When liquid chromatography–high-resolution mass spectrometry (LC–HRMS) is used to screen complex mixtures for families of structurally analogous nontarget compounds, the workflow