risk factor or similar. With this review we aim to highlight the advantages and gaps in the approaches used for concentration and toxicity assessment and provide guidelines for more homogeneous data interrogation and extrapolation.

Nontarget analysis by liquid chromatography-high-resolutionmass spectrometry (LC-HRMS) is now widely used to detect pollutants in the environment. Shifting away from targeted methods has led to detection of previously unseen chemicals, and assessing the risk posed by these newly detected chemicals is an important challenge. Assessing exposure and toxicity of chemicals detected with nontarget HRMS is highly dependent on the knowledge of the structure of the chemical. However, the majority of features detected in nontarget screening remain unidentified and therefore the risk assessment with conventional tools is hampered. Here, we developed MS2Quant, a machine learning model that enables prediction of concentration from fragmentation(MS2) spectra of detected, but unidentified chemicals. MS2Quant is an xgbTree algorithm-based regression model developed using ionization efficiency data for 1191 unique chemicals that spans 8 orders of magnitude. The ionization efficiency values are predicted from structural fingerprints that can be computed from the SMILES notation of the identified chemicals or from MS2 spectra of unidentified chemicals using SIRIUS+CSI: FingerID software. The root mean square errors of the training and test sets were 0.55(3.5x) and 0.80 (6.3x) log-units, respectively. In comparison, ionization efficiency prediction approaches that depend on assigning an unequivocal structure typically yield errors from 2x to 6x. The MS2Quant quantification model was validated on a set of 39 environmental pollutants and resulted in a mean prediction error of 7.4x, ageometric mean of 4.5x, and a median of 4.0x. For comparison, a model based on PaDEL descriptors that depends on unequivocal structural assignment was developed using the same dataset. The latter approach yielded a comparable mean prediction error of 9.5x, a geometricmean of 5.6x, and a median of 5.2x on the validation set chemicals when the top structural assignment was used as input. This confirms that MS2Quant enables to extract exposure information for unidentified chemicals which, although detected, have thus far been disregarded due to lack of accurate tools for quantification. TheMS2Quant model is available as an R-package in GitHub for improving discovery and monitoring of potentially hazardous environmental pollutants with nontarget screening.

Discovering hazardous pollutants currently relies on the tedious and often inaccurate structural identification step, required for further toxicity and exposure studies. Here, we propose and validate a workflow for prioritizing the detected features solely from their mass spectrometric data based on the priority score reflecting the risk posed by these chemicals. The workflow integrates two machine learning approaches, MS2Tox and MS2Quant, that predict toxicity and concentration of unidentified molecular features, respectively, and does not require any additional standards to be measured in the same run, allowing the application to digitally frozen data. Validation by using priority score for classifying 23 chemicals with available risk quotient values into high and low risk categories yielded recall value of 0.55 to 0.74, precision of 0.14 to 0.40, and accuracy of 0.46 to 0.69, depending on the acquisition mode and fish species. Applying the developed workflow to wastewater effluent prioritized 20-27% of featureswith predicted fingerprints as “precautionary risk features” with a risk quotient ≥1 based on the lower limit of the 95% prediction interval. All prioritized features were subject to spectral library matching, with 11% of the features yielding level 2 identification. Features categorized as high-risk were further subject to structural annotation using in silico identification tools SIRIUS+CSI:FingerID and MetFrag. While plausible candidates were suggested, the in silico tools rarely agreed within the top 10 suggested structures, highlighting structural assignment as a bottleneck that can be bypassed by data-driven feature prioritization.

Estimating the risk posed by a mixture of environmental contaminants is a complex task. Here, we aimed to predict toxic units for chemicals in mixtures to evaluate their contribution to the overall mixture toxicity. In order to validate the workflow, 36 designed mixtures representing different cumulative mixture toxicity scenarios were analyzed using liquid chromatography coupled to high-resolution mass spectrometry. Two fragmentation spectra approaches were investigated: mixture-specific fragmentation spectra (MS2) and MS2 spectra merged over a range of collision energies. To enable experimental validation, a fish embryo acute toxicity prediction model was developed, and the toxicity was experimentally determined for eight chemicals. Chemical-specific toxic units were predicted with a geometric mean fold error of 6-7× in positive and 6-21× in negative electrospray ionization mode, while the cumulative mixture toxicity was predicted with a geometric mean fold error of 2-3× for both structure and MS2-based methods. The merged MS2 spectra improved predictions compared to mixture-specific MS2 spectra and performed comparably with the structure-based predictions. These results demonstrate that both structure and MS2-based mixture toxicity predictions can be equally useful for further application on real-life samples.