Nontarget screening with liquid chromatography–high-resolution mass spectrometry (LC-HRMS) detects thousands of environmental chemicals that cannot be structurally identified and therefore lack the analytical standards needed for conventional quantification. MS2Quant (Sepman et al., 2023) bypasses identification by predicting ionization efficiency directly from MS² fragmentation spectra, but it was trained exclusively on positive-mode electrospray ionization data, so it cannot serve chemicals that are detectable only in negative mode. Later literature has already implemented the coupling of MS2Quant with MS2Tox for risk-based prioritization, and has extensively tested predicted-ionization-efficiency quantification across laboratories, methods, and chemical classes — all in positive mode. The surviving open core is the development and validation of a negative-ion-mode MS²-based concentration-estimation model for unidentified chemicals, so that standard-free exposure estimation becomes available in both electrospray polarities. The Kruve group's MS2Quant2 repository (October 2025, 'in development') announces negative-mode support, but no negative-mode model, paper, or preprint has been released to date.

Background

Non-target screening (NTS) with liquid chromatography coupled to high-resolution mass spectrometry (LC-HRMS) is the standard way to detect organic micropollutants — pesticides, pharmaceuticals, industrial chemicals and their transformation products — in environmental and biological samples. In electrospray ionization (ESI), the LC effluent is dispersed into charged droplets and analytes enter the gas phase as ions: in positive mode mostly as protonated molecules [M+H]+[M+H]^+, in negative mode mostly as deprotonated molecules [M−H]−[M-H]^-, with acidic compounds (carboxylic acids, phenols, sulfonates, many per- and polyfluoroalkyl substances) often detectable only in negative mode. Signal intensity depends on the analyte's ionization efficiency (IE) — how readily it is ionized relative to other molecules — so conventional quantification calibrates each compound with an authentic analytical standard via its response factor (the proportionality constant between injected concentration and peak area). In NTS most detected features cannot be structurally identified, so no standard can ever exist for them, and the concentration information needed for exposure and risk assessment is normally unobtainable.

MS2Quant (Sepman et al., 2023, Analytical Chemistry 95:12329–12338) addresses this by predicting IE directly from the tandem mass spectrum (MS², the fragment spectrum produced by collision-induced dissociation of the precursor ion): molecular fingerprints predicted from the MS² spectrum by the SIRIUS+CSI:FingerID framework feed an xgbTree (extreme gradient-boosted tree) regression, trained on IE data for 1191 chemicals spanning eight orders of magnitude, that outputs log⁡\log IE; calibrants measured in the same run then map predicted IE to the instrument-specific response factor, yielding concentration estimates for entirely unidentified features (on the held-out test set the root-mean-square error (RMSE) was 0.80 log⁡\log-units, about 6.3×; validation on 39 environmental pollutants gave a median error of 4.0×). A companion tool, MS2Tox (Peets et al., 2022), predicts fish lethality (LC50, the concentration lethal to 50% of exposed test fish) from MS² spectra by the same identification-free logic.

The training scope of the 2023 model defines the open question: it was trained exclusively on positive-mode ESI data ([M+H]+[M+H]^+ and M+M^+ ions), so it cannot serve the many environmentally important chemicals that ionize in negative mode; although IE prediction from molecular structure was extended to negative-mode ESI as early as Kruve et al. (2014), that route presupposes a confident structural identification, which is exactly what NTS lacks. The source also notes, as a limitation of standard-free concentration estimation, that matrix effects — suppression or enhancement of ionization by co-eluting sample constituents — can occur and cannot be corrected in the usual way with isotope-labeled internal standards, since such standards cannot be prepared for unidentified compounds, while citing evidence that the model prediction error significantly dominates over the error arising from matrix effects (Wang et al., Food Chemistry 2020, 318:126460, DOI 10.1016/j.foodchem.2020.126460); matrix-containing samples therefore remain part of a realistic validation rather than a separate open question. Subsequent work resolved the coupling of MS2Quant with MS2Tox for risk-based prioritization of unidentified features, and extensively tested predicted-IE quantification across laboratories, instrumental setups, and chemical classes — all in positive mode, leaving the negative-mode model as the surviving core.

Problem Statement

The open question, narrowed to the part not resolved by subsequent literature, is: can a negative-ion-mode counterpart of MS2Quant be developed and validated — a model that predicts ionization efficiency, and hence concentration, directly from the MS² spectra of unidentified chemicals detected in negative-mode electrospray LC-HRMS, without structural identification and without analytical standards — so that standard-free exposure estimation becomes available in both ESI polarities? Addressing the question requires a negative-mode model trained on negative-ion ionization-efficiency data spanning a broad chemical space and validated on independent chemicals and matrix-containing samples. Alternatively, a complete answer may establish, with supporting evidence, that no such model can be developed and validated at present; a bare claim of feasibility or infeasibility does not resolve the question. Because isotope-labeled internal standards cannot be prepared for unidentified compounds, matrix effects on ionization cannot be corrected in the usual way and enter the problem as a condition on validation with matrix-containing samples.

References

  1. Sepman H, Malm L, Peets P, MacLeod M, Martin JW, Breitholtz M, Kruve A (2023) Bypassing the Identification: MS2Quant for Concentration Estimations of Chemicals Detected with Nontarget LC-HRMS from MS² Data. Analytical Chemistry 95(33):12329–12338. DOI 10.1021/acs.analchem.3c01744, https://doi.org/10.1021/acs.analchem.3c01744
  2. Peets P, Wang W-C, MacLeod M, Breitholtz M, Martin JW, Kruve A (2022) MS2Tox Machine Learning Tool for Predicting the Ecotoxicity of Unidentified Chemicals in Water by Nontarget LC-HRMS. Environmental Science & Technology 56(22):15508–15517. DOI 10.1021/acs.est.2c02536, https://doi.org/10.1021/acs.est.2c02536
  3. Sepman H, Jonsson L, Posselt M, Johansson T, Breitholtz M, Martin JW, MacLeod M, Kruve A (2026) High-risk contaminants detected in wastewater effluent samples can be prioritized prior to structural assignment using machine learning tools. Water Research 306:126581. DOI 10.1016/j.watres.2026.126581, https://doi.org/10.1016/j.watres.2026.126581
  4. Malm L, Liigand J, Aalizadeh R, Alygizakis N, Ng K, Frøkjær EE, Nanusha MY, Hansen M, Plassmann M, Bieber S, Letzel T, Balest L, Abis PP, Mazzetti M, Kasprzyk-Hordern B, Ceolotto N, Kumari S, Hann S, Kochmann S, Steininger-Mairinger T, Soulier C, Mascolo G, Murgolo S, Garcia-Vara M, López de Alda M, Hollender J, Arturi K, Coppola G, Peruzzo M, Joerss H, van der Neut-Marchand C, Pieke EN, Gago-Ferrero P, Gil-Solsona R, Licul-Kucera V, Roscioli C, Valsecchi S, Luckute A, Christensen JH, Tisler S, Vughs D, Meekel N, Talavera Andújar B, Aurich D, Schymanski EL, Frigerio G, Macherius A, Kunkel U, Bader T, Rostkowski P, Gundersen H, Valdecanas B, Davis WC, Schulze B, Kaserzon S, Pijnappels M, Esperanza M, Fildier A, Vulliet E, Wiest L, Covaci A, Macan Schönleben A, Belova L, Celma A, Bijlsma L, Caupos E, Mebold E, Le Roux J, Troia E, de Rijke E, Helmus R, Leroy G, Haelewyck N, Chrastina D, Verwoert M, Thomaidis NS, Kruve A (2024) Quantification Approaches in Non-Target LC/ESI/HRMS Analysis: An Interlaboratory Comparison. Analytical Chemistry 96(41):16215–16226. DOI 10.1021/acs.analchem.4c02902, https://doi.org/10.1021/acs.analchem.4c02902
  5. Malm L, Alygizakis N, Aalizadeh R, Kruve A (2026) In-Depth Analysis of the Data from an Interlaboratory Study of Quantitative Non-Target Screening—How Do the Instrumental Methods Compare? Molecules 31(5):875. DOI 10.3390/molecules31050875, https://doi.org/10.3390/molecules31050875
  6. Wang W-C, Amini N, Huber C, Kull M, Kruve A (2025) Active Learning Improves Ionization Efficiency Predictions and Quantification in Nontargeted LC/HRMS. Analytical Chemistry 97(25):13131–13139. DOI 10.1021/acs.analchem.5c00816, https://doi.org/10.1021/acs.analchem.5c00816
  7. Nikolopoulos A, van Herwerden D, Turkina V, Kruve A, Baerenfaenger M, Samanipour S (2025) Ionization efficiency prediction of electrospray ionization mass spectrometry analytes based on molecular fingerprints and cumulative neutral losses. Journal of Cheminformatics 17(1):183. DOI 10.1186/s13321-025-01129-7, https://doi.org/10.1186/s13321-025-01129-7
  8. Pu S, McCord JP, Dickman R, Sayresmith N, Sepman H, Kruve A, Aga DM, Sobus JR (2025) Examining environmental matrix effects on quantitative non-targeted analysis estimates of per- and polyfluoroalkyl substances. Analytical and Bioanalytical Chemistry 417(10):2097–2110. DOI 10.1007/s00216-025-05796-1, https://doi.org/10.1007/s00216-025-05796-1
  9. Peets P, Rian MB, Martin JW, Kruve A (2024) Evaluation of Nontargeted Mass Spectral Data Acquisition Strategies for Water Analysis and Toxicity-Based Feature Prioritization by MS2Tox. Environmental Science & Technology 58(39):17406–17418. DOI 10.1021/acs.est.4c02833, https://doi.org/10.1021/acs.est.4c02833
  10. Kruve A, Kaupmees K, Liigand J, Leito I (2014) Negative Electrospray Ionization via Deprotonation: Predicting the Ionization Efficiency. Analytical Chemistry 86(10):4822–4830. DOI 10.1021/ac404066v, https://doi.org/10.1021/ac404066v
  11. Souihi A, Kruve A (2024) Estimating LoD-s Based on the Ionization Efficiency Values for the Reporting and Harmonization of Amenable Chemical Space in Nontargeted Screening LC/ESI/HRMS. Analytical Chemistry 96(28):11263–11272. DOI 10.1021/acs.analchem.4c01002, https://doi.org/10.1021/acs.analchem.4c01002
  12. Sepman H, Peets P, Jonsson L, Malm L, Posselt M, MacLeod M, Martin J, Breitholtz M, McLachlan M, Kruve A (2023) Machine Learning Tools Can Pinpoint High-Risk Water Pollutants. Proceedings 2023(20):68. DOI 10.3390/proceedings2023092068, https://doi.org/10.3390/proceedings2023092068