Non-target analysis for wastewater treatment evaluation: strategies, applications, and future directions

Abstract

Non-target analysis (NTA), based on high-resolution mass spectrometry (HRMS), is increasingly used to characterize the complex chemical composition of wastewater and evaluate the fate of emerging contaminants (EC) during treatment. However, despite its growing application, the use of NTA for wastewater treatment evaluation remains fragmented across the literature, with existing reviews focusing primarily on analytical workflows, contaminant identification, or environmental monitoring rather than treatment performance assessment. This review addresses this gap by providing a comprehensive synthesis of NTA as a framework for evaluating wastewater treatment processes. The principal NTA-based evaluation strategies reported in the literature are consolidated and classified, including assessment of changes in chemical complexity, investigation of contaminant fate throughout treatment trains, and analysis of shifts in physicochemical characteristics inferred from molecular information. Also, the review examines the wastewater-specific NTA workflow and synthesizes emerging themes across recent studies. Particular emphasis is placed on how these approaches can be translated into engineering-relevant metrics for assessing treatment efficiency and evaluating residual chemical risks. By bridging analytical chemistry and wastewater engineering perspectives, this review demonstrates how NTA can move beyond contaminant discovery to support technology comparison, process optimization, and risk-informed decision-making, while highlighting key challenges and research priorities for its broader implementation in wastewater treatment assessment.

Graphical Abstract

1 Introduction

Wastewater treatment plants (WWTPs) function as barriers, preventing contamination from reaching the natural environment. The treatment technologies implemented in WWTPs are selected based on several factors such as the characteristics of the wastewater, the prevailing conditions, and the availability of economic resources and technical capacity to establish and operate the facilities. The performance of WWTPs is commonly evaluated using bulk metrics of effluent water quality, including total suspended solids (TSS), biological oxygen demand (BOD), and the total nitrogen and phosphorus content (McLachlan et al., 2022). However, as domestic wastewater treatment plants are not designed to handle emerging contaminants (EC) which are of relatively polar nature, these micropollutants are not completely sorbed into the activated sludge process, rendering WWTP effluents a significant source of emerging contaminants to the receiving aquatic environment (Pandey et al., 2024). Zhang et al. (2023) estimated that pollutants discharged from WWTPs contributed 33.6% to the total pollutant load in a recipient river (Zhang et al., 2023). Therefore, a comprehensive assessment of WWTP efficiency is critical for evaluating the fate of these trace contaminants and their potential to drive ecological and human health risks.

Traditional targeted analysis for evaluating wastewater treatment performance is analyte-specific, expensive, and has limitations in capturing the full spectrum of contaminants and their transformation products, which may exhibit similar toxicity to parent compounds (Petrie et al., 2015; Yadav et al., 2017). The advent of high-resolution mass spectrometry (HRMS) technologies, such as quadrupole time-of-flight (QTOF) MS and orbitrap MS, has further accelerated EC research by improving mass accuracy and resolving power (Rout et al., 2021). These techniques have represented the backbone for non-target analysis (NTA) which emerged as a powerful technique for detecting and identifying overlooked and unknown chemicals in various environmental and biological samples (Manz et al., 2023). With the aid of NTA, thousands of compounds can be detected, including unknown or unexpected micropollutants and transformation products (TPs) (Li et al., 2018; Nürenberg et al., 2019).

Taking advantage of its ability to estimate removal efficiencies for a large number of contaminants with relatively low effort, NTA has been introduced as a tool for evaluating the performance of various wastewater treatment processes, such as biological treatment, sand filtration, and activated carbon filtration (Li et al., 2018; Nürenberg et al., 2015; 2019). As an illustration for its potential, an NTA workflow, employing both positive and negative ionization modes, detected 70,631 compounds in positive mode and 14,423 in negative mode in wastewater influent, confirmed the elimination of 94.5% of these compounds after biological treatment, tentatively identified 510 chemicals and designated 191 compounds as TPs (Chen et al., 2024). However, the application of NTA is surrounded by numerous uncertainties, including its qualitative nature and the varying levels of confidence in compound identification (McCord et al., 2022; Schymanski et al., 2014).

Table 1 provides a critical comparison of the most relevant review articles related to NTA and wastewater treatment evaluation, highlighting the unique contribution of the present review. The reviewed articles fall into three categories. The first category focuses on the overall NTA workflow for aquatic samples including wastewater (Hollender et al., 2017; Menger et al., 2020; Paszkiewicz et al., 2022). These reviews provide valuable discussions of NTA methodology and include selected treatment-related applications; however, they do not place wastewater treatment performance evaluation at the core of their discussion. The second category reviews wastewater treatment technologies and contaminant fate (Prasse et al., 2015; Rout et al., 2021). Although these reviews discuss treatment technologies, they do not provide a systematic NTA-based framework for evaluating wastewater treatment performance. The third category is more closely related to NTA-based wastewater treatment evaluation, as represented by the review of Qiu et al. (2021). This review provides a structured framework for interpreting complex ultrahigh-resolution mass spectrometry (UHRMS) molecular data, but its scope is narrower than that of the present review, particularly regarding QA/QC measures, matrix effects, toxicity and risk assessment, and regulatory implications (Qiu et al., 2021). These comparisons indicate that no previous review has provided a comprehensive framework integrating NTA results for the evaluation of wastewater treatment performance.

The present review addresses this gap. It synthesizes and classifies the principal NTA-based strategies used to evaluate treatment performance, including assessing changes in chemical complexity, tracking the fate of contaminants and molecular signals throughout treatment trains, and examining shifts in physicochemical characteristics inferred from molecular information. By consolidating these approaches into a unified framework, the review demonstrates how NTA can move beyond contaminant discovery to support process evaluation, technology comparison, optimization, and decision-making.

To support the interpretation of NTA-derived results, the review further examines the wastewater-specific analytical workflow, from sampling and sample preparation to chromatographic separation, HRMS acquisition, data processing, compound identification, quality assurance and quality control practices, and matrix effect mitigation measures. In addition, it synthesizes emerging themes that are increasingly shaping the field, including prediction of removal efficiencies based on compound properties, integration of ecological and human health risk assessment into treatment evaluation, and identification of transformation products that may contribute to residual risks. By critically assessing the opportunities, limitations, and uncertainties associated with current approaches, this review provides a roadmap for future research and for the broader integration of NTA into wastewater treatment assessment and regulatory monitoring.

2 Data search methodology

Scopus was selected as the principal database for the literature search. Although reliance on a single database may raise concerns, comparative studies have provided a reasonable indication of Scopus broad journal coverage. For example, approximately 99% of the journals listed in Web of Science, a widely used database in bibliometric research, have also been reported to be indexed in Scopus (Singh et al., 2021; Wu et al., 2024).

Articles were included if they: (i) applied HRMS-based non-target analysis or suspect screening to wastewater samples, and (ii) analyzed changes in the chemical profile of samples under the effect of treatment. Articles were excluded if only analyzed influent or effluent samples independently without examining changes occurring throughout the treatment process or evaluating treatment performance. The initial search returned 327 records. After duplicate removal, 301 records remained for screening which resulted in a final set of 90 studies included in this review. The flow diagram of the literature search and study selection process is shown in Figure 1.

FIGURE 1

In addition to extracting information on NTA-based strategies for evaluating wastewater treatment performance, these articles were also examined to capture the characteristics of NTA workflow used in wastewater research. Figure 2 illustrates the geographic distribution of the included studies by country of origin. Among the 90 studies, contributions were identified from 20 countries. China accounted for the largest share of articles, while substantial research activity was also observed across Europe, East and South Asia, Türkiye and the United States. African representation was limited to studies conducted in Botswana, Eswatini and South Africa. In addition to the 90 studies included in the systematic review, supplementary literature covering relevant methodological topics—including quality control and quality assurance (QC/QA), data processing, risk assessment, and machine learning—was consulted to support the discussion.

FIGURE 2

3 NTA workflow

The non-target analysis (NTA) workflow typically consists of several key steps: (i) sampling; (ii) sample preparation; (iii) chromatographic separation (liquid chromatography (LC), two-dimensional liquid chromatography (LC × LC), gas chromatography (GC), two-dimensional gas chromatography GC × GC, and supercritical fluid chromatography (SFC)); (iv) data acquisition (using a high-resolution mass spectrometer); (v) data processing and (vi) data analysis tailored to the research question (Schulze et al., 2020).

Despite that NTA detects a broad spectrum of chemical compounds which might reach several thousand (Wiest et al., 2021); the actual range of detectable substances is limited to the “detectable domain” which is governed by the intersection of the chemical spaces covered by the techniques used at the different stages of the workflow (Black et al., 2023). Therefore, knowing the effect of these steps on the nature of detected chemicals allows one to infer the chemical boundary of the results even though the range of analyzed compounds is not known in advance (Hollender et al., 2023).

The following subsections briefly outline the various steps in the NTA workflow and highlight their domain of applicability in the context of wastewater analysis (Figure 3).

FIGURE 3

3.1 Sampling methods

The sample collection methods used in the reviewed studies included grab sampling, composite sampling, and passive sampling (Figure 4). Grab sampling is the collection of a single, discrete sample at a specific location or time. Composite sampling involves combining multiple discrete samples or using continuous automated samplers to create a single representative sample that captures temporal variation of chemical occurrence over a given period. Time-proportional and flow-proportional sampling are the two main types of composite sampling (Hollender et al., 2023). Flow-proportional composite sampling provides more accurate data for tracking organic micropollutant (OMP) concentrations and removal efficiencies in wastewater compared to time-proportional sampling (Gutierrez et al., 2024). Passive sampling adsorbs contaminants in situ by employing a suitable sorbent medium which determines the selectivity of the sampler (Hollender et al., 2023; Menger et al., 2020). After reaching equilibrium, the adsorbed chemicals are eluted by means of a solvent to be instrumentally analyzed (Sobotka et al., 2021). To ensure these sampling methods yield comparable influent-effluent data, timing between influent and effluent sampling is often adjusted based on the hydraulic retention time (HRT) to ensure that these samples are properly matched (Johnson et al., 2024; McLachlan et al., 2022; Verkh et al., 2018; Xiao et al., 2024).

FIGURE 4

Among the reviewed studies, composite sampling was reported 65% more often than grab sampling. This is likely driven by the fact that composite sampling produces more representative samples. However, operational factors can dictate the choice of sampling method. For example, composite sampling can be challenging in cases of short residence times, such as in tertiary treatment, which makes grab sampling more convenient (Verkh et al., 2018). The number of studies that reported the use of flow-proportional sampling exceeded that of the studies applying time-proportional sampling by 45%. The popularity of flow-proportional sampling can be justified by its superior performance in measuring contaminant concentration and removal compared with time-proportional sampling (Gutierrez et al., 2024). Despite its potential, the application of passive sampling in wastewater analysis is limited; it was reported in only two studies representing 2% of the studies that involved wastewater experimental analysis (Ouyang et al., 2015; Tadić et al., 2022).

Several studies have shown that different sampling methods yield inequivalent results. In a comparative study of grab sampling and passive sampling for the identification of pesticides and pharmaceuticals in wastewater, the Polar Organic Chemical Integrative Sampler (POCIS) performed better than grab sampling as it detected more antibiotics representing 6% of the total compounds. This was attributed to the higher signal intensities and the high quality of MS2 spectra achieved by passive sampling which enhanced the isotope pattern matching (Tadić et al., 2022). Verkh et al. (2018) compared composite and grab sampling of secondary effluent based on retention time, mass, nitrogen atom count (as an indicator of elemental properties), and logarithmic intensity. The results showed statistically significant differences in mass and retention time distribution, highlighting the need for caution when comparing data from these two methods. However, since this study did not evaluate the two methods at the molecular level, no preference for either method was concluded (Verkh et al., 2018).

Grab and composite sampling are the most common sampling methods. However, each of these methods has its advantages and disadvantages, and there is no single best solution that fits all purposes, as operational factors can significantly affect sampling method selection. On the other hand, passive sampling is a promising technique, but further research is encouraged to gain more insight into its practicality.

3.2 Sample preparation

A satisfactory sample pretreatment method for NTA is a method that retains the widest possible range of analytes of interest, with plausible recoveries (Ruan et al., 2023). Figure 5 presents a mechanism-based classification of the sample preparation methods employed in the NTA workflows of the reviewed studies. Table 2 provides a comparative overview of these methods, focusing on operational considerations, advantages, and limitations.

FIGURE 5

According to a recent review, the most widely applied sample preparation method for NTA is Solid phase extraction (SPE) (Hajeb et al., 2022). This predominance is evidenced by its high reported frequency in approximately 80% of the reviewed studies. Polymeric reverse-phase sorbents, characterized by a wide polarity range (Hollender et al., 2023), were the most commonly used materials in the reviewed articles. Examples of such sorbents include Oasis HLB (Reverbel et al., 2023), Strata-X (Kempińska and Kot-Wasik, 2018), Isolute ENV+ (Tisler et al., 2025), ENVI-Carb (Gollong et al., 2022), styrene-divinylbenzene (LiChrolut EN) (Ponce-Robles et al., 2018), and HR-X (Bergé et al., 2018). On the other hand, silica-based C 18 reverse phase was used when highly non-polar compounds (with octanol-water partitioning coefficient log K_OW >1) were expected (Gollong et al., 2022; Ponce-Robles et al., 2018; Wang et al., 2019). To further expand the chemical coverage to include ionic analytes, various ionof-exchange sorbents have been applied in wastewater analysis. These include weak anion exchange for strong acids (e.g., Strata-X-AW (Nihemaiti et al., 2022), Oasis WAX (Qian et al., 2021)), strong anion exchange for weak acids (e.g., Oasis MAX (Deeb et al., 2017)), weak cation exchange for strong bases (e.g., Strata-X-CW (Nihemaiti et al., 2022), Oasis WCX (Qian et al., 2021)), and strong cation exchange for weak bases (e.g., Oasis MCX (Deeb et al., 2017; Halwatura and Aga, 2023)).

Oasis HLB was the most common single sorbent applied–reported in about 40% of the reviewed studies that applied SPE. This is likely due to its popularity for extracting broad polarity range compounds (Johnson et al., 2024; Mladenov et al., 2022). The remaining 60% of papers applied combinations of sorbents in sequence or in a mixed-bed cartridge. In several comparative studies, combining sorbent materials outperformed the use of a single sorbent, likely due to their synergistic effect. For instance, out of 66 compounds (log K_OW from −1.9–5.5) extracted by tandem HLB–MCX SPE, 17 compounds were missed by HLB alone (Halwatura and Aga, 2023). Based on target analysis of wastewater samples, coupling strong anion and cation exchangers (Strata-X-A and Strata-X-C or Oasis MAX and Oasis MCX) in a sequence provided the best results. This approach achieved ≥90% recovery for all analytes with log K_OW values that ranged from −1.1 to 4.3 (at pH 7) without pH modification (Deeb and Schmidt, 2016). In another study based on non-target analysis, multilayer cartridges composed of HLB, ENV+, XCW, and XAW performed better than individual cartridges at retaining a wide range of HRMS features and were less influenced by dissolved organic matter (DOM), reducing matrix effect potential (Huynh et al., 2021).

Customized selective SPE substances have recently been developed to avoid the genericity of the previously mentioned commercial adsorbents. One example is the molecularly imprinted polymers (MIPs) synthesized from β-cyclodextrin–epichlorohydrin for steroid absorbance. Compared with conventional adsorbents, MIPs enhanced qualitative-selectivity by 30% and reduced ion suppression rate range from 45%–80% to 15%–30% (Kopperi and Riekkola, 2016).

In addition to the properties of the SPE cartridge, the determination of captured compounds is significantly affected by the sample pH, conditioning and elution solvents (Huynh et al., 2021). The solvents, eluents, eluent modifiers, and extraction additives used in the sample preparation methods are presented in Table 3. In LC-based systems, combinations of methanol and water are the most used conditioning solvents in LC based systems (reported in more than 70% of papers using SPE, e.g., (Delanka-Pedige et al., 2024a; Nihemaiti et al., 2022; Zang et al., 2022)). Dichloromethane, hexane (Wang et al., 2018), acetone (Ulucan-Altuntas et al., 2023), acetonitrile (Halwatura and Aga, 2023), have been also used for conditioning.

Methanol is the predominant eluent, used alone (Delanka-Pedige et al., 2024a; Delanka-Pedige, Young, et al., 2024; Doyle et al., 2022; Gollong et al., 2022; Hu et al., 2023; Kiss et al., 2018; Sapkota and Pariatamby, 2025; Ulucan-Altuntas et al., 2023; Wang et al., 2020; Wiest et al., 2021; Zang et al., 2022) or in combination with acetonitrile (Bergé et al., 2018; El-Deen and Shimizu, 2022; Halwatura and Aga, 2023; Parry and Young, 2016), ethyl acetate (Gago-Ferrero et al., 2020; Halwatura and Aga, 2023; Nihemaiti et al., 2022), dichloromethane (Ponce-Robles et al., 2018; Reverbel et al., 2023) and hexane (Ponce-Robles et al., 2018). Methyl tertiary butyl ether (MTBE) was also used as an eluent (Itzel et al., 2020). Ammonia and formic acid are added to adjust the acidity/basicity of the eluent specifically when ion exchange sorbents are applied (Nihemaiti et al., 2022).

In GC instruments, dichloromethane was the most used eluent with SPE; the solvent was individually used (Blum et al., 2017) or combined with acetone (Mladenov et al., 2022), acetonitrile (Blum et al., 2017), and hexane (Wang et al., 2019). Methanol was used as an eluent in a study performed by Tisler et al. (2025) on a GC-based system but the extracts were derivatized before undergoing chromatographic separation (Tisler et al., 2025). In addition to methanol and water, other solvents, such as dichloromethane, acetone (Mladenov et al., 2022) and acetonitrile (Blum et al., 2017) have been used for conditioning SPE sorbents applied with GC.

The popularity of SPE stems from its high recovery rates for compounds of interest, improved selectivity, specificity, and reproducibility, its applicability to a wide range of sample matrices, and its use of low solvent volumes during extraction steps (Deeb and Schmidt, 2016). SPE is particularly recommended for low-concentration liquid samples (Hollender et al., 2023). Additionally, efficient removal of matrix interferences is one of the commonly reported advantages of SPE (Halwatura and Aga, 2023). However, several studies have shown that matrix effects can still occur (Gutierrez et al., 2024). For example, Li et al. (2018) assessed matrix effects by comparing the peak areas of isotope-labeled standards in influent and effluent matrices. While no significant differences were found for 85% of compounds using direct injections, statistically significant differences were observed for 70% of compounds when samples were enriched using SPE (Li et al., 2018). Thus, direct injection has been proposed as a valuable alternative for sample introduction, particularly for complex or high-strength samples such as raw wastewater (Gutierrez et al., 2024). However, a disadvantage of direct injection highlighted by the work of Li et al. (2018) on non-target analysis was the occurrence of false negatives due to the relatively low concentration levels of some compounds. Using large volume injections was proposed as a solution for low concentrations despite concerns about potential matrix effect (Li et al., 2018).

Liquid-liquid extraction (LLE) is another wastewater sample preparation technique reported mainly in GC-MS systems across the inspected studies (Blum et al., 2017; Murrell and Dorman, 2021; Tian et al., 2024; Zhang et al., 2023). With dichloromethane as the main organic solvent applied, varying the pH of the extracted samples (pH 2,7, and 11) was the approach adopted by Zhang et al. (2023) to broaden the range of captured compounds in a GC-MS (Zhang et al., 2023). NaCl is added to facilitate the separation of the organic phase from the aqueous phase (Blum et al., 2017; Zhang et al., 2023) whereas Na₂SO₄ is added to dry the organic phase (Blum et al., 2017; Tian et al., 2024).

A modified QuEChERS (quick, easy, cheap, effective, rugged, and safe) method was also applied as an alternative to SPE in direct infusion-high resolution mass spectrometry (DI-HRMS) to avoid matrix suppression which was observed primarily with SPE in the negative electrospray ionization (ESI) mode. Dry freezing was used to increase QuEChERS volumetric capacity, and further clean-up using dispersive SPE (dSPE) was applied. Employing Fourier-transform ion cyclotron resonance mass spectrometry (FTICR-MS), the modified method successfully extracted a broad range of pharmaceuticals proving applicability beyond QuEChERS conventional pesticide scope (Perkons et al., 2021).

While the mechanism of the previously described pretreatment techniques focuses on extracting compounds of interest, developing strategies based on excluding matrix interferences is recommended (Hajeb et al., 2022). One example is the addition of methanol to wastewater samples at a ratio of 1:1 to precipitate protein (Sapkota and Pariatamby, 2025).

3.3 Chromatographic separation

In agreement with literature, liquid chromatography (LC) and gas chromatography (GC) are the two commonly used chromatographic methods employed with LC predominating (∼80%) across the reviewed studies. The predominance of LC is attributed to the aquatic nature of the wastewater samples (Schulze et al., 2020; Tintrop et al., 2024). In addition to the matrix nature, the physicochemical properties of the analytes are determinant factors in the selection of chromatographic method; volatile and semi-volatiles compounds are analyzed with GC, whereas LC is used for polar and less volatile compounds (García-Córcoles et al., 2019). Figure 6 shows the classes of chemicals detected by different systems as reported in the reviewed studies.

FIGURE 6

Key factors in LC approaches include: (1) The column chemistry; (2) the composition of the mobile phase; (3) pH; and (4) the type and concentration of additives (e.g., formic acid, ammonium acetate) (Souihi et al., 2023). Among the reviewed articles, reverse phase (RP) is the most popular type with a prevalence of C18 columns. This column type captures non-polar to moderately polar compounds, raising concern about overlooking more polar compounds. To enhance the detection of polar compounds several alternative techniques have been introduced including hydrophilic interaction liquid chromatography (HILIC), supercritical fluid chromatography (SFC) and sample derivatization prior to analysis with GC-based systems. The ability of hydrophilic interaction liquid chromatography (HILIC) at retaining highly polar compounds is gaining interest (Armin et al., 2025). This column retains hydrophilic compounds by using a polar stationary phase eluted by an organic mobile phase of low aqueous content (Batt et al., 2024; Gollong et al., 2022) preferentially extracted polar compounds using Multi-layer SPE composed of weak (acidic and basic) ion exchangers, and graphitized carbon black. The polar extract was analyzed via HILIC (Gollong et al., 2022). The low median m/z value of the detected features was considered as an indicator for their polar nature. Armin et al., 2025 successfully applied zwitterionic hydrophilic interaction liquid chromatography (ZIC-HILIC) – which is an advanced version of HILIC containing positively and negatively charged functional groups–to extend the polarity range beyond that captured by traditional RPLC. This was demonstrated by the high percentage of the overall features (44%) uniquely detected by ZIC-HILIC in the effluent widely known for its high polar content (Armin et al., 2025). In a comparative study by Tisler et al., 2025, supercritical fluid chromatography (SFC) successfully detected very polar compounds that were overlooked by RPLC. Furthermore, derivatization was employed in the same study to extend the coverage of two-dimensional gas chromatography GC × GC to detect highly polar and semi-volatile compounds (Tisler et al., 2025). Beyond GC systems, derivatization enhanced also the detection of highly polar compounds in LC systems (Manasfi et al., 2023).

As an emerging configuration, two-dimensional liquid chromatography (LC × LC) is recognized for its ability to increase the chromatographic separation peak capacity (Beschnitt et al., 2022), that is the estimated maximum number of equal-height peaks that can be fit side-by-side at equal resolution (typically 1.0) within a given separation space (Stoll and Carr, 2017). Peak capacity of LC × LC ranged from hundreds to 10,000, while the maximum peak capacity achieved by one dimension LC is 1,000, with a typical range of 100–500. High peak capacity reduces the potential of co-elution of compounds to the ionization source, which minimizes matrix effect (Kronik et al., 2024). Modes of operation in LC × LC systems range from the heartcutting technique where a single or multiple fractions of the first dimension effluent are collected and injected into the second dimension column, to the comprehensive LC × LC in which all fractions undergo separation in the second dimension (Duarte et al., 2023; Ouyang et al., 2015; Pirok et al., 2019; Purschke et al., 2020). In a study comparing the performance of LC-HRMs and LC x LC-HRMS in the evaluation of pharmaceutical manufacturing wastewater treatment, LC × LC-HRMS yielded higher spectral quality and twice the number of detected compound peaks relative to LC-HRMS (Saint Germain et al., 2022). Despite these advantages, the application of two-dimensional liquid LC × LC chromatography in environmental analysis is limited compared to GC × GC (Duarte et al., 2023; Ouyang et al., 2015). The limited application of LC × LC is attributed to challenges in mobile phases compatibility between LC interfaced dimensions, the risk of first dimension effluent undersampling, the need to optimize the selection and matching of columns and multifaceted separation conditions, and the scarcity of equipment and data analysis software for LC × LC (Duarte et al., 2023). However, advancements are being made to address these challenges, rendering LC × LC a promising technique for environmental analysis. For instance (Kronik et al., 2024), has recently introduced the pulsed elution-LC × LC as a strategy to decouple the two dimensions and increase the system flexibility.

As for GC-MS methods, a couple of advantages distinguish them from LC-based methods in the field of NTA. These include (1) the highly reproducible fragmentation pattern and retention time; (2) the availability of several libraries and database associated with GC-MS such as National Institute of Standards and Technology (NIST) library which facilitate the identification of unknown compounds (Zhang et al., 2023); (3) efficient separation; (4) lower operational costs; and (5) low matrix effect (Devers et al., 2025). Compounds amenable to GC are non-polar to mid-polar in terms of polarity (Devers et al., 2025) and volatile to semi-volatile in terms of volatility (Wang et al., 2019; Zhang et al., 2023). The classes of detected chemicals that were reported in the reviewed articles addressing the application of one dimension GC methods in NTA include petroleum related hydrocarbons and their derivatives (Zhang et al., 2023), Pharmaceuticals and Personal Care Products (PPCPs), Benzotriazole corrosion inhibitors (Murrell and Dorman, 2021), PPCPs, plasticizers, flame retardants, insecticide TPs, pesticides, Polycyclic Aromatic Hydrocarbons (PAHs), Steroids & Hormones (Endocrine-disrupting chemicals) EDCs, volatile organic compounds (VOCs), Persistent Organic Pollutants (POPs) and chemicals with other uses (Wang et al., 2019; Zhang et al., 2023). GC received samples dissolved in hexane (Zhang et al., 2023) or hexane-dichloromethane mixture (Tian et al., 2024; Wang et al., 2019). The most widely applied column was the low-polarity (5%-phenyl)-methylpolysiloxane phase and its equivalents (Blum et al., 2017; 2019; Johnson et al., 2024; Lopez-Herguedas et al., 2024; Mladenov et al., 2022; Mok et al., 2023; Murrell and Dorman, 2021; Tan et al., 2025; Tian et al., 2024; Tintrop et al., 2024; Tisler et al., 2025; Wang et al., 2019; Yang et al., 2025; Zhang et al., 2023).

On the other hand, the use of GC × GC is reported in almost 40% of the reviewed articles that applied GC-based methods, indicating its popularity in wastewater NTA (Blum et al., 2017; 2019; Johnson et al., 2024; Mladenov et al., 2022; Murrell and Dorman, 2021; Tisler et al., 2025). GC × GC/HRMS provides high resolution and sensitivity compared to GC-MS and has been identified as highly complementary to LC-based NTA approaches (Blum et al., 2019; Johnson et al., 2024). GC × GC-HRMS has successfully detected small molecular weight compounds with high water solubility, highlighting its suitability for wastewater analysis (Mladenov et al., 2022). Furthermore, derivatization improved the capabilities of GC × GC-HRMS to detect very polar and semi-volatile compounds with median Henry’s law constant of 3 × 10⁻⁸ kPa m³/mol (Tisler et al., 2025). To fully exploit the capabilities of GC × GC, the two dimensions of the GC × GC system must operate in different modes (Devers et al., 2025). Using a nonpolar column in the first dimension followed by polar column in the second dimension is the scheme adopted in the majority of the papers that addressed the application of GC × GC (Blum et al., 2019; Johnson et al., 2024; Mladenov et al., 2022; Murrell and Dorman, 2021; Tisler et al., 2025). Only one paper applied the reversed combination: polar followed by nonpolar column as they suspected high hydrocarbon content in the sample (Blum et al., 2017).

3.4 Mass spectrometric analysis

High-resolution instruments, such as Orbitrap and time-of-flight (TOF) mass spectrometers and their variants, are the most common HRMS techniques for NTA (Schulze et al., 2020). Mass resolution is defined as m/Δm—the ratio of the measured mass (m/z) to the peak width (Δm) measured at a defined height; the most widely adopted convention is to measure (Δm) at half the maximum peak height, known as the Full Width at Half Maximum (FWHM) (Murray et al., 2013). Orbitrap systems can reach a mass resolution of up to 1,000,000, which exceeds that of TOF instruments at 80,000, indicating that Orbitrap instruments provide higher mass resolution than TOF instruments (Menger et al., 2020; Schulze et al., 2020). However, the application of QTOF-MS in the reviewed articles was double that of Orbitrap-MS, emphasizing the popularity of QTOF-MS.

Another mass spectrometry technique employed for wastewater NTA is Fourier-transform ion cyclotron resonance mass spectrometry (FTICR). It is characterized by the highest resolving power but its intricacy and expensiveness has limited its application in analyzing emergent contaminants (Menger et al., 2020). Low acquisition rate renders FTICR coupling with chromatographic separation uncommon. This limitation hinders data acquisition of retention time (RT) which is a critical dimension for compound identification (Ghaste et al., 2016; Schulze et al., 2020). However, there are some instances where FTICR was successfully used with LC to separate isomers (Jennings et al., 2025). Perkons et al. (2021) highlighted FTICR compatibility with direct infusion (DI) and flow injection (FI). Their study showed the rapidness of DI-FTICR-based NTA method, its significant reduced post-processing steps compared to LC-HRMS methods and its ability to analyze compounds across a wide polarity range. Its suitability for screening substances with at least two different heteroatoms and its ability to achieve higher selectivity to halogenated compounds were also noted. However, the drawbacks reported in the study included a higher susceptibility to false positive rate, particularly for compounds with numerous isomers, lower sensitivity compared to LC-HRMS systems, and limited accuracy in quantification (Perkons et al., 2021).

Unlike FTICR technique, which might lack the RT dimension, ion mobility–high resolution mass spectrometry (IM-HRMS) devices help in providing the collision cross section (CCS) as an additional separation dimension. CCS is calculated based on the drift time measured by the ion mobility separation (IMS). This technique is applied successfully in discriminating isobars, isomers and even enantiomers (Menger et al., 2020). Instrument independence and matrix–effect freedom are two characteristics that increase the confidence in CCS-based compound identification (Celma et al., 2020; Hinnenkamp et al., 2022).

Prior to being detected by the mass analyzers described earlier, molecules undergo ionization using various techniques (Schulze et al., 2020). Electrospray Ionization (ESI) coupled with LC-HRMS systems was the most widely used ionization in the reviewed articles. Most of the studies reported the dual utilization of positive and negative ionization modes to achieve broader coverage of organic micropollutants (OMP) with varying chemical properties. Positive ionization mode detected more compounds than negative ionization mode in wastewater samples (Chen et al., 2024; Gutierrez et al., 2024; Kempińska and Kot-Wasik, 2018; Kiss et al., 2018). For example, the percentage of compounds detected by the negative ionization mode was <0.5% of a set of targeted compounds during optimization of an LC-HRMS method for wastewater analysis (Huidobro-López et al., 2023). Positive ionization mode efficiently ionizes basic compounds such as those containing amino or amide groups. Negative ionization mode targets acidic compounds with functional groups like carboxy, ester, or hydroxyl group. It has been also applied mainly for suspect per- and polyfluoroalkyl substances (PFAS) detection (Halwatura and Aga, 2023; Wang et al., 2018). Compounds containing multifunctional groups respond in both modes (Kempińska and Kot-Wasik, 2018). Beyond chemical functionality, a physical trend has been observed where compounds ionized in negative mode tend to have larger m/z values than those in positive mode (Kiss et al., 2018).

In addition to ESI which is the most common in wastewater analysis, other ionization techniques compatible with LC-MS include the atmospheric pressure chemical ionization (APCI) and atmospheric pressure photoionization (APPI). APCI and APPI are characterized by their ability to detect less polar compounds, usually not ionized by ESI (Menger et al., 2020).

For GC-MS systems, electronic ionization (EI) is regarded as the golden standard (Tintrop et al., 2024). This is evidenced by its popularity in the reviewed GC based articles (Blum et al., 2017; 2019; Johnson et al., 2024; Lopez-Herguedas et al., 2024; Mladenov et al., 2022; Murrell and Dorman, 2021; Tian et al., 2024; Tisler et al., 2025; Wang et al., 2019; Zhang et al., 2023). Despite its highly-reproducible fragmentation, which aids in library matching, EI strong in-source fragmentation often leads to the loss of molecular ions, hindering the identification of unknown compounds. To address this limitation, hybrid approaches combining EI with chemical ionization (CI) – a softer ionization technique that keeps the molecular ion intact–have been employed to improve compound identification and expand chemical coverage (Tintrop et al., 2024).

3.5 MS data acquisition

Data acquisition in non-target analysis is performed mainly via three key methods: full scan, data independent acquisition (DIA) or data dependent acquisition modes (DDA) (Defossez et al., 2023) (DDA is known also as information dependent acquisition (IDA) (Yang et al., 2023)). Full scan is the simplest data acquisition method that produces MS1 data including ions of the molecular species, adducts and insource fragments (Defossez et al., 2023). On the other hand, DIA and DDA modes were the most applied in reviewed studies (Gutierrez et al., 2024). Both methods acquire MS2 data following MS1 scan (Ledesma-Escobar et al., 2023), providing the complete datasets required for compound identification and qualitative analysis (Ruan et al., 2023). Each of these strategies has its unique advantages and disadvantages. DDA allows only the fragmentation of MS signals that fulfil predefined conditions (Defossez et al., 2023). An example is the selection of top N most intense ions (Ledesma-Escobar et al., 2023), where N ranged from 3 to 10 in the inspected studies. Despite the limited number of fragmented precursors, DDA results in a high-quality MS2 spectra (Ruan et al., 2023). By contrast, DIA is characterized by its wide mass coverage as all precursor ions are fragmented, which makes it difficult to link fragments to their parent ions (Defossez et al., 2023). Some studies combined DIA and DDA (Wang A.-G. et al., 2025). This is apparently to make use of the advantages of both strategies (Ruan et al., 2023).

3.6 Data processing

The conventional NTA data processing begins with peak picking or feature extraction. A feature is generally defined as a unique combination of exact mass-to-charge ratio (m/z) and retention time, associated with an intensity value (Hollender et al., 2023). In cases where online centroiding is not enabled during data acquisition, offline centroiding may need to be performed prior to feature extraction. Conversion of data from vendor format to open format may also be needed if open access software tools are used for data processing. Following peak picking, the general NTA data processing workflow includes, componentization where related peaks belonging to the same compounds such as isotopes and adducts are grouped together; and blank subtraction, where peaks belonging to matrix, contamination, or background are removed from the sample data, and finally peak alignment across samples (Blum et al., 2017; Hohrenk et al., 2019). Each step of the NTA data processing workflow requires a careful selection of input parameters to ensure the validity of results. The types of parameters might vary based on the algorithm or software tool. However, parameters like mass tolerance, signal-to-noise (S/N) ratio, and minimum peak intensity are common across feature extraction algorithms. The large number of required parameters (about 10–15 parameters) reflects the complexity of data processing and highlights the need for high level of expertise (Lennon et al., 2024). Additionally, results inconsistency due to variations in the software underlying algorithms has been an issue of concern (Hohrenk et al., 2020). Representative examples of conventional feature extraction algorithms are centWave in XCMS and MZmine 2, the Automated Data Analysis Pipeline (ADAP) in MZmine three and 4, and the linear-weighted moving average algorithm in MS-DIAL. More modern statistically based algorithms are Asari and Bayesian 2D (Padilla-González et al., 2026; Renner and Reuschenbach, 2023).

Diverse software tools are employed in processing the extensive NTA data. Examples of commercial software supplied by instrument manufacturers are Compound Discoverer (Thermo Fisher Scientific) (Li et al., 2018); Unknown Analysis (Blum et al., 2019) and MassHunter Profinder (El-Deen and Shimizu, 2022) from Agilent Technologies; UNIFI (Waters) (Gros et al., 2017); and DataAnalysis (Bruker Daltonics) (Bergé et al., 2018). Some tools are reported with specific applications such as ChromaTOF (LECO Corp) (Blum et al., 2019) for GC × GC; GC-Image, LLC for GC × GC (Tisler et al., 2025) and LC × LC (Ouyang et al., 2015); and IM-MS Browser and IM-MS reprocessor (Agilent) for systems coupled with ion mobility separation (Blum et al., 2019). On the other hand, a collection of open-access software tools is applied in numerous reviewed studies (e.g., PyHRMS (Chen et al., 2024), MZmine 3 (Delanka-Pedige, Young, et al., 2024), MS-DIAL (Tisler et al., 2025))

It is worth noting that data processing methods beyond peak picking-based approach described earlier have been developed (Lennon et al., 2024; Vosough et al., 2024). For instance (Hohrenk et al., 2019), combined the data compression method Regions of Interest (ROI) and Multivariate Curve Resolution-Alternating Least Squares (MCR-ALS) to assess wastewater samples pre- and post-ozonation treatment. Data size reduction, tracing of analytes without alignment, and overcoming spectral interferences and overlapped organic micropollutants (OMPs) elution profiles are the advantages exhibited by this strategy. Multivariate Curve Resolution based methods were also successfully applied to process LC × LC data widely known for the limited tools of data processing (Schneide and Munk Kronik, 2025).

In the context of NTA applications in wastewater treatment assessment, extracted features are further analyzed following two approaches. The first approach focuses on the holistic processing of features applying statistics and chemometric methods. This approach enables trend monitoring and process evaluation (Minkus et al., 2022b; Minkus et al., 2022a). Strategies for evaluating wastewater treatment based on this approach are discussed in section “4. NTA-based wastewater treatment evaluation”. The second approach focuses on the Identification of compounds (Minkus et al., 2022b; Minkus et al., 2022a). Typical strategies for identification of NTA-based features include (1) matching the spectra of the unknown features with experimental or in silico spectra (generated to compensate for lacking experimental spectra) in spectra libraries; (2) predicting structural information such as molecular formula and fingerprints from experimental MS2 spectra; (3) similar structure search for assisting in annotating missing structures in databases; (4) the emerging in silico structure generation based on MS2 data (Hupatz et al., 2024; Milman, 2015). In general, identification tasks comprise several components including software tools, spectral and structural libraries and databases, and similarity metrics (Hupatz et al., 2024).

The capabilities of software tools and algorithms used for NTA data processing vary, as not all of them can be used for both data processing and identification workflow. Some tools have a wide range of applications while others are task-specific (Canchola et al., 2025). MetFrag for predicting fragmentation data (Xiao et al., 2024), SIRIUS for assigning molecular formulas, CSI: FingerID for generating candidate structures (Delanka-Pedige, Young, et al., 2024), HaloSeeker for identifying halogenated compounds (Tian et al., 2024) are examples of software tools used in the identification workflow. Following compound identification, hierarchical confidence levels are assigned to identified compounds to reflect the certainty of their structural identification (Alygizakis et al., 2023). These levels range from Level 5 (exact mass), Level 4 (unequivocal molecular formula), Level 3 (tentative structure), Level 2 (probable structure) through to Level 1 (confirmed identification) (Schymanski et al., 2014). Recently, these confidence levels have been updated to account for the inclusion of CCS value derived by IMS and introduced as an additional dimension supporting compounds identification (Celma et al., 2020).

3.6.1 Machine learning-assisted data processing

NTA data processing algorithms generally fall into two categories: traditional, rule-based algorithms, which rely on mathematical criteria and threshold parameters (Padilla-González et al., 2026; Renner and Reuschenbach, 2023), and data-driven methods which use machine learning models capable of handling complex large datasets (Padilla-González et al., 2026). Therefore, the implementation of machine learning (ML) in the various data processing stages, particularly in feature annotation, compound prioritization, and transformation product prediction, is expected to amplify the capabilities of NTA.

In feature extraction, several deep learning-based algorithms have been introduced to improve detection accuracy compared to conventional algorithms (see “3.6 Data Processing”). Examples of ML-based feature extraction algorithms are Peakonly, a deep learning algorithm based on convolutional neural networks (CNNs) that detects, integrates, and filters out low-intensity noise peaks directly from raw LC-MS data (Melnikov et al., 2020), NeatMS, a CNN based model that reduce the number of false peaks among a previously extracted features (Gloaguen et al., 2022), and peakDetective that classifies detected peaks into true chemical signals or noise artifacts with high accuracy employing unsupervised autoencoder for dimensionality reduction with an active learning classifier that require a relatively small training dataset (Stancliffe and Patti, 2023). A concern surrounding these emerging algorithms is that they were originally developed for metabolomic data processing, which necessitates cautious application to environmental datasets and may require further tuning.

Artificial Intelligence (AI) has also transformed compound annotation by improving both structure-to-spectrum and spectrum-to-structure prediction (Jin et al., 2025; Liu et al., 2026). Traditional mass spectral prediction tools, such as MetFrag and older versions of CFM-ID, rely on rule-based fragmentation (Liu et al., 2026). A key shortcoming of traditional models is the generation of an excessive number of candidate fragments, which degrades the accuracy of spectral matching (Hupatz et al., 2024; Liu et al., 2026). To improve performance, machine learning (ML)-based models have been introduced. Recent examples include the graph neural network FIORA (Nowatzky et al., 2025), the transformer-based model ICEBERG (Goldman et al., 2024) and 3DMolMS, a deep neural network that utilizes 3D compound conformations (Hong et al., 2023).

For spectrum-to-structure prediction, SIRIUS CSI:FingerID is a representative example. It combines fragmentation trees with a support vector machine (SVM) to predict molecular fingerprints (Canchola et al., 2025). More recently, de novo approaches based on generative AI have been introduced (Jin et al., 2025; Liu et al., 2026). Unlike database-dependent methods, generative models predict molecular structures directly from MS2 spectra, enabling the identification of emerging contaminants and transformation products that are absent from existing databases. Deep learning neural networks such as recurrent neural networks (RNN), variational autoencoders (VAE), and transformers forms the bases of generative model’s architecture (Hupatz et al., 2024; Liu et al., 2026). Examples of these models are MS2Mol (Butler et al., 2023), Mass2SMILES (Elser et al., 2023) and DeePFAS (Wang H. et al., 2025).

Beyond compound annotation, ML is increasingly being applied to compound prioritization. Such prioritization strategies helps in supporting compound identification and facilitating downstream processing (Hupatz et al., 2024; Jin et al., 2025; Lin et al., 2026; Liu et al., 2026). Learning structure-property relationships to predict orthogonal chemical properties, such as Retention Time (RT), CCS (Song et al., 2023) and bioactivity (Wang et al., 2024) reduces susceptible false positives. To make these predictions, models typically utilize inputs such as molecular descriptors, fingerprints, structures, and operational conditions, employing algorithms that range from classical machine learning to deep neural networks (Lin et al., 2026; Song et al., 2023).

Another rapidly advancing application is the prediction and identification of biotic and abiotic transformation products. BioTransformer 3.0 predicts human metabolites, and microbial and environmental transformation products based on a combination of machine learning and rule-based biotransformation (Wishart et al., 2022). The recent transformer-based models, TP-Transformer, predicts both transformation products and their degradation pathways from chemical structures (Dai et al., 2025).

To fully exploit the potential of ML-assisted NTA, several challenges have to be tackled, including the requirement of large amounts of high-quality training data, bias toward the chemical space represented in the training data, and poor generalizability across instruments and laboratories. Multiple innovative approaches have been proposed to overcome these bottlenecks. Generative AI models can augment existing datasets by generating realistic synthetic data, thereby mitigating data scarcity and reducing biases associated with limited training datasets. Multimodal learning, which integrates complementary information such as MS2 spectra, retention time (RT), collision cross section (CCS), molecular structures, and physicochemical properties, can improve model robustness and generalizability across diverse analytical scenarios. Transfer learning enables pre-trained models to be fine-tuned for specific applications, allowing them to be adapted to different analytical platforms and laboratory instruments while maintaining predictive performance. Transfer learning also reduces the need for the large datasets required to train deep learning models from scratch. Representation learning methods based on molecular sequences or graph representations, when combined with self-supervised learning, enable more accurate prediction by learning informative molecular features directly from unlabeled data, often outperforming traditional task-specific molecular fingerprints and physicochemical descriptors (Jin et al., 2025; Liu et al., 2026).

3.7 Quality control and quality assurance

The key objective of NTA QA and QC protocols is to reduce false positives (i.e., mistakenly reporting compounds which are absent as present), and false negatives (i.e., falsely reporting compounds which are present in the sample as absent). These errors can occur at any stage of the NTA workflow (Hollender et al., 2023). In the reviewed studies, mass calibrations have been undertaken at an early stage of the analysis to detect and correct any deviation in the mass measurement using stable and well-characterized reference masses recognized for their stability (Han et al., 2024). Internal standards (IS), predominated by isotopically labeled ones, are commonly used in QA/QC practices. These practices include the estimation of recoveries at the sample preparation step (Guy et al., 2024), matrix effect (Gollong et al., 2022), mass accuracy, retention times, sensitivity, instrument performance (Brunelle et al., 2024; Reverbel et al., 2023) and evaluating data processing in determining false positives and false negatives (Sobus et al., 2025). Blank samples are consistently used for identifying contamination from sample preparation or instrumentation (Hollender et al., 2023), subsequent filtering out false positives (Perkons et al., 2021; Reverbel et al., 2023) and eliminating carry-over (Manasfi et al., 2023). A pooled sample composed of a mixture of different samples is a representative quality control sample (Kiss et al., 2018). It is injected at the beginning of an analytical sequence to ensure the column is properly equilibrated before analyzing actual samples, applied to evaluate the repeatability (Schollée et al., 2021) and stability (Guy et al., 2024) of the analytical method, and used to evaluate the false positives and false negatives (Sobus et al., 2025). In addition to QA/QC samples, other practices such as randomized analytical sequences used to offset systematic errors caused by carry over, and replicating samples used for assessing method repeatability, filtering out susceptible false positives and acquiring statistically significant data are frequently reported (Doyle et al., 2022; Gutierrez et al., 2024; Hollender et al., 2023; Sobus et al., 2025).

The main concern of the current QA/QC practices is that they based on those of target analysis, therefore, efforts have been recently directed to develop specific QA/QC protocols to account for the different nature of NTA and its complex methods including analysis and data processing, and to ensure harmonized practices (Lennon et al., 2024; Schulze et al., 2020; Sobus et al., 2025; Vosough et al., 2024).

4 NTA-BASED wastewater treatment evaluation

It is recognized that process evaluation approaches that do not focus on extensive feature identification minimize effort (Bader et al., 2016). This section discusses such approaches using NTA data, along with their applicable statistical methods and visualization tools (Figure 7). It is worth noting that lack of required calibration curves or standards for all concerned compounds, and the simplified quality assurance requirements further facilitate the analytical effort for NTA-generated data (McLachlan et al., 2022).

FIGURE 7

4.1 Tracking changes in the total number of features and overall peak areas

The most straightforward NTA-based performance evaluation strategy is to track changes in the total number of features detected after the corresponding treatment (Nováková et al., 2023). This approach provides more comprehensive information without the need for laboriously identifying each compound or risking compound omission (Qian et al., 2021). The decrease in the number of features indicates a reduction of chemical complexity of the sample (Nürenberg et al., 2019). Hence, the greater the decrease in feature number, the more efficient the treatment process is (Johnson et al., 2024). In this case the removal rate can be quantified as the percentage change in feature numbers relative to the number of features in the influent (Kiss et al., 2018; Nürenberg et al., 2019). Similarly, treatment efficiency can be estimated based on changes in features abundance, as measured by total or average peak area (Chen et al., 2024; Verkh et al., 2018; Zhang et al., 2023). Despite their simplicity, evaluation methods based on total compound numbers (and peak areas) remain valuable for showing the trend of the bulk of compounds during different treatment steps in data-limited contexts. Although these approaches do not identify specific degraded or generated compounds, these approaches can indicate general degradation patterns throughout the treatment process (Ponce-Robles et al., 2018). However, the dependence of the feature numbers on data processing parameters during peak picking and alignment is a concern raised against these approaches (Kiss et al., 2018). Johnson et al. (2024) applied both approaches to evaluate the performance of aerobic (AeMBR) and anaerobic (AnMBR) membrane bioreactors fed with a shared inflow of municipal wastewater. The study compared the average number of chromatographic features in the permeates of aerobic (AeMBR) and anaerobic (AnMBR) membrane reactors across several treatment cycles. The AeMBR treatment demonstrated superior performance at the complete removal of contaminants compared to AnMBR. This result was confirmed by the significant reduction in total peak area of the features in the effluent of AeMBR compared to AnMBR relative to the influent (Johnson et al., 2024). Another example is illustrated in Figure 8. It displays the overall number of detectable chemical features during UV/H₂O₂ and UV/peroxydisulfate (PDS) treatment of wastewater effluent at increasing irradiation doses. Both processes produced broadly similar feature numbers at low and intermediate doses, indicating limited changes in the overall detectable chemical profile under these conditions. At the highest irradiation dose, however, feature numbers declined in both ionization modes, with a somewhat greater decrease observed for UV/PDS, particularly in ESI⁺ mode. These results suggest that high-dose treatment decreased the number of detectable features and that UV/PDS may have exerted a slightly stronger effect than UV/H₂O₂ under the most intensive treatment conditions (Xu et al., 2024). When sufficient data are available to classify features (e.g., by elemental composition (Zang et al., 2022), chemical structure (Zhang et al., 2023), application (Wang et al., 2019), or extraction method amenability (Ponce-Robles et al., 2018)), changes in total number and abundance of features can be tracked at the class-level. This approach provides deeper insight on how the treatment mechanism correlates with the compound nature.

FIGURE 8

At low identification levels, where features are characterized by mass-to-charge ratio and retention time, multivariate chemometric tools can be utilized for exploratory data analysis and visualizing the effect of the treatment based on the profile of sample features. Common examples include unsupervised methods such as principal component analysis (PCA), and supervised methods such as partial least-squares-discriminant analysis (PLS-DA) and its advanced variant Orthogonal Projections to Latent Structures Discriminant Analysis (OPLS-DA). Additionally, the information derived from multivariate methods can be used effectively for prioritization of features for further identification (Bergé et al., 2018; Hohrenk et al., 2019; Nihemaiti et al., 2022; Ponce-Robles et al., 2018; Schollée et al., 2016). PCA reduces the dimensionality of samples so that they can be described in principal components (PC) of much fewer dimensions (PC1, PC2, PC3,.etc.), while retaining most of their characteristics (Ponce-Robles et al., 2018). These components are linear combinations of the original dimensions, which makes it easy to detect trends and patterns (Kiss et al., 2018). In a study assessing the performance of integrated coagulation/flocculation and photo-Fenton treatments on cork boiling wastewater, the two-dimensional PCA score plot based on sample peak area (Figure 9a) distinguished between the raw and treated wastewater samples and differentiated between samples subjected to different treatments, indicating the impact of the treatment on the sample characteristics. Moreover, features concentrated in the center of the PCA loading plot in Figure 9b represent masses common to all sample types and therefore have not undergone significant change. In contrast, outliers—the 17 marked features in Figure 9b—are characteristic of certain samples and, hence, might exhibit a degrading (Figure 9c) or transformation (Figure 9d) trend, triggering their further identification (Ponce-Robles et al., 2018). Similarly, Nihemaiti et al. (2022) applied OPLS-DA and its associated s-plot which distinctly clustered untreated and treated features and allowed prioritization of the discriminant features for each sample group (before and after Performic Acid oxidation) for further identification (Nihemaiti et al., 2022).

FIGURE 9

4.2 Investigating the Fate of Individual Features

Removal efficiency (RE) and fold change (FC) are two equivalent metrics used to evaluate wastewater treatment performance at the feature level. RE (Equation 1) is the ratio of the amount of change in a feature area (intensity) resulting from the treatment to the area of that feature in the influent (El-Deen and Shimizu, 2022)

FC (Equation 2) is the ratio of the feature area after treatment to its area before treatment (Bader et al., 2017) (sometimes it is referred to as breakthrough (B) (McLachlan et al., 2022)):

The relationship between RE and FC is given by Equation 3 (McLachlan et al., 2022):

FC is often expressed on the logarithmic scale (Doyle et al., 2022). The peak area of a non-target feature acts as a surrogate for absolute concentration. In some studies, removal efficiencies for tentatively identified compounds were calculated using semi-quantified concentrations, rather than peak areas. These concentrations were estimated by comparing their responses to those of similar compounds for which standards were available (Sepman et al., 2023; Xiao et al., 2024). For example, Zang et al. (2022) based the estimation of removal efficiencies on closely eluting surrogates amenable to same ionization mode (Zang et al., 2022).

FIGURE 10

The common features group can be further classified into decreasing, consistent, or increasing categories based on changes in feature intensities (Bader et al., 2016). Furthermore, the definition of the eliminated features category may be expanded to include significantly decreased features. Similarly, highly increased features and newly formed features could be grouped together. An example is the five-level categorization system adopted by Bader et al. (2017), where elimination is assigned for 0.00 ≤ FC < 0.20, decrease for 0.20 ≤ FC < 0.50, consistency for 0.50 ≤ FC ≤ 2.00, increase for 2.00 < FC ≤ 5.00 and formation for 5.00 < FC ≤ (Bader et al., 2017). It is worth noting that the range of the consistency category is widened to compensate for the matrix effect (Peter et al., 2019). The performance evaluation can be based on examining the proportions of compounds belonging to each of these categories (Delanka-Pedige et al., 2024a). A treatment technology is considered more efficient if it tends to achieve high removal efficiency, which means having more eliminated and decreased features and fewer increased and formed ones (Bader et al., 2016). The removal categorization method is more illustrative compared to the previously described approach based on the comparison of the total number of features in a sample without taking into consideration the various fates of the features (Nováková et al., 2023). In addition, this approach highlights the advantage of NTA over target analysis by accounting for the formation of transformation products. The categorization approach was followed in several studies across the reviewed articles (Bader et al., 2016; Delanka-Pedige et al., 2024a; Tisler, Tüchsen, et al., 2022; Xu et al., 2024). Additionally, comparing the area of the features in each removal category (Schollée et al., 2021), and adding the qualitative dimension by classifying tentatively identified features in each category based on their chemical structure or application are some variations of the removal category approach (Delanka-Pedige et al., 2024a; Doyle et al., 2022; Zang et al., 2022; Zhang et al., 2023).

Another approach is to use the median (or the average value) RE or FC as a single measure for a set of features to characterize the performance of a treatment process. Based on this method, the lower average FC (or higher average RE) is, the more efficient the treatment process is (Doyle et al., 2022; Gollong et al., 2022; McLachlan et al., 2022; Xu et al., 2024).

However, Nürenberg et al. (2019) highlighted the risk of overestimating the true removal efficiency of emerging contaminants based on the average removal of non-target features in biological treatment. The reason is that the presence of a vast number of easily degradable biomolecules might skew the removal efficiency to higher values not reflecting the true removal efficiency of emerging contaminants. Nevertheless, to solve this issue, setting the range of MS method between 100–1,200 Da to exclude highly polar biomolecules such as sugar as well as large molecules such as lipids or proteins was suggested. Additionally, they found that the average RE calculated for a subset of features where 50% of the features have an average RE > 99.99% can act as a surrogate for the entire set of features. In an attempt to link the feature number-based evaluation and the average removal efficiency of a treatment process (RE), Nürenberg et al. (2019) found a strong correlation between the proportion of the number of the eliminated features in the influent and RE. Consequently, using the average removal efficiency or the proportion of the number of the eliminated features in the influent as a basis for ranking the performance of the wastewater treatment plant was demonstrated to be reliable and provide equivalent results (Nürenberg et al., 2019).

Expressing feature removals in a cumulative or aggregative manner provides another means of evaluating and comparing treatment performance. That is to report the percentage of compounds (target) or features (nontarget) removed to specified performance levels (e.g., > 0% or >50%), based on the distribution of features across removal levels. For example, an ultraviolet only (UV-only) treatment, in which 8.3% of target compounds and 2.2% of non-target compounds demonstrated removals above 50%, was judged less effective than the optimum hybrid UV/H₂O₂, in which the percentages of compounds that achieved the same removal level were 100% of target compounds and 74% of non-target features (Parry and Young, 2016).

Besides boxplots (Figure 11), there are several visualization tools employing FCs to study the impact of treatment on the entire community of features. One of these tools is the volcano plot (Figure 12). Log₂ FCs for individual features are plotted against their -log10 (p values). P values resulted from testing the significance of the feature intensity change under the effect of the treatment. Based on their p values, significantly changed features are differentiated from those who are not. Features undergoing significant changes are further divided into increasing and decreasing features. Examining these patterns gives an overall view of the treatment process performance (Doyle et al., 2022).

FIGURE 11

FIGURE 12

Scatter plots (Figure 13), where features FCs under one process are represented on one axis and the features FCs under another process are represented on the other axis, serve as an effective tool for evaluating the relative performance of the two treatment processes (Parry and Young, 2016; Xu et al., 2024). The features with data above the 1:1 line are more likely to be removed by the process represented by the y-axis, whereas features whose data are below this line are more prone to removal by the process represented by the x-axis. An even distribution of points across the 1:1 line indicates equivalent overall performance under the two conditions (Parry and Young, 2016; Peter et al., 2019).

FIGURE 13

PCA is also employed to evaluate treatment facilities. McLachlan et al., 2022 performed PCA on between the measured log-fold change (log FC) of each chemical and its average log FC across all wastewater treatment plants (WWTPs), rather than on features intensities as explained in section “4.1. Tracking Changes in the Total Number of Features and Overall Peak Areas”. The resulting score plot clustered the wastewater treatment plant into three groups according to similarity in performance (Figure 14) (McLachlan et al., 2022).

FIGURE 14

Hierarchical clustering analysis (HCA) is an unsupervised multivariate method, like PCA, that is used to track changes in peak intensity at the feature level. Hierarchical clustered heatmaps based on the change of feature intensities are utilized to visualize which features within the clusters were eliminated, partially removed, or newly formed (Johnson et al., 2024; Rodriguez-Otero et al., 2026). This approach allowed for the comparison of removal between aerobic and anaerobic membrane bioreactors treating municipal wastewater (Figure 15). Schollée et al. (2018) relied on HCA to assign clustered features to 11 trends describing their expected fates in a WWTP, which facilitated comparing ozonation-equipped WWTPs with different post treatment configurations (Schollée et al., 2018). In a subsequent study, Schollée et al. (2021) developed an alternative “(1,0,-1) – barcoding” method based on feature intensity normalization across samples representing the investigated treatments. This method enabled the automation of feature trend assignment, outperforming HCA. It further guided the classification of features detected in the effluent of each treatment step into (i) those carried over from previous steps and (ii) newly transformed features, forming a basis for comparing treatment step performance.

FIGURE 15

Qiu et al. (2021), Qiu et al. (2022) introduced Venn diagram-based models that can be used to track the fate of HRMS-detected features across wastewater treatment stages. Three models were discussed: a 2-set Venn diagram that uses two sampling points and evaluates one treatment stage; a 3-set diagram that uses three sampling points and evaluates two consecutive stages (Qiu et al., 2021); and a 4-set diagram that uses four sampling points and evaluates three consecutive stages (Qiu et al., 2022). The number of Venn intersection regions is related to the number of sets (n) according to 2ⁿ - 1. These regions enable distinguishing influent-derived features removed at different stages from features formed in the subsequent units, highlighting the performance of the intermediate steps in a treatment plant. In addition to the number of units, these models can account for external recirculation and influent bypass streams, which require configuration-specific interpretation of the resulting Venn regions (Qiu et al., 2021; Qiu et al., 2022).

4.3 Tracking changes in physicochemical characteristics of features

In addition to tracking changes in the number and intensity of features (as described in section “4.1. Tracking Changes in the Total Number of Features and Overall Peak Areas”), insightful information on the impact of wastewater treatment can be obtained from examining changes in features physicochemical properties, which are predictable at a low level of identification (Hollender et al., 2023). For instance, changes in m/z values and retention times can provide approximate indications of molecular weight and polarity, respectively. Wastewater effluents are commonly characterized by smaller molecular weight (lower average m/z values) and higher polarity (lower average retention time) compared to influents (Nihemaiti et al., 2022; Nováková et al., 2023). However, this trend may not be generalizable across all treatment systems. Delanka-Pedige et al. (2024b) reported no significant differences in the range or mean m/z values of detected features between the influent and effluent of algal and activated sludge treatment systems or between the effluents of the two systems (Delanka-Pedige et al., 2024a).

Additionally, assessing changes in molecular indexes derived from the assigned molecular formula without extensive identity confirmation is adopted for investigating the effect of the treatment on the chemical nature of wastewater samples (Hollender et al., 2023; Qiu et al., 2021). Examples of such strategies include (1) tracking the change of unsaturation degree as an indicator for hydrophobicity-altering reactions such as hydrolysis or oxidation via monitoring of double bond equivalents (DBE) and its variant DBE-O; (2) examining chemical transformation of different samples by comparing their van Krevelen diagrams plotting the atomic ratio X/C (where X is an element of interest) against H/C; (3) recognizing removal and transformation of specific homologous series such as those containing -CH₂- or -C₂H₄O- moieties using Kendrick Mass Defect (KMD) plots (Verkh et al., 2018). Figure 16 demonstrates how van Krevelen plots can provide mechanistic insight into wastewater treatment by tracking shifts in molecular H/C and O/C ratios across process stages. Their application to a full-scale inverted anaerobic/anoxic/oxic (A/A/O) system configuration indicated that the treatment process preferentially removed nitrogen- and sulfur-containing molecules relative to CHO molecules, which suggest a shift toward more oxygenated and less saturated compounds in the effluent compared to the influent (Wen et al., 2024).

FIGURE 16

4.4 Comparative framework of NTA-based evaluation methods

4.4.1 Information provided by the evaluation methods

Table 4 compares the NTA-based evaluation strategies discussed in this review. The different approaches provide complementary perspectives on treatment performance. Approaches based on feature numbers and peak areas provide rapid, overall indicators of changes in the sample chemical profile and are particularly suitable for preliminary screening and the comparison of treatment performance when compound identification is limited. However, they provide little information on the fate of individual contaminants or transformation products.

Feature-level analyses based on removal efficiency or fold change provide substantially greater detail by distinguishing eliminated, persistent, and newly formed features. Therefore, they are especially valuable for advanced treatment processes, such as oxidation, where chemical transformation is expected, highlighting one of the principal advantages of NTA over conventional target analysis. Their main limitation is the greater analytical effort required for reliable feature alignment.

Analyses based on physicochemical characteristics provide qualitative insight into the nature of the chemical transformations occurring during treatment. By examining changes in molecular weight, retention time, or formula-derived indexes, such as DBE, DBE-O, and van Krevelen diagrams, they help explain why certain compounds are preferentially removed or transformed and are therefore particularly useful for understanding treatment mechanisms.

4.4.2 Practical considerations

The applicability of each evaluation strategy depends on the information generated during NTA processing. Tracking changes in the total number of features and overall peak areas requires reproducible feature detection and therefore can be applied without compound identification. Similarly, RE- and FC-based approaches rely on consistent feature matching between samples and use peak area as a surrogate for concentration, making them applicable without compound identification. However, in addition to the need for robust feature detection and alignment, a common challenge shared by peak-area-based approaches is their susceptibility to matrix effect, which may influence feature responses. Therefore, matrix effect should be considered when interpreting treatment performance and estimating removal efficiencies (see “5. Impact of matrix effect and offsetting strategies”).

The analysis of molecular indexes generally requires molecular formula assignment but not complete structural elucidation. Consequently, these approaches occupy an intermediate position between feature-based evaluation and compound-specific interpretation.

The analytical platform should also be considered when selecting an evaluation strategy. Peak-area-based metrics can be applied to both chromatographic and direct-infusion MS workflows, whereas approaches relying on retention time as a surrogate for polarity require chromatographic separation and are therefore not applicable to direct-infusion MS. On the other hand, formula-derived descriptors, such as DBE, van Krevelen diagrams, and Kendrick mass defect analysis, remain applicable whenever reliable molecular formula assignment is achievable.

Overall, no single approach is sufficient to comprehensively evaluate wastewater treatment performance. Tracking overall feature numbers and abundances provides a broad assessment of treatment performance, feature-fate analyses quantify removal and indicate the formation of transformation products, and physicochemical characterization explains the underlying mechanisms. Therefore, combining these complementary approaches yields the most informative assessment of treatment performance.

5 Impact of matrix effect and offsetting strategies

Matrix effect is the suppression or enhancement in the analyte response caused by the constituents of the matrix (Hollender et al., 2023). Consequently, the matrix effect yields variations in area counts for identical concentrations in different background matrices (Parry and Young, 2016). This might bias the removal efficiency estimated based on peak area or signal intensity (Bader et al., 2017; Gollong et al., 2022; Parry and Young, 2016). An example of an erroneous result caused by matrix effects is mistakenly classifying a persistent feature as transformed after treatment, simply because the matrix effect obscured its detection in the influent (Pandey et al., 2024). The matrix effect is estimated by spiking internal standards (IS) into matrix samples and blanks (solvents), then, calculating the IS response in the matrix samples relative to its response in the blank sample which is free from matrix components (Li et al., 2018). However, the matrix effect can be neglected when calculating removal efficiency if the matrix exerts a similar influence across all samples. This conclusion can be straightforwardly deduced by comparing the IS response in the effluent directly to that in the influent (Nürenberg et al., 2019). If the quotient of the peak areas of the internal standards in the influent and effluent lies within the range of 1 ± 0.25, the responses are considered similar, implying a negligible matrix effect (Li et al., 2018; McLachlan et al., 2022).

To offset matrix effects, several strategies have been developed. Diluting sample extracts has been demonstrated to reduce the matrix effect. For instance, a dilution factor of 1: 3 enhanced the recovery of the internal standard intensities by 33% in the diluted influent samples and 23% in diluted effluent samples compared to undiluted samples (Nürenberg et al., 2015). For comparable matrices, Tisler et al. (2021) recommended the analysis of influent and effluent extracts at relative enrichment factors (REF) that achieve low matrix effects (<20%), which was REF 50 for effluent and REF 10 for influent water extracts in their study (Tisler et al., 2021). However, the omission of low-intensity features is a key drawback of dilution, limiting its applicability primarily to high-intensity feature (Nürenberg et al., 2015; Tisler et al., 2021).

Normalizing analyte responses to those of internal standards is a common strategy for compensating for matrix effects in target analysis. Hence, Nurenberg et al., 2015 assessed the feasibility of this approach for NTA utilizing three strategies for the selection of IS: (i) IS with a comparable mass, (ii) IS with similar RT, or (iii) one IS was used for all features. No improvement was observed in the median relative standard deviation (RSD) of the feature responses compared to uncorrected responses, rendering IS normalization effectiveness questionable for non-target analysis (Nürenberg et al., 2015). Moreover, addition of ISs may increase the matrix effect (Tisler, Engler, et al., 2022). Nevertheless, Wang et al. (2020) successfully applied this normalization approach to the peak areas of 63 PF AS, using the closest eluting internal standards, before assessing PFAS removal in local wastewater treatment plants. The relatively high level of identification achieved for those compounds (level 3 and above) might have facilitated better matching between the analytes and the internal standards (Wang et al., 2020).

As highlighted in section “4.2. Investigating the Fate of Individual Features”, broadening the range of the FC values for the consistent features category (e.g., 0.5–2 (Bader et al., 2017)) is a strategy applied to account for the matrix effect (Peter et al., 2019). In the same regard, Wang et al. (2018) defined the FC values range for consistent features as 1/6–six based on the mean values of the ratio A_effluent/A_influent expanded at threefold of the standard deviations for a set of spiked standards (Wang et al., 2018).

Moreover, two recent methods have been introduced to address limitations in NTA matrix effect correction: (1) a total ion chromatogram (TIC)-based correction method combined with quantitative structure-property relationship (QSPR) models to determine structure-specific matrix effects (Tisler, Engler, et al., 2022; Tisler et al., 2021), and (2) an individual sample-matched internal standard (IS-MIS) method, which selects an internal standard that achieves the lowest relative standard deviation (RSD) across various relative enrichment factor (REF) values (Molnár Karlsson and Christensen, 2025).

Scaling features to equal intensities is required to compare samples of different matrix content (Minkus et al., 2022b). A relevant approach is that proposed by Motteau et al. (2024) to compensate for the matrix effect. In this method, the area ratios of standards spiked into influent and effluent samples were determined. The mean and median ratios of the standard signals in effluent to those in influent were 2.1 and 2.0 respectively. Therefore, a correction factor of two was applied to features detected in the influent samples to offset the matrix effect (Motteau et al., 2024).

Another scaling approach is the one introduced by Schollée et al. (2021), a median intensity per internal standard is calculated over all samples. In each sample, the deviation of each IS to the median intensity is then calculated. Subsequently, the median deviation is calculated for each sample, which is then indicative of the matrix suppression in that sample and is then applied as a normalization factor to all intensities in that sample (Schollée et al., 2021).

Apart from offsetting matrix effect purposes, other data scaling and normalization approaches aimed at eliminating systematic variations are common practices for data preparation before statistical analysis (Hollender et al., 2023). Some examples are volume-based normalization where peak areas are divided by the extracted sample volume (Mladenov et al., 2022); relative abundance normalization calculated as the ratio of a monoisotopic peak area to the total peak area of all monoisotopic masses detected in a sample (Delanka-Pedige, Young, et al., 2024); pareto scaling based on the square root of the standard deviation (Ponce-Robles et al., 2018); intensity based normalization where peaks area are normalized to the maximum intensity (Schollée et al., 2018) or to a selected intensity percentile value for a compound in a sample, e.g., 75th percentile (Wang et al., 2019); and standardizing feature peak areas using the corresponding features in the QC pooled samples followed by log₂-normalization (Xu et al., 2024). Investigating the interaction between these normalization methods and matrix effect estimation would be critical for reliable treatment efficiency evaluation in wastewater studies, considering the results of a previous non-target metabolomics study that demonstrated that the suitability of data normalization method depends on sample matrix complexity (Cuevas-Delgado et al., 2020).

6 Recurring themes in NTA applications for wastewater treatment

6.1 Inferring removal efficiencies from compound physicochemical properties

Several of the inspected studies have explored how feature characteristics can indicate removal patterns under specific treatments. In this regard, NTA can offer important insights into how the nature of compounds relates to their fate (Parry and Young, 2016). This can be achieved through examining retention times and m/z values - the basic data provided by NTA (Li et al., 2026). For instance, among features detected under ESI-, Kiss et al. (2018) observed that the biological treatment coupled with ozonation most significantly decreased the number of features in the m/z range of 500–600 Da. Similarly, features detected in ESI+ with middle retention time values (between 4 and 6 min) were the most affected by the integrated treatment (Kiss et al., 2018). Motteau et al. (2024) demonstrated that conventional activated sludge (CAS) treatment removed features with retention times that exceeded the mid-point of the retention time range (>6 min) and a mass of >600 Da (Motteau et al., 2024). Nihemaiti et al. (2022) observed that both persistent and newly formed features under disinfection with performic acid were in average smaller and more polar (low m/z, low RT) than the removed ones (Nihemaiti et al., 2022). Further supporting these trends, Schollée et al. (2018) reported that ozonation transformation products (OTPs) with higher RT were more removable, suggesting that some polar OTPs remain recalcitrant (Schollée et al., 2018). This aligns with the observation of Motteau et al., 2024 that more apolar and higher-molecular-mass compounds were more likely to be eliminated or substantially reduced, regardless of the WWTP treatment line (Motteau et al., 2024).

As the confidence level of compound identification increases, removal patterns can be investigated in terms of more specific physicochemical properties. For example, Delanka-Pedige et al. (2024b) found that certain physicochemical properties of compounds such as vapor pressure, Henry’s Law constant, and boiling point, influenced contaminant rejection in the different membrane distillation configurations used to treat oil field produced water. Despite their theoretical ability of 100% salt retention, vacuum membrane distillation (VMD) and photocatalytic membrane distillation (PMD) (with and without UV) failed to purify distillates from volatile/evaporative compounds with high Henry’s law constant and vapor pressure, and low boiling (Delanka-Pedige, Young, et al., 2024). Making use of the data of 111 compounds tentatively identified via suspect screening, Mladenov et al. (2022) assessed the association between molecular descriptors related to shape (Geometric shape, Kier shape, Zagreb group, polar surface area) and bonds (no. Of hydrogen bond acceptors, no. Of hydrogen bond donors, no. Of atoms, no. Of functional groups, no. Of aromatic rings), and removal in WWTPs employing biological treatments. Recalcitrant compounds tended to have lower numbers of H bond acceptors-1, a lower fraction of rotatable bonds, and lower numbers of atoms (Mladenov et al., 2022). Blum et al. (2019) found that high hydrophobicity, low polarity, few heteroatoms and limited H-bond donors or acceptors are the characteristics of significantly removed compounds in sand and char-fortified filter beds treated wastewater. In sorption-based treatment technology such as sedimentation and soil filtration, the octanol-water partition coefficient (K_OW) plays a pivotal role in determining removal efficacy (Blum et al., 2019). Highly hydrophobic compounds (log K_OW 4.8–12) consistently achieve high overall median removal efficiencies (≥90%) and vice versa (Blum et al., 2017).

The extensive amount of data provided by NTA analysis has motivated the establishment of quantitative structure-property relationship (QSPR) mathematical models to predict the removal of compounds based on structural and molecular predictors (Blum et al., 2019; Chirico et al., 2024; Johnson et al., 2024). These models, replacing tedious experiments, have the advantage of saving cost and effort. Chirico et al. (2024) explicitly modelled the relationship between log FC and the molecular descriptors of chemicals that underwent mechanical, chemical and biological treatment and sand filtration using a simple multiple linear regression (MLR) model. Bootstrap and randomization procedures were applied to avoid overfitting and coincidental relationships between the selected descriptors and the removal metric. For generalizability, the model applicability was further extended to similar compounds having Tanimoto similarity index >0.83. Finally, class-specific models for polyethylene glycol (PEG) and polypropylene glycol (PPG), representing homogenous class of chemicals, were additionally developed. For the three scenarios, the improved QSPR models were based on limited number of descriptors (1–2) with R² ranging from 0.54 to 0.94 and Mean Average Error (MAE) ranging from 0.059 to 0.51 (Chirico et al., 2024).

6.2 Correlation of emerging contaminant removal and bulk water quality parameters

A couple of the reviewed articles addressed the potential of correlating the occurrence of emerging contaminants (EC) with frequently monitored conventional chemical and spectroscopic water-quality parameters such as chemical oxygen demand (COD) (Nürenberg et al., 2019), biological oxygen demand (BOD), total suspended solids (TSS), total nitrogen (N), total phosphorus (P) (McLachlan et al., 2022) and the specific UV absorbance at 254 nm (SUVA254) (Delanka-Pedige, Young, et al., 2024). These parameters are used to monitor the overall removal of organic material and nutrients. The main reasons driving the selection of these parameters are their availability and ease of their measurement via standardized methods (Prasse et al., 2015).

Delanka-Pedige et al. (2024a) highlighted the importance of SUVA254 in indicating the presence of ECs. It is reported in the study that SUVA values less than two indicated a greater fraction of hydrophilic, non-humic organic matter based on the United State Environmental Protection Agency (US EPA) drinking water guidance note. The values of SUVA <2 detected for the effluents of CAS and algal treatments, the treatment systems examined in this study, triggered further analysis for determining the presence of ECs (Delanka-Pedige et al., 2024a).

McLachlan et al. (2022) found a strong correlation between log(FC) for each chemical in the effluents and log BOD, log TSS and log N, with log TSS giving superior results. In contrast, log P exhibited the weakest correlation for most of the chemicals. This was attributed to the fact that BOD, TSS and N are primarily removed through biotransformation, the predominant mechanism for organic contaminant removal, whereas P is mainly removed through chemical treatment, which has minimal impact on organic contaminants (McLachlan et al., 2022). In the same regard, Nürenberg et al. (2019) concluded that the average contaminant removal did not correlate with the conventional water quality parameters such as solid retention time (SRT), hydraulic retention time (HRT), sludge concentration, COD removal and N removal which was contrarily found by (McLachlan et al., 2022) to correlate with micro contaminant removal (Nürenberg et al., 2019). These contrasting conclusions motivate further investigation to determine whether the apparent conflict reflects differences in contaminant classes, plant configuration, or analytical design rather than a true contradiction.

6.3 Ecological risk assessment

Several of the reviewed studies addressed health and ecological risks associated with treatment residuals. This trend highlights that wastewater treatment evaluation should consider not only removal rates but also the risks posed by residual compounds including transformation products (Lopez-Herguedas et al., 2024). Risk assessment studies rely on toxicity data which can be experimentally acquired, literature-derived, or in-silico-predicted, with their scope varying from limited number of compounds to whole samples (Sepman et al., 2023).

While experimental toxicological assessment approaches are broadly divided into in vivo and in vitro studies (Sepman et al., 2023), only in vitro methods were applied in the reviewed studies assessing risk. For instance, Lopez-Herguedas et al. (2024) used the Microtox® bioassay to evaluate the efficiency of membrane bioreactor (MBR) technology in toxicity removal (Lopez-Herguedas et al., 2024). Additionally, Gollong et al. (2022) conducted multiple in vitro assays (cytotoxicity, mutagenicity, genotoxicity and endocrine disruption) to assess the toxicity of ozonation and activated carbon treated wastewater using whole samples (Gollong et al., 2022). In both studies, the toxicity effect did not exceed thresholds. Moreover, it is worth noting that compound identification or quantification was not required to assess the toxicity of wastewater samples in experimental toxicological assessment. Two main approaches are used for experimental toxicity assessment. Effect-based methods (EBMs) assess the toxicity of an intact sample using in vitro or in vivo bioassays (Alvarez-Mora et al., 2025; Li et al., 2026), whereas effect-directed analysis (EDA) sequentially examines sample fractions obtained through chromatographic fractionation (McCord et al., 2022). Directing subsequent identification efforts toward risk-driving components in bioactive fractions is a recognized strategy for prioritizing unknown features in NTA. For instance, Alvarez-Mora et al. (2025) identified testosterone, androsterone, and norgestrel as androgenic effect drivers in hospital wastewater effluent, using a workflow that combined microplate fractionation, AR-CALUX bioassay, NTA, and MLinvitroTox–a machine learning toxicity prediction model. This approach reduced the number of candidate features by more than 95% (Alvarez-Mora et al., 2025).

To overcome the labor-intensive nature of conventional EDA, virtual EDA (vEDA) is emerging as a data-driven approach that eliminates the need for physical sample fractionation by prioritizing potential toxicants directly within the HRMS feature space (Li et al., 2026; Zweigle et al., 2025).

Toxicity information extracted from literature, databases or in silico tools, and quantified environmental exposure are combined to evaluate the risk posed by individual chemicals or chemical mixtures (Sepman et al., 2023). These methods are applied post identification (Han et al., 2024; Sepman et al., 2023). Risk Quotient (RQ) indicator, frequently used for assessing ecological risk, is an example of these methods. RQ is calculated as the ratio of measured environmental concentration (MEC) and predicted no-effect concentration (PNEC). PNEC represents the lowest values obtained from the most sensitive species (e.g., bacteria, algae, invertebrates, and fish). RQ values higher than one is generally considered as highly hazardous, values between one and 0.1 as intermediate and RQ < 0.1 shows low risk to the environment (Sepman et al., 2023). Since it is a concentration-based method, it applies when concentration information is available (Lopez-Herguedas et al., 2024). In the context of non-target and suspect screening, RQ has been estimated based on semi-quantification for compounds lacking standards (Xiao et al., 2024) (see “4.2 Investigating the Fate of Individual Features”). Toxicological Priority (ToxPi) score is another toxicity assessment metric (Equation 4).where R represents the logarithmic transformation of the maximum instrumental peak area responses value, T is log and K stands for the logarithmic transformation of (log K_OW) of the compounds. W_R, W_T and W_D are the weightings assigned to each respective variable (Xiao et al., 2024). NORMAN Ecotoxicology Database was reported as a source for PNEC whereas ECOSAR module of the EPI WEB software was used to predict PNEC and log K_OW. Examples of sets of weighting values are W_R = 2, W_T = 2 and W_D = 1 (Xiao et al., 2024) and W_R = 1, W_T = 1 and W_D = 0 (Lopez-Herguedas et al., 2024). A high ToxPi score represents increased potential environmental risk (Xiao et al., 2024).

In addition, Cheminformatics analysis modules, EPA-developed web-based tools, that combine experimental and in silico based toxicity predicted via quantitative structure-activity relationships (QSAR), has been utilized to generate hazard profiles based on compounds identifiers or structures. Toxicity criteria are classified into human health-based toxicity, aquatic life based toxicity and environmental fate-based toxicity (Delanka-Pedige et al., 2024a; Delanka-Pedige et al., 2024b). Applying this method to compare hazard profiles of municipal wastewaters algal treatment effluent to that of activated sludge treatment, the alga treatment showed a lower risk under all criteria except in the chronic aquatic toxicity category (Delanka-Pedige et al., 2024a). Vacuum and photocatalytic membrane distillation treated hyper-saline oil and gas (O&G) produced water were also assessed using this tool. The association of the majority of the tentative identified compounds in the water samples with some form of human health, ecotoxicity, or environmental fate-based risks raised concern about the reuse of this water in application of high potential contact of humans or the environment. While not accounting for concentration is a key limitation of this risk assessment approach, the results can still guide more rigorous targeted investigations of prioritized compounds (Delanka-Pedige et al., 2024a).

Advances in machine learning (ML) have helped overcome key challenges associated with NTA-based risk assessment, particularly those related to quantification and gaps in toxicity data. In the absence of reference standards, ML-based models can predict mass-spectrometric response factors (Hu et al., 2023), and ionization efficiencies, both of which are fundamental to quantitative NTA (Johnson and Abrahamsson, 2024). In addition, ML models can predict contaminant toxicity by learning relationships between chemical structure, physicochemical properties, and toxicological activity. These approaches have substantially strengthened the application of NTA in environmental risk assessment (Lin et al., 2026).

6.4 Identification of transformation products (TPs)

Current wastewater treatment assessment relying only on the removal efficiency of a few selected targeted organic micropollutants (OMPs) may lead to an overestimation of treatment efficiency and omission of potential risks posed by non-targeted compounds including transformation products (TPs) formed during treatment (Gollong et al., 2022). Therefore, addressing the formation of TPs is critical for evaluating wastewater treatment (Chen et al., 2024), particularly as TPs have been shown in many cases to be more hazardous than their parent compounds (Hu et al., 2023; Petromelidou et al., 2024). Lack of reference standards (Deeb et al., 2017) and incomplete knowledge of transformation pathways are the key challenges related to the identification of transformation products (TPs) in environmental samples (Wang et al., 2020). Nevertheless, several NTA-based strategies have been developed to address these challenges, taking advantage of the advances in high-resolution mass spectrometry (HRMS) (Schollée et al., 2015). Differential analysis associated with univariate and multivariate statistical methods (e.g., significance testing, principal component analysis (PCA), Hierarchical clustering analysis (HCA)), are used to prioritize potential TP features (Schollée et al., 2015; 2016; 2018). These prioritized features are then further subjected to identification and structural elucidation. Suspect screening and non-target methods based on structural and molecular analysis are employed for the prioritization and identification of unknown compounds including TPs (Ponce-Robles et al., 2018; Szabo et al., 2024).

Suspect screening of transformation products (TPs) depends on the prior compilation of candidate TPs lists, which may be derived from various sources, including literature-reported compounds, specialized TPs libraries, laboratory-induced transformation experiments, metabolic logic analysis, pathway of expected transformation reactions or in silico prediction tools based on established chemical transformation rules (Gulde et al., 2021; Hu et al., 2023; Jaén-Gil et al., 2018; Schollée et al., 2015; 2018; Wang et al., 2025). The key limitations of suspect screening for transformation products include the necessity for prior knowledge of parent compounds, omission of transformation products formed through unknown pathways, reliance on matching HRMS data against databases of limited coverage, and a high risk of false positives when long suspect lists are used for prioritization (Chen et al., 2024; Schollée et al., 2015; Szabo et al., 2024; Wang et al., 2020).

To overcome suspect screening limitations, non-target analysis approaches based on structural analysis and spectral similarities are employed (Hollender et al., 2023). These methods include grouping parents compounds and potential transformation products based on common diagnostic fragment ion (DFI) (Wang et al., 2020), applying the molecular networking algorithm of Global Natural Products Social Molecular Networking (GNPS) or the newly introduced structural molecular networking technique (Xia et al., 2025) to search for structurally related compounds (Chen et al., 2024; Hollender et al., 2023; Wu et al., 2023), linking TPs with their potential precursors using paired mass distance (Tu et al., 2026), and utilizing mass defect and its variant Kendrick mass defect filters used for identification of homologous compounds with shared repetitive masses (Szabo et al., 2024; Wang A.-G. et al., 2025; Zweigle et al., 2025).

Isotope labeling methods have also been proven effective in distinguishing TPs from similar unreactive originally presented isomers and exploring reaction mechanisms. Utilization of D₂O in a UV irradiated treatment (Dwinandha et al., 2024) and ozonation with mass-labeled ozone (¹⁸O) (Jennings et al., 2025) were successfully applied to investigate the production of TPs.

7 Potential of NTA to support wastewater engineering and regulatory monitoring

NTA holds significant potential for wastewater engineering by assessing how effectively treatment processes remove contaminants. Findings from NTA studies can help investigate the fate of contaminants (Chen et al., 2024; Schollée et al., 2021), identify optimal treatment conditions (Parry and Young, 2016), evaluate alternative technologies (Delanka-Pedige et al., 2024a; Delanka-Pedige et al., 2024b; Johnson et al., 2024; Rodriguez-Otero et al., 2026; Xu et al., 2024), and compare overall performance (García-Martínez et al., 2025; Mandlazi et al., 2026; McLachlan et al., 2022). The comprehensive information generated by NTA can also support regulatory monitoring by identifying contaminants that may require future regulatory attention and by informing wastewater discharge policies (Hollender et al., 2019; Kronsbein et al., 2026).

Despite its potential, wider adoption of NTA in routine analysis is hindered by several barriers: (i) the lack of standardized guidelines for analysis workflow, data quality assessment, and results reporting; (ii) the complexity of processing and interpreting large NTA datasets for engineering and regulatory applications, together with the limited availability and user-friendliness of advanced data-processing tools; (iii) the qualitative nature of NTA, which prevents accurate quantification and thus limits its utility in exposure and risk assessment frameworks that rely on precise concentration data; (iv) limited expertise and capacity among practitioners and end users; and (v) the absence of credentialing programs that would allow NTA providers to benchmark their performance (Kronsbein et al., 2026; Nason et al., 2025).

To support the routine use of NTA in wastewater treatment evaluation and regulatory monitoring, future efforts should focus on harmonizing analytical workflows including sampling, sample preparation, instrumental analysis, data processing, and QA/QC procedures. Adoption of best practices (Hollender et al., 2023) and standardized reporting guidelines (Peter et al., 2021) developed by organizations such as BP4NTA and the NORMAN network would substantially improve the comparability and reproducibility of NTA studies.

Of paramount importance is the development of automated and user-friendly data-processing platforms that enhance efficiency and reduce the computational expertise required for NTA data processing. Continued advances in machine learning (ML) and artificial intelligence (AI) are expected to play a significant role in achieving these objectives.

At the same time, advances in quantitative non-target analysis (qNTA) are required to enable the integration of NTA into treatment performance assessments, risk evaluations, and future water-quality regulations.

Continued investment in training and knowledge transfer is also essential to improve the generation, interpretation, and practical use of NTA data by laboratory scientists, wastewater professionals, and regulators.

The lack of established credentialing pathways for NTA providers remains an important challenge. In the short term, interlaboratory comparisons studies using wastewater samples, building on previous initiatives such as ENTACT (Sobus et al., 2018) and NORMAN collaborative trials (Schymanski et al., 2015), can provide performance benchmarks and training, and support harmonization efforts. In the longer term, formal credentialing will require the development of agreed QA/QC criteria against which the proficiency of NTA providers will be assessed.

Achieving these goals will require collaboration among existing NTA community networks (e.g., BP4NTA and the NORMAN network), standardization organizations (e.g., ISO, AOAC), researchers, and water-sector regulators to establish harmonized practices (Nason et al., 2025) that support the routine implementation of NTA in both wastewater treatment evaluation and regulatory monitoring frameworks.

8 Conclusion and future directions

Progress in these priority areas will build on the current achievements of NTA and support the greater realization of its potential for wastewater treatment evaluation, process optimization, environmental risk assessment, and regulatory monitoring.

References

• 1

  Alvarez-MoraI.MuratulyA.JohannS.ArturiK.JüngerF.HuberC.et al (2025). High-throughput effect-directed analysis of androgenic compounds in hospital wastewater: identifying effect drivers through non-target screening supported by toxicity prediction. Environ. Sci. and Technol.59 (45), 24513–24525. 10.1021/acs.est.4c09942

  • CrossRef
  • Google Scholar

• 2

  AlygizakisN.LestremauF.Gago-FerreroP.Gil-SolsonaR.ArturiK.HollenderJ.et al (2023). Towards a harmonized identification scoring system in LC-HRMS/MS based non-target screening (NTS) of emerging contaminants. TrAC Trends Anal. Chem.159, 116944. 10.1016/j.trac.2023.116944

  • CrossRef
  • Google Scholar

• 3

  ArminR.WachendorfJ.WeberM.SchmidtT. C. (2025). Enhanced industrial wastewater monitoring: method development for non-target screening of highly polar substances using ZIC-HILIC-HRMS. Anal. Bioanal. Chem.417 (1), 167–181. 10.1007/s00216-024-05635-9

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 4

  BaderT.SchulzW.LuckeT.SeitzW.WinzenbacherR. (2016). Application of non-target analysis with LC-HRMS for the monitoring of raw and potable water: strategy and results. ACS Symp. Ser.1242, 49–70. 10.1021/bk-2016-1242.ch003

  • CrossRef
  • Google Scholar

• 5

  BaderT.SchulzW.KümmererK.WinzenbacherR. (2017). LC-HRMS data processing strategy for reliable sample comparison exemplified by the assessment of water treatment processes. Anal. Chem.89 (24), 13219–13226. 10.1021/acs.analchem.7b03037

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 6

  BattA. L.BrunelleL. D.QuineteN. S.StebelE. K.NgB.GardinaliP.et al (2024). Investigating the chemical space coverage of multiple chromatographic and ionization methods using non-targeted analysis on surface and drinking water collected using passive sampling. Sci. Total Environ.955, 176922. 10.1016/j.scitotenv.2024.176922

  • CrossRef
  • Google Scholar

• 7

  BergéA.BuletéA.FildierA.MaillerR.GasperiJ.CoquetY.et al (2018). Non-target strategies by HRMS to evaluate fluidized micro-grain activated carbon as a tertiary treatment of wastewater. Chemosphere213, 587–595. 10.1016/j.chemosphere.2018.09.101

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 8

  BeschnittA.SchwikowskiM.HoffmannT. (2022). Towards comprehensive non-target screening using heart-cut two-dimensional liquid chromatography for the analysis of organic atmospheric tracers in ice cores. J. Chromatogr. A1661, 462706. 10.1016/j.chroma.2021.462706

  • CrossRef
  • Google Scholar

• 9

  BlackG.LoweC.AnumolT.BadeJ.FavelaK.FengY.-L.et al (2023). Exploring chemical space in non-targeted analysis: a proposed ChemSpace tool. Anal. Bioanal. Chem.415 (1), 35–44. 10.1007/s00216-022-04434-4

  • CrossRef
  • Google Scholar

• 10

  BlumK. M.AnderssonP. L.RenmanG.AhrensL.GrosM.WibergK.et al (2017). Non-target screening and prioritization of potentially persistent, bioaccumulating and toxic domestic wastewater contaminants and their removal in on-site and large-scale sewage treatment plants. Sci. Total Environ.575, 265–275. 10.1016/j.scitotenv.2016.09.135

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 11

  BlumK. M.GallampoisC.AnderssonP. L.RenmanG.RenmanA.HaglundP. (2019). Comprehensive assessment of organic contaminant removal from on-site sewage treatment facility effluent by char-fortified filter beds. J. Hazard. Mater.361, 111–122. 10.1016/j.jhazmat.2018.08.009

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 12

  BrunelleL. D.BattA. L.ChaoA.GlassmeyerS. T.QuineteN.AlvarezD. A.et al (2024). De facto Water Reuse: investigating the Fate and Transport of Chemicals of Emerging Concern from Wastewater Discharge through Drinking Water Treatment Using Non-targeted Analysis and Suspect Screening. Environ. Sci. and Technol.58 (5), 2468–2478. 10.1021/acs.est.3c07514

  • CrossRef
  • Google Scholar

• 13

  ButlerT.FrandsenA.LightheartR.BarghB.KerbyT.WestK.et al (2023). MS2Mol: A Transformer Model for Illuminating Dark Chemical Space from Mass Spectra. 10.26434/chemrxiv-2023-vsmpx-v4

  • CrossRef
  • Google Scholar

• 14

  CancholaA.TranL. N.WooW.TianL.LinY.-H.ChouW.-C. (2025). Advancing non-target analysis of emerging environmental contaminants with machine learning: current status and future implications. Environ. Int.198, 109404. 10.1016/j.envint.2025.109404

  • CrossRef
  • Google Scholar

• 15

  CelmaA.SanchoJ. V.SchymanskiE. L.Fabregat-SafontD.IbáñezM.GoshawkJ.et al (2020). Improving target and suspect screening high-resolution mass spectrometry workflows in environmental analysis by ion mobility separation. Environ. Sci. and Technol.54 (23), 15120–15131. 10.1021/acs.est.0c05713

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 16

  ChenZ.XiaD.LiuH.WangR.HuangM.TangT.et al (2024). Tracing contaminants of emerging concern and their transformations in the whole treatment process of a municipal wastewater treatment plant using nontarget screening and molecular networking strategies. Water Res.267 (Scopus), 122522. 10.1016/j.watres.2024.122522

  • CrossRef
  • Google Scholar

• 17

  ChiricoN.McLachlanM. S.LiZ.PapaE. (2024). In silico approaches for the prediction of the breakthrough of organic contaminants in wastewater treatment plants. Environ. Sci. Process. and Impacts26 (2), 400–410. 10.1039/D3EM00267E

  • CrossRef
  • Google Scholar

• 18

  Cuevas-DelgadoP.DudzikD.MiguelV.LamasS.BarbasC. (2020). Data-dependent normalization strategies for untargeted metabolomics—A case study. Anal. Bioanal. Chem.412 (24), 6391–6405. 10.1007/s00216-020-02594-9

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 19

  DaiZ.XuJ.GuanJ.FengM.LiuY.XingC.et al (2025). TP-Transformer: an interpretable model for predicting the transformation pathways of organic pollutants in chemical oxidation processes. Environ. Sci. and Technol.59 (30), 15853–15864. 10.1021/acs.est.5c02701

  • CrossRef
  • Google Scholar

• 20

  DeebA. A.SchmidtT. C. (2016). Tandem anion and cation exchange solid phase extraction for the enrichment of micropollutants and their transformation products from ozonation in a wastewater treatment plant. Anal. Bioanal. Chem.408 (16), 4219–4232. 10.1007/s00216-016-9523-y

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 21

  DeebA. A.StephanS.SchmitzO. J.SchmidtT. C. (2017). Suspect screening of micropollutants and their transformation products in advanced wastewater treatment. Sci. Total Environ.601–602, 1247–1253. 10.1016/j.scitotenv.2017.05.271

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 22

  DefossezE.BourquinJ.Von ReussS.RasmannS.GlauserG. (2023). Eight key rules for successful data‐dependent acquisition in mass spectrometry‐based metabolomics. Mass Spectrom. Rev.42 (1), 131–143. 10.1002/mas.21715

  • CrossRef
  • Google Scholar

• 23

  Delanka-PedigeH. M. K.AhmedA.SchaubT.TrainorP.NirmalakhandanN. (2024a). Feasibility of organic micropollutant removal from municipal wastewaters by algal treatment vs. activated sludge treatment: Comparison based on non-targeted organic compound analysis. J. Environ. Manag.367, 121936. 10.1016/j.jenvman.2024.121936

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 24

  Delanka-PedigeH. M. K.YoungR. B.AbutokaikahM. T.ChenL.WangH.ImihamillageK. A. B. I.et al (2024b). Non-targeted analysis and toxicity prediction for evaluation of photocatalytic membrane distillation removing organic contaminants from hypersaline oil and gas field-produced water. J. Hazard. Mater.471, 134436. 10.1016/j.jhazmat.2024.134436

  • CrossRef
  • Google Scholar

• 25

  DeversJ.PattisonD. I.HansenA. B.ChristensenJ. H. (2025). Comprehensive two-dimensional gas chromatography as a tool for targeted and non-targeted analysis of contaminants of emerging concern in wastewater. Talanta282, 127032. 10.1016/j.talanta.2024.127032

  • CrossRef
  • Google Scholar

• 26

  DoyleM. G.OdenkirkM. T.StewartA. K.NelsonJ. P.BakerE. S.De La CruzF. (2022). Assessing the fate of dissolved organic compounds in landfill leachate and wastewater treatment systems. ACS ES&T Water2 (12), 2502–2509. 10.1021/acsestwater.2c00320

  • CrossRef
  • Google Scholar

• 27

  DuarteR. M. B. O.BrandãoP. F.DuarteA. C. (2023). Multidimensional chromatography in environmental analysis: comprehensive two-dimensional liquid versus gas chromatography. J. Chromatogr. A1706, 464288. 10.1016/j.chroma.2023.464288

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 28

  DwinandhaD.ElsamadonyM.GaoR.FuQ.-L.LiuJ.FujiiM. (2024). Interpretable machine learning and reactomics assisted isotopically labeled FT-ICR-MS for exploring the reactivity and transformation of natural organic matter during ultraviolet photolysis. Environ. Sci. and Technol.58 (1), 816–825. 10.1021/acs.est.3c05213

  • CrossRef
  • Google Scholar

• 29

  El-DeenA. K.ShimizuK. (2022). Suspect and non-target screening workflow for studying the occurrence, fate, and environmental risk of contaminants in wastewater using data-independent acquisition. J. Chromatogr. A1667, 462905. 10.1016/j.chroma.2022.462905

  • CrossRef
  • Google Scholar

• 30

  ElserD.HuberF.GaquerelE. (2023). Mass2SMILES: deep learning based fast prediction of structures and functional groups directly from high-resolution MS/MS spectra. bioRxiv. 10.1101/2023.07.06.547963

  • CrossRef
  • Google Scholar

• 31

  Gago-FerreroP.BletsouA. A.DamalasD. E.AalizadehR.AlygizakisN. A.SingerH. P.et al (2020). Wide-scope target screening of >2000 emerging contaminants in wastewater samples with UPLC-Q-ToF-HRMS/MS and smart evaluation of its performance through the validation of 195 selected representative analytes. J. Hazard. Mater.387, 121712. 10.1016/j.jhazmat.2019.121712

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 32

  García-CórcolesM. T.Rodríguez-GómezR.De Alarcón-GómezB.ÇipaM.Martín-PozoL.KauffmannJ.-M.et al (2019). Chromatographic methods for the determination of emerging contaminants in natural water and wastewater samples: a review. Crit. Rev. Anal. Chem.49 (2), 160–186. 10.1080/10408347.2018.1496010

  • CrossRef
  • Google Scholar

• 33

  García-MartínezJ.AkasoglouI.GolovkoO.FreyM.HenserC.SpégelP.et al (2025). Evaluating removal efficiency in wastewater treatment plants using non-target screening analysis. J. Environ. Manag.395, 127820. 10.1016/j.jenvman.2025.127820

  • CrossRef
  • Google Scholar

• 34

  GhasteM.MistrikR.ShulaevV. (2016). Applications of fourier transform ion cyclotron resonance (FT-ICR) and orbitrap based high resolution mass spectrometry in metabolomics and lipidomics. Int. J. Mol. Sci.17 (6), 816. 10.3390/ijms17060816

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 35

  GloaguenY.KirwanJ. A.BeuleD. (2022). Deep learning-assisted peak curation for large-scale LC-MS metabolomics. Anal. Chem.94 (12), 4930–4937. 10.1021/acs.analchem.1c02220

  • CrossRef
  • Google Scholar

• 36

  GoldmanS.LiJ.ColeyC. W. (2024). Generating molecular fragmentation graphs with autoregressive neural networks. Anal. Chem.96 (8), 3419–3428. 10.1021/acs.analchem.3c04654

  • CrossRef
  • Google Scholar

• 37

  GollongG.NeuwaldI. J.KuckelkornJ.JunekR.ZahnD. (2022). Assessing the protection gap for Mobile and persistent chemicals during advanced water treatment – a study in a drinking water production and wastewater treatment plant. Water Res.221 (Scopus), 118847. 10.1016/j.watres.2022.118847

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 38

  GrosM.BlumK. M.JernstedtH.RenmanG.Rodríguez-MozazS.HaglundP.et al (2017). Screening and prioritization of micropollutants in wastewaters from on-site sewage treatment facilities. J. Hazard. Mater.328, 37–45. 10.1016/j.jhazmat.2016.12.055

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 39

  GuldeR.RutschM.ClercB.ScholléeJ. E.Von GuntenU.McArdellC. S. (2021). Formation of transformation products during ozonation of secondary wastewater effluent and their fate in post-treatment: from laboratory-to full-scale. Water Res.200, 117200. 10.1016/j.watres.2021.117200

  • CrossRef
  • Google Scholar

• 40

  GutierrezM.Mutavdžić PavlovićD.StipaničevD.RepecS.AvolioF.ZanellaM.et al (2024). A thorough analysis of the occurrence, removal and environmental risks of organic micropollutants in a full-scale hybrid membrane bioreactor fed by hospital wastewater. Sci. Total Environ.914, 169848. 10.1016/j.scitotenv.2023.169848

  • CrossRef
  • Google Scholar

• 41

  GuyC.DuportéG.LuquotL.GomezE. (2024). Non-target screening to track contaminant removal and release during nature-based water treatment. Front. Environ. Sci.12, 1385806. 10.3389/fenvs.2024.1385806

  • CrossRef
  • Google Scholar

• 42

  HajebP.ZhuL.BossiR.VorkampK. (2022). Sample preparation techniques for suspect and non-target screening of emerging contaminants. Chemosphere287, 132306. 10.1016/j.chemosphere.2021.132306

  • CrossRef
  • Google Scholar

• 43

  HalwaturaL. M.AgaD. S. (2023). Broad-range extraction of highly polar to non-polar organic contaminants for inclusive target analysis and suspect screening of environmental samples. Sci. Total Environ.893, 164707. 10.1016/j.scitotenv.2023.164707

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 44

  HanY.HuL.-X.LiuT.DongL.-L.LiuY.-S.ZhaoJ.-L.et al (2024). Discovering transformation products of pharmaceuticals in domestic wastewaters and receiving Rivers by using non-target screening and machine learning approaches. Sci. Total Environ.948, 174715. 10.1016/j.scitotenv.2024.174715

  • CrossRef
  • Google Scholar

• 45

  HinnenkampV.BalsaaP.SchmidtT. C. (2022). Target, suspect and non-target screening analysis from wastewater treatment plant effluents to drinking water using collision cross section values as additional identification criterion. Anal. Bioanal. Chem.414 (1), 425–438. 10.1007/s00216-021-03263-1

  • CrossRef
  • Google Scholar

• 46

  HohrenkL. L.VosoughM.SchmidtT. C. (2019). Implementation of chemometric tools to improve data mining and prioritization in LC-HRMS for nontarget screening of organic micropollutants in complex water matrixes. Anal. Chem.91 (14), 9213–9220. 10.1021/acs.analchem.9b01984

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 47

  HohrenkL. L.ItzelF.BaetzN.TuerkJ.VosoughM.SchmidtT. C. (2020). Comparison of software tools for liquid Chromatography–high-resolution mass spectrometry data processing in nontarget screening of environmental samples. Anal. Chem.92 (2), 1898–1907. 10.1021/acs.analchem.9b04095

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 48

  HollenderJ.SchymanskiE. L.SingerH. P.FergusonP. L. (2017). Nontarget screening with high resolution mass spectrometry in the environment: ready to go?Environ. Sci. and Technol.51 (20), 11505–11512. 10.1021/acs.est.7b02184

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 49

  HollenderJ.Van BavelB.DulioV.FarmenE.FurtmannK.KoschorreckJ.et al (2019). High resolution mass spectrometry-based non-target screening can support regulatory environmental monitoring and chemicals management. Environ. Sci. Eur.31 (1), 42. 10.1186/s12302-019-0225-x

  • CrossRef
  • Google Scholar

• 50

  HollenderJ.SchymanskiE. L.AhrensL.AlygizakisN.BéenF.BijlsmaL.et al (2023). NORMAN guidance on suspect and non-target screening in environmental monitoring. Environ. Sci. Eur.35 (1), 75. 10.1186/s12302-023-00779-4

  • CrossRef
  • Google Scholar

• 51

  HongY.LiS.WelchC. J.TichyS.YeY.TangH. (2023). 3DMolMS: prediction of tandem mass spectra from 3D molecular conformations. Bioinformatics39 (6), btad354. 10.1093/bioinformatics/btad354

  • CrossRef
  • Google Scholar

• 52

  HuJ.LyuY.ChenH.LiS.SunW. (2023). Suspect and nontarget screening reveal the underestimated risks of antibiotic transformation products in wastewater treatment plant effluents. Environ. Sci. and Technol.57 (45), 17439–17451. 10.1021/acs.est.3c05008

  • CrossRef
  • Google Scholar

• 53

  Huidobro-LópezB.LeónC.López-HerasI.Martínez-HernándezV.NozalL.CregoA. L.et al (2023). Untargeted metabolomic analysis to explore the impact of soil amendments in a non-conventional wastewater treatment. Sci. Total Environ.870, 161890. 10.1016/j.scitotenv.2023.161890

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 54

  HupatzH.RahuI.WangW.-C.PeetsP.PalmE. H.KruveA. (2024). Critical review on in silico methods for structural annotation of chemicals detected with LC/HRMS non-targeted screening. Anal. Bioanal. Chem.417, 473–493. 10.1007/s00216-024-05471-x

  • CrossRef
  • Google Scholar

• 55

  HuynhN.CauposE.Soares PeireraC.Le RouxJ.BressyA.MoilleronR. (2021). Evaluation of sample preparation methods for non-target screening of organic micropollutants in urban waters using high-resolution mass spectrometry. Molecules26 (23), 7064. 10.3390/molecules26237064

  • CrossRef
  • Google Scholar

• 56

  ItzelF.BaetzN.HohrenkL. L.GehrmannL.AntakyaliD.SchmidtT. C.et al (2020). Evaluation of a biological post-treatment after full-scale ozonation at a municipal wastewater treatment plant. Water Res.170, 115316. 10.1016/j.watres.2019.115316

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 57

  Jaén-GilA.Hom-DiazA.LlorcaM.VicentT.BlánquezP.BarcelóD.et al (2018). An automated on-line turbulent flow liquid-chromatography technology coupled to a high resolution mass spectrometer LTQ-orbitrap for suspect screening of antibiotic transformation products during microalgae wastewater treatment. J. Chromatogr. A1568, 57–68. 10.1016/j.chroma.2018.06.027

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 58

  JenningsE. K.Sierra OleaM.HübnerU.Rodrigues MatosR.ReemtsmaT.LechtenfeldO. J. (2025). Molecular-level insights into recalcitrant ozonation products from effluent organic matter. Environ. Sci. and Technol.59 (1), 823–833. 10.1021/acs.est.4c10212

  • CrossRef
  • Google Scholar

• 59

  JinZ.-L.ChenL.WangY.ShiC.-T.ZhouY.XiaB. (2025). Application of machine learning in LC-MS-based non-targeted analysis. TrAC Trends Anal. Chem.189, 118243. 10.1016/j.trac.2025.118243

  • CrossRef
  • Google Scholar

• 60

  JohnsonT. A.AbrahamssonD. P. (2024). Quantification of chemicals in non-targeted analysis without analytical standards – understanding the mechanism of electrospray ionization and making predictions. Curr. Opin. Environ. Sci. and Health37, 100529. 10.1016/j.coesh.2023.100529

  • CrossRef
  • Google Scholar

• 61

  JohnsonJ. L.DodderN. G.MladenovN.SteinbergL.RichardotW. H.HohE. (2024). Comparison of trace organic chemical removal efficiencies between aerobic and anaerobic membrane bioreactors treating municipal wastewater. ACS ES&T Water4 (4), 1381–1392. 10.1021/acsestwater.3c00542

  • CrossRef
  • Google Scholar

• 62

  KempińskaD.Kot-WasikA. (2018). The use of RP-HPLC–Q-TOF–MS as a powerful tool for wastewater composition profiling and selection of water pollution marker specific to wastewater contamination. Monatsh. Für Chem. - Chem. Mon.149 (9), 1595–1604. 10.1007/s00706-018-2259-y

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 63

  KissA.BergéA.DomenjoudB.Gonzalez-OspinaA.VullietE. (2018). Chemometric and high-resolution mass spectrometry tools for the characterization and comparison of raw and treated wastewater samples of a pilot plant on the SIPIBEL site. Environ. Sci. Pollut. Res.25 (10), 9230–9242. 10.1007/s11356-017-0748-x

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 64

  KopperiM.RiekkolaM.-L. (2016). Non-targeted evaluation of selectivity of water-compatible class selective adsorbents for the analysis of steroids in wastewater. Anal. Chim. Acta920, 47–53. 10.1016/j.aca.2016.03.036

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 65

  KronikO. M.ChristensenJ. H.NielsenN. J. (2024). Instrumental and theoretical advancements in pulsed elution-LC × LC: investigation of pulse parameters and application to wastewater effluent. J. Chromatogr. A1730, 465079. 10.1016/j.chroma.2024.465079

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 66

  KronsbeinA. L.WallisR. M.WinterE.BandowN.JewellK. S.DierkesG.et al (2026). Use cases from the German NTSPortal on the systematic use of high-resolution mass spectrometry non-target screening data in environmental monitoring and chemicals management. Environ. Sci. Eur.38 (1), 85. 10.1186/s12302-026-01368-x

  • CrossRef
  • Google Scholar

• 67

  Ledesma-EscobarC. A.Priego-CapoteF.Calderón-SantiagoM. (2023). MetaboMSDIA: a tool for implementing data-independent acquisition in metabolomic-based mass spectrometry analysis. Anal. Chim. Acta1266, 341308. 10.1016/j.aca.2023.341308

  • CrossRef
  • Google Scholar

• 68

  LennonS.ChakerJ.PriceE. J.HollenderJ.HuberC.SchulzeT.et al (2024). Harmonized quality assurance/quality control provisions to assess completeness and robustness of MS1 data preprocessing for LC-HRMS-based suspect screening and non-targeted analysis. TrAC Trends Anal. Chem.174, 117674. 10.1016/j.trac.2024.117674

  • CrossRef
  • Google Scholar

• 69

  LiZ.UndemanE.PapaE.McLachlanM. S. (2018). High-throughput evaluation of organic contaminant removal efficiency in a wastewater treatment plant using direct injection UHPLC-Orbitrap-MS/MS. Environ. Sci. Process. and Impacts20 (3), 561–571. 10.1039/C7EM00552K

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 70

  LiX.ChengF.QiuJ.LingY.YouJ. (2026). Advancing the identification of toxic by-products in disinfected water by integrating effect-directed analysis and nontarget screening. J. Hazard. Mater.513, 142536. 10.1016/j.jhazmat.2026.142536

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 71

  LiY.ZhangX.LiY.AbrahamssonD.GaoY.WuW.et al (2026). High-risk compounds and transformation products in wastewater treatment plants: a non-targeted analysis workflow coupled with rapid risk prioritization. Environ. Res.301, 124510. 10.1016/j.envres.2026.124510

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 72

  LinD.WangZ.LiaoJ.LiN.LiX. (2026). From analysis to assessment: machine learning for non-target screening of pollutants using chromatography coupled with (ion mobility) mass spectrometry. Toxics14 (4), 322. 10.3390/toxics14040322

  • CrossRef
  • Google Scholar

• 73

  LiuY.XiongH.LiuJ.XieH.ChenJ. (2026). Application of machine learning in non-targeted analysis for environmental organic pollutants. Artif. Intell. and Environ.1 (1), 11–22. 10.66178/aie-0026-0003

  • CrossRef
  • Google Scholar

• 74

  Lopez-HerguedasN.IrazolaM.Alvarez-MoraI.MijangosL.BilbaoD.EtxebarriaN.et al (2024). Evaluating membrane bioreactor treatment for the elimination of emerging contaminants using different analytical methods. J. Hazard. Mater.463, 132833. 10.1016/j.jhazmat.2023.132833

  • CrossRef
  • Google Scholar

• 75

  ManasfiT.HouskaJ.GebhardtI.Von GuntenU. (2023). Formation of carbonyl compounds during ozonation of Lake water and wastewater: development of a non-target screening method and quantification of target compounds. Water Res.237, 119751. 10.1016/j.watres.2023.119751

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 76

  MandlaziN. B.VilakatiG. D.MadikizelaL.KgomoH. (2026). Non-targeted screening of emerging pollutants across wastewater and receiving waters: a case study in Eswatini. Sci. Total Environ.1014, 181353. 10.1016/j.scitotenv.2026.181353

  • CrossRef
  • Google Scholar

• 77

  ManzK. E.FeerickA.BraunJ. M.FengY.-L.HallA.KoelmelJ.et al (2023). Non-targeted analysis (NTA) and suspect screening analysis (SSA): a review of examining the chemical exposome. J. Expo. Sci. and Environ. Epidemiol.33 (4), 524–536. 10.1038/s41370-023-00574-6

  • CrossRef
  • Google Scholar

• 78

  McCordJ. P.GroffL. C.SobusJ. R. (2022). Quantitative non-targeted analysis: bridging the gap between contaminant discovery and risk characterization. Environ. Int.158, 107011. 10.1016/j.envint.2021.107011

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 79

  McLachlanM. S.LiZ.JonssonL.KaserzonS.O’BrienJ. W.MuellerJ. F. (2022). Removal of 293 organic compounds in 15 WWTPs studied with non-targeted suspect screening. Environ. Sci. Water Res. and Technol.8 (7), 1423–1433. 10.1039/D2EW00088A

  • CrossRef
  • Google Scholar

• 80

  MelnikovA. D.TsentalovichY. P.YansholeV. V. (2020). Deep learning for the precise peak detection in high-resolution LC–MS data. Anal. Chem.92 (1), 588–592. 10.1021/acs.analchem.9b04811

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 81

  MengerF.Gago-FerreroP.WibergK.AhrensL. (2020). Wide-scope screening of polar contaminants of concern in water: a critical review of liquid chromatography-high resolution mass spectrometry-based strategies. Trends Environ. Anal. Chem.28, e00102. 10.1016/j.teac.2020.e00102

  • CrossRef
  • Google Scholar

• 82

  MilmanB. L. (2015). General principles of identification by mass spectrometry. TrAC Trends Anal. Chem.69, 24–33. 10.1016/j.trac.2014.12.009

  • CrossRef
  • Google Scholar

• 83

  MinkusS.BieberS.LetzelT. (2022a). Characterizing powdered activated carbon treatment of surface water samples using polarity-extended non-target screening analysis. Molecules27 (16), Scopus. 10.3390/molecules27165214

  • CrossRef
  • Google Scholar

• 84

  MinkusS.BieberS.LetzelT. (2022b). Spotlight on mass spectrometric non‐target screening analysis: advanced data processing methods recently communicated for extracting, prioritizing and quantifying features. Anal. Sci. Adv.3 (3–4), 103–112. 10.1002/ansa.202200001

  • CrossRef
  • Google Scholar

• 85

  MladenovN.DodderN. G.SteinbergL.RichardotW.JohnsonJ.MartincighB. S.et al (2022). Persistence and removal of trace organic compounds in centralized and decentralized wastewater treatment systems. Chemosphere286, 131621. 10.1016/j.chemosphere.2021.131621

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 86

  MokS.LeeS.ChoiY.JeonJ.KimY. H.MoonH.-B. (2023). Target and non-target analyses of neutral per- and polyfluoroalkyl substances from fluorochemical industries using GC-MS/MS and GC-TOF: insights on their environmental fate. Environ. Int.182, 108311. 10.1016/j.envint.2023.108311

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 87

  Molnár KarlssonT.ChristensenJ. H. (2025). Overcoming matrix effects in non-target screening: a novel strategy for urban runoff analysis. Anal. Chem.97 (25), 13274–13281. 10.1021/acs.analchem.5c01303

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 88

  MotteauS.DebordeM.GombertB.Karpel Vel LeitnerN. (2024). Non-target analysis for water characterization: wastewater treatment impact and selection of relevant features. Environ. Sci. Pollut. Res.31 (3), 4154–4173. 10.1007/s11356-023-30972-0

  • CrossRef
  • Google Scholar

• 89

  MurrayK. K.BoydR. K.EberlinM. N.LangleyG. J.LiL.NaitoY. (2013). Definitions of terms relating to mass spectrometry (IUPAC recommendations 2013). Pure Appl. Chem.85 (7), 1515–1609. 10.1351/PAC-REC-06-04-06

  • CrossRef
  • Google Scholar

• 90

  MurrellK. A.DormanF. L. (2021). A comparison of liquid-liquid extraction and stir bar sorptive extraction for multiclass organic contaminants in wastewater by comprehensive two-dimensional gas chromatography time of flight mass spectrometry. Talanta221, 121481. 10.1016/j.talanta.2020.121481

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 91

  NasonS. L.McCordJ.FengY.-L.SobusJ. R.FisherC. M.Marfil-VegaR.et al (2025). Communicating with stakeholders to identify high-impact research directions for non-targeted analysis. Anal. Chem.97 (5), 2567–2578. 10.1021/acs.analchem.4c04801

  • CrossRef
  • Google Scholar

• 92

  NihemaitiM.HuynhN.MaillerR.Mèche-AnanitP.RocherV.BarhdadiR.et al (2022). High-resolution mass spectrometry screening of wastewater effluent for micropollutants and their transformation products during disinfection with performic acid. ACS ES&T Water2 (7), 1225–1233. 10.1021/acsestwater.2c00075

  • CrossRef
  • Google Scholar

• 93

  NovákováP.ŠvecováH.BoříkA.GrabicR. (2023). Novel nontarget LC-HRMS-based approaches for evaluation of drinking water treatment. Environ. Monit. Assess.195 (6), 739. 10.1007/s10661-023-11348-w

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 94

  NowatzkyY.RussoF. F.LisecJ.KisterA.ReinertK.MuthT.et al (2025). FIORA: local neighborhood-based prediction of compound mass spectra from single fragmentation events. Nat. Commun.16 (1), 2298. 10.1038/s41467-025-57422-4

  • CrossRef
  • Google Scholar

• 95

  NürenbergG.SchulzM.KunkelU.TernesT. A. (2015). Development and validation of a generic nontarget method based on liquid chromatography – high resolution mass spectrometry analysis for the evaluation of different wastewater treatment options. J. Chromatogr. A1426, 77–90. 10.1016/j.chroma.2015.11.014

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 96

  NürenbergG.KunkelU.WickA.FalåsP.JossA.TernesT. A. (2019). Nontarget analysis: a new tool for the evaluation of wastewater processes. Water Res.163, 114842. 10.1016/j.watres.2019.07.009

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 97

  OuyangX.LeonardsP.LeglerJ.Van Der OostR.De BoerJ.LamoreeM. (2015). Comprehensive two-dimensional liquid chromatography coupled to high resolution time of flight mass spectrometry for chemical characterization of sewage treatment plant effluents. J. Chromatogr. A1380, 139–145. 10.1016/j.chroma.2014.12.075

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 98

  Padilla-GonzálezF.BloklandM.AlewijnM. (2026). Processing of non-targeted LC-HRMS data: progress and challenges in automated data-driven optimisations. TrAC Trends Anal. Chem.202, 118964. 10.1016/j.trac.2026.118964

  • CrossRef
  • Google Scholar

• 99

  PandeyA.KasugaI.FurumaiH.KurisuF. (2024). Non-target liquid chromatography high-resolution mass spectrometry screening to prioritize unregulated micropollutants that persist through domestic wastewater treatment. Sci. Total Environ.947, 174486. 10.1016/j.scitotenv.2024.174486

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 100

  ParryE.YoungT. M. (2016). Comparing targeted and non-targeted high-resolution mass spectrometric approaches for assessing advanced oxidation reactor performance. Water Res.104, 72–81. 10.1016/j.watres.2016.07.056

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 101

  PaszkiewiczM.GodlewskaK.LisH.CabanM.Białk-BielińskaA.StepnowskiP. (2022). Advances in suspect screening and non-target analysis of polar emerging contaminants in the environmental monitoring. TrAC Trends Anal. Chem.154, 116671. 10.1016/j.trac.2022.116671

  • CrossRef
  • Google Scholar

• 102

  PerkonsI.RuskoJ.ZacsD.BartkevicsV. (2021). Rapid determination of pharmaceuticals in wastewater by direct infusion HRMS using target and suspect screening analysis. Sci. Total Environ.755, 142688. 10.1016/j.scitotenv.2020.142688

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 103

  PeterK. T.HerzogS.TianZ.WuC.McCrayJ. E.LynchK.et al (2019). Evaluating emerging organic contaminant removal in an engineered hyporheic zone using high resolution mass spectrometry. Water Res.150, 140–152. 10.1016/j.watres.2018.11.050

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 104

  PeterK. T.PhillipsA. L.KnolhoffA. M.GardinaliP. R.ManzanoC. A.MillerK. E.et al (2021). Nontargeted analysis study reporting tool: a framework to improve research transparency and reproducibility. Anal. Chem.93 (41), 13870–13879. 10.1021/acs.analchem.1c02621

  • CrossRef
  • Google Scholar

• 105

  PetrieB.BardenR.Kasprzyk-HordernB. (2015). A review on emerging contaminants in wastewaters and the environment: current knowledge, understudied areas and recommendations for future monitoring. Water Res.72, 3–27. 10.1016/j.watres.2014.08.053

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 106

  PetromelidouS.EvgenidouE.TziouvalekasM.LambropoulouD. A. (2024). Unravelling psychoactive substances and their metabolites and transformation products: high-resolution mass spectrometry approaches for comprehensive target and suspect screening in wastewater. Sci. Total Environ.931, 172867. 10.1016/j.scitotenv.2024.172867

  • CrossRef
  • Google Scholar

• 107

  PirokB. W. J.StollD. R.SchoenmakersP. J. (2019). Recent developments in two-dimensional liquid chromatography: fundamental improvements for practical applications. Anal. Chem.91 (1), 240–263. 10.1021/acs.analchem.8b04841

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 108

  Ponce-RoblesL.OllerI.AgüeraA.Trinidad-LozanoM. J.YusteF. J.MalatoS.et al (2018). Application of a multivariate analysis method for non-target screening detection of persistent transformation products during the Cork boiling wastewater treatment. Sci. Total Environ.633, 508–517. 10.1016/j.scitotenv.2018.03.179

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 109

  PrasseC.StalterD.Schulte-OehlmannU.OehlmannJ.TernesT. A. (2015). Spoilt for choice: a critical review on the chemical and biological assessment of current wastewater treatment technologies. Water Res.87, 237–270. 10.1016/j.watres.2015.09.023

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 110

  PurschkeK.ZoellC.LeonhardtJ.WeberM.SchmidtT. C. (2020). Identification of unknowns in industrial wastewater using offline 2D chromatography and non-target screening. Sci. Total Environ.706, 135835. 10.1016/j.scitotenv.2019.135835

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 111

  QianY.WangX.WuG.WangL.GengJ.YuN.et al (2021). Screening priority indicator pollutants in full-scale wastewater treatment plants by non-target analysis. J. Hazard. Mater.414, Scopus. 10.1016/j.jhazmat.2021.125490

  • CrossRef
  • Google Scholar

• 112

  QiuJ.LüF.ZhangH.ShaoL.HeP. (2021). Data mining strategies of molecular information for inspecting wastewater treatment by using UHRMS. Trends Environ. Anal. Chem.31, e00134. 10.1016/j.teac.2021.e00134

  • CrossRef
  • Google Scholar

• 113

  QiuJ.LüF.LiT.ZhangH.HeP. (2022). A novel 4-Set venn diagram model based on high-resolution mass spectrometry to monitor wastewater treatment. Environ. Sci. and Technol.56 (20), 14753–14762. 10.1021/acs.est.2c02229

  • CrossRef
  • Google Scholar

• 114

  RennerG.ReuschenbachM. (2023). Critical review on data processing algorithms in non-target screening: challenges and opportunities to improve result comparability. Anal. Bioanal. Chem.415 (18), 4111–4123. 10.1007/s00216-023-04776-7

  • CrossRef
  • Google Scholar

• 115

  ReverbelS.DévierM.-H.DuprazV.GenesteE.BudzinskiH. (2023). Assessment of the presence of transformation products of certain pharmaceutical products (psychotropic family) by suspect and non-targeted HRMS screening in wastewater treatment plants. Toxics11 (8), 713. 10.3390/toxics11080713

  • CrossRef
  • Google Scholar

• 116

  Rodriguez-OteroA.TislerS.ReinhardtL. M.JørgensenM. B.PattisonD. I.BouyssiereB.et al (2026). Developing a cost and time efficient strategy for assessing novel materials for wastewater treatment with non-target screening analysis. J. Hazard. Mater.503, 141115. 10.1016/j.jhazmat.2026.141115

  • CrossRef
  • Google Scholar

• 117

  RoutP. R.ZhangT. C.BhuniaP.SurampalliR. Y. (2021). Treatment technologies for emerging contaminants in wastewater treatment plants: a review. Sci. Total Environ.753, 141990. 10.1016/j.scitotenv.2020.141990

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 118

  RuanT.LiP.WangH.LiT.JiangG. (2023). Identification and prioritization of environmental organic pollutants: from an analytical and toxicological perspective. Chem. Rev.123 (17), 10584–10640. 10.1021/acs.chemrev.3c00056

  • CrossRef
  • Google Scholar

• 119

  Saint GermainF. M.FaureK.SaunierE.LerestifJ.-M.HeinischS. (2022). On-line 2D-RPLC x RPLC – HRMS to assess wastewater treatment in a pharmaceutical plant. J. Pharm. Biomed. Analysis208, 114465. 10.1016/j.jpba.2021.114465

  • CrossRef
  • Google Scholar

• 120

  SapkotaB.PariatambyA. (2025). Comprehensive screening and analysis of pharmaceuticals and pharmaceutically active chemicals in wastewater: health and environmental hazards and removal efficiency of wastewater treatment plant in Malaysia. Environ. Sci. Pollut. Res.32 (5), 2577–2591. 10.1007/s11356-024-35817-y

  • CrossRef
  • Google Scholar

• 121

  SchneideP.-A.Munk KronikO. (2025). Signal processing workflow for suspect screening in LC × LC-HRMS: efficient extraction of pure mass spectra for identification of suspects in complex samples using a mass filtering algorithm. Anal. Chem.97 (2), 1180–1189. 10.1021/acs.analchem.4c04288

  • CrossRef
  • Google Scholar

• 122

  ScholléeJ. E.SchymanskiE. L.AvakS. E.LoosM.HollenderJ. (2015). Prioritizing unknown transformation products from biologically-treated wastewater using high-resolution mass spectrometry, multivariate statistics, and metabolic logic. Anal. Chem.87 (24), 12121–12129. 10.1021/acs.analchem.5b02905

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 123

  ScholléeJ. E.SchymanskiE. L.HollenderJ. (2016). “Statistical approaches for LC-HRMS data to characterize, prioritize, and identify transformation products from water treatment processes. ACS Symposium Series Editors DrewesJ. E.LetzelT. (American Chemical Society), 1241, 45–65. 10.1021/bk-2016-1241.ch004

  • CrossRef
  • Google Scholar

• 124

  ScholléeJ. E.BourginM.Von GuntenU.McArdellC. S.HollenderJ. (2018). Non-target screening to trace ozonation transformation products in a wastewater treatment train including different post-treatments. Water Res.142, 267–278. 10.1016/j.watres.2018.05.045

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 125

  ScholléeJ. E.HollenderJ.McArdellC. S. (2021). Characterization of advanced wastewater treatment with ozone and activated carbon using LC-HRMS based non-target screening with automated trend assignment. Water Res.200, 117209. 10.1016/j.watres.2021.117209

  • CrossRef
  • Google Scholar

• 126

  SchulzeB.JeonY.KaserzonS.HeffernanA. L.DewapriyaP.O’BrienJ.et al (2020). An assessment of quality assurance/quality control efforts in high resolution mass spectrometry non-target workflows for analysis of environmental samples. TrAC Trends Anal. Chem.133, 116063. 10.1016/j.trac.2020.116063

  • CrossRef
  • Google Scholar

• 127

  SchymanskiE. L.JeonJ.GuldeR.FennerK.RuffM.SingerH. P.et al (2014). Identifying small molecules via high resolution mass spectrometry: communicating confidence. Environ. Sci. and Technol.48 (4), 2097–2098. 10.1021/es5002105

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 128

  SchymanskiE. L.SingerH. P.SlobodnikJ.IpolyiI. M.OswaldP.KraussM.et al (2015). Non-target screening with high-resolution mass spectrometry: critical review using a collaborative trial on water analysis. Anal. Bioanal. Chem.407 (21), 6237–6255. 10.1007/s00216-015-8681-7

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 129

  SepmanH.MalmL.PeetsP.KruveA. (2023). Scientometric review: concentration and toxicity assessment in environmental non-targeted LC/HRMS analysis. Trends Environ. Anal. Chem.40, e00217. 10.1016/j.teac.2023.e00217

  • CrossRef
  • Google Scholar

• 130

  SinghV. K.SinghP.KarmakarM.LetaJ.MayrP. (2021). The journal coverage of web of science, scopus and dimensions: a comparative analysis. Scientometrics126 (6), 5113–5142. 10.1007/s11192-021-03948-5

  • CrossRef
  • Google Scholar

• 131

  SobotkaJ.LammelG.SlobodníkJ.SchinkA.ProkešR.VranaB. (2021). Dynamic passive sampling of hydrophobic organic compounds in surface seawater along the south Atlantic Ocean east-to-west transect and across the black sea. Mar. Pollut. Bull.168, 112375. 10.1016/j.marpolbul.2021.112375

  • CrossRef
  • Google Scholar

• 132

  SobusJ. R.WambaughJ. F.IsaacsK. K.WilliamsA. J.McEachranA. D.RichardA. M.et al (2018). Integrating tools for non-targeted analysis research and chemical safety evaluations at the US EPA. J. Expo. Sci. and Environ. Epidemiol.28 (5), 411–426. 10.1038/s41370-017-0012-y

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 133

  SobusJ. R.Sayre-SmithN. A.ChaoA.FerlandT. M.MinucciJ. M.CarrE. T.et al (2025). Automated QA/QC reporting for non-targeted analysis: a demonstration of “INTERPRET NTA” with de facto water reuse data. Anal. Bioanal. Chem.417 (9), 1897–1914. 10.1007/s00216-025-05771-w

  • CrossRef
  • Google Scholar

• 134

  SongX.-C.CanellasE.DreolinN.GoshawkJ.LvM.QuG.et al (2023). Application of ion mobility spectrometry and the derived collision cross section in the analysis of environmental organic micropollutants. Environ. Sci. and Technol.57 (51), 21485–21502. 10.1021/acs.est.3c03686

  • CrossRef
  • Google Scholar

• 135

  SouihiA.MohaiM. P.MartinJ. W.KruveA. (2023). Mobile phase and column chemistry selection for high sensitivity non-targeted LC/ESI/HRMS screening of water. Anal. Chim. Acta1274, 341573. 10.1016/j.aca.2023.341573

  • CrossRef
  • Google Scholar

• 136

  StancliffeE.PattiG. J. (2023). PeakDetective: a semisupervised deep learning-based approach for peak curation in untargeted metabolomics. Anal. Chem.95 (25), 9397–9403. 10.1021/acs.analchem.3c00764

  • CrossRef
  • Google Scholar

• 137

  StollD. R.CarrP. W. (2017). Two-dimensional liquid chromatography: a state of the art tutorial. Anal. Chem.89 (1), 519–531. 10.1021/acs.analchem.6b03506

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 138

  SzaboD.FalconerT. M.FisherC. M.HeiseT.PhillipsA. L.VasG.et al (2024). Online and offline prioritization of chemicals of interest in suspect screening and non-targeted screening with high-resolution mass spectrometry. Anal. Chem.96 (9), 3707–3716. 10.1021/acs.analchem.3c05705

  • CrossRef
  • Google Scholar

• 139

  TadićĐ.ManasfiR.BertrandM.SauvêtreA.ChironS. (2022). Use of passive and grab sampling and high-resolution mass spectrometry for non-targeted analysis of emerging contaminants and their semi-quantification in water. Molecules27 (10), 3167. 10.3390/molecules27103167

  • CrossRef
  • Google Scholar

• 140

  TanX.SunS.HouS.WuL.BaiY.XiH.et al (2025). Non-targeted screening and characterization of organic pollutants in wastewater from an industrial park in China. Emerg. Contam.11 (3), Scopus. 10.1016/j.emcon.2025.100519

  • CrossRef
  • Google Scholar

• 141

  TianY.JiangF.LiuN.XuL.CaiY. (2024). Non-target analysis of organic pollutants in oil-production wastewater treatment stations and surrounding soils: their profiles, electro-transformation, and environmental risks. Chemosphere368, 143779. 10.1016/j.chemosphere.2024.143779

  • CrossRef
  • Google Scholar

• 142

  TintropL. K.BräklingS.VetterM.EßerW.DreesF.SalemiA.et al (2024). Evaluation of GC-EI&CI-TOFMS for nontarget analysis of industrial wastewater using hydrophilic-lipophilic-balanced SPME. Anal. Chem.96 (16), 6122–6130. 10.1021/acs.analchem.3c04114

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 143

  TislerS.EnglerN.JørgensenM. B.KilpinenK.TomasiG.ChristensenJ. H. (2022). From data to reliable conclusions: identification and comparison of persistent micropollutants and transformation products in 37 wastewater samples by non-target screening prioritization. Water Res.219, 118599. 10.1016/j.watres.2022.118599

  • CrossRef
  • Google Scholar

• 144

  TislerS.PattisonD. I.ChristensenJ. H. (2021). Correction of matrix effects for reliable non-target screening LC–ESI–MS analysis of wastewater. Anal. Chem.93 (24), 8432–8441. 10.1021/acs.analchem.1c00357

  • CrossRef
  • Google Scholar

• 145

  TislerS.KilpinenK.DeversJ.CastroM.JørgensenM. B.MandavaG.et al (2025). Mapping emerging contaminants in wastewater effluents through multichromatographic platform analysis and source correlations. Environ. Sci. and Technol.59 (11), 5766–5774. 10.1021/acs.est.5c01139

  • CrossRef
  • Google Scholar

• 146

  TislerS.TüchsenP. L.ChristensenJ. H. (2022). Non-target screening of micropollutants and transformation products for assessing AOP-BAC treatment in groundwater. Environ. Pollut.309, 119758. 10.1016/j.envpol.2022.119758

  • CrossRef
  • Google Scholar

• 147

  TuK.LiuY.ChenX.LiW.XuZ.QiaC.et al (2026). Insights into contaminant composition variations and reactomics during wastewater treatment processes based on nontargeted analysis and paired mass distance. Environ. Sci. and Technol.60 (6), 5086–5096. 10.1021/acs.est.5c14774

  • CrossRef
  • Google Scholar

• 148

  Ulucan-AltuntasK.Manav-DemirN.IlhanF.GelgorH. B.HuddersmanK.TiwaryA.et al (2023). Emerging pollutants removal in full-scale biological treatment plants: a case study. J. Water Process Eng.51, 103336. 10.1016/j.jwpe.2022.103336

  • CrossRef
  • Google Scholar

• 149

  VerkhY.RozmanM.PetrovicM. (2018). A non-targeted high-resolution mass spectrometry data analysis of dissolved organic matter in wastewater treatment. Chemosphere200, 397–404. 10.1016/j.chemosphere.2018.02.095

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 150

  VosoughM.SchmidtT. C.RennerG. (2024). Non-target screening in water analysis: recent trends of data evaluation, quality assurance, and their future perspectives. Anal. Bioanal. Chem.416 (9), 2125–2136. 10.1007/s00216-024-05153-8

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 151

  WangA.-G.ZhangJ.-N.HuL.-X.LiuY.-S.YingG.-G.LiangY.-X.et al (2025). Fate and transformation of psychotropic drugs in urban wastewater systems and receiving Rivers via the integration of targeted and suspect screening analysis. Water Res.276, 123292. 10.1016/j.watres.2025.123292

  • CrossRef
  • Google Scholar

• 152

  WangY.YuN.ZhuX.GuoH.JiangJ.WangX.et al (2018). Suspect and nontarget screening of Per- and polyfluoroalkyl substances in wastewater from a fluorochemical manufacturing park. Environ. Sci. and Technol.52 (19), 11007–11016. 10.1021/acs.est.8b03030

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 153

  WangY.GaoW.WangY.JiangG. (2019). Suspect screening analysis of the occurrence and removal of micropollutants by GC-QTOF MS during wastewater treatment processes. J. Hazard. Mater.376, 153–159. 10.1016/j.jhazmat.2019.05.031

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 154

  WangX.YuN.YangJ.JinL.GuoH.ShiW.et al (2020). Suspect and non-target screening of pesticides and pharmaceuticals transformation products in wastewater using QTOF-MS. Environ. Int.137, 105599. 10.1016/j.envint.2020.105599

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 155

  WangH.FengX.SuW.ZhongL.LiuY.LiangY.et al (2024). Identifying organic chemicals with acetylcholinesterase inhibition in nationwide estuarine waters by machine learning-assisted mass spectrometric screening. Environ. Sci. and Technol.58 (50), 22379–22390. 10.1021/acs.est.4c10230

  • CrossRef
  • Google Scholar

• 156

  WangH.KuoT.-C.TsengY. J. (2025). DeePFAS: deep-learning-enabled rapid annotation of PFAS: enhancing nontargeted screening through spectral encoding and latent space analysis. Environ. Sci. and Technol.59 (46), 24841–24852. 10.1021/acs.est.5c09769

  • CrossRef
  • Google Scholar

• 157

  WenY.DongL.XuJ.ZhangX.LianL.GuanX. (2024). Molecular insights into the transformation of dissolved organic matter in a full-scale wastewater treatment plant. ACS ES&T Water4 (9), 4228–4238. 10.1021/acsestwater.4c00534

  • CrossRef
  • Google Scholar

• 158

  WiestL.GossetA.FildierA.LibertC.HervéM.SibeudE.et al (2021). Occurrence and removal of emerging pollutants in urban sewage treatment plants using LC-QToF-MS suspect screening and quantification. Sci. Total Environ.774, 145779. 10.1016/j.scitotenv.2021.145779

  • CrossRef
  • Google Scholar

• 159

  WishartD. S.TianS.AllenD.OlerE.PetersH.LuiV. W.et al (2022). BioTransformer 3.0—A web server for accurately predicting metabolic transformation products. Nucleic Acids Res.50 (W1), W115–W123. 10.1093/nar/gkac313

  • CrossRef
  • Google Scholar

• 160

  WuG.WangX.ZhangX.RenH.WangY.YuQ.et al (2023). Nontarget screening based on molecular networking strategy to identify transformation products of citalopram and sertraline in wastewater. Water Res.232, 119509. 10.1016/j.watres.2022.119509

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 161

  WuY.ShiY.GuT.DuX.DuZ.ZhangC.et al (2024). Global trends and hotspots in non-targeted screening of water pollution research: bibliometric and visual analysis. Toxics12 (12), 844. 10.3390/toxics12120844

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 162

  XiaY.SunX.YangL.WangX.LiY.WangM.et al (2025). Structural molecular network for nontargeted screening and prioritization of new pollutants in urban wastewater. Environ. Sci. and Technol.59 (33), 17846–17856. 10.1021/acs.est.5c07353

  • CrossRef
  • Google Scholar

• 163

  XiaoS.LiuT.HuL.-X.YangB.YingG.-G. (2024). Non-target and target screening and risk assessment of per- and polyfluoroalkyl substances in textile wastewater and receiving river. Sci. Total Environ.927, 171876. 10.1016/j.scitotenv.2024.171876

  • CrossRef
  • Google Scholar

• 164

  XuJ.SongY.JiaX.WenY.SunW.GuanX.et al (2024). Comprehensive characterization of the transformation of wastewater effluent via advanced oxidation processes using nontarget mass spectrometric analysis. ACS ES T Water4, 1720–1729. 10.1021/acsestwater.3c00765

  • CrossRef
  • Google Scholar

• 165

  YadavM. K.ShortM. D.AryalR.GerberC.Van Den AkkerB.SaintC. P. (2017). Occurrence of illicit drugs in water and wastewater and their removal during wastewater treatment. Water Res.124, 713–727. 10.1016/j.watres.2017.07.068

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 166

  YangY.YangL.ZhengM.CaoD.LiuG. (2023). Data acquisition methods for non-targeted screening in environmental analysis. TrAC Trends Anal. Chem.160, 116966. 10.1016/j.trac.2023.116966

  • CrossRef
  • Google Scholar

• 167

  YangC.YangJ.LvY.ShiY.CuiD.YinM.et al (2025). Suspect and non-target screening for emerging contaminants of potential concern and risk assessment in wastewater treatment plants: a case study of a typical industrial city in China. Sci. Total Environ.989, Scopus. 10.1016/j.scitotenv.2025.179857

  • CrossRef
  • Google Scholar

• 168

  ZangL.WanY.ZhangH.ZhangY.GaoY.HeY.et al (2022). Characterization of non-volatile organic contaminants in coking wastewater using non-target screening: dominance of nitrogen, sulfur, and oxygen-containing compounds in biological effluents. Sci. Total Environ.837, 155768. 10.1016/j.scitotenv.2022.155768

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 169

  ZhangQ.XuH.SongN.LiuS.WangY.YeF.et al (2023). New insight into fate and transport of organic compounds from pollution sources to aquatic environment using non-targeted screening: a wastewater treatment plant case study. Sci. Total Environ.863, 161031. 10.1016/j.scitotenv.2022.161031

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 170

  ZweigleJ.TislerS.BevilacquaM.TomasiG.NielsenN. J.GawlittaN.et al (2025). Prioritization strategies for non-target screening in environmental samples by chromatography – high-Resolution mass spectrometry: a tutorial. J. Chromatogr. A1751, 465944. 10.1016/j.chroma.2025.465944

  • Pubmed Abstract
  • CrossRef
  • Google Scholar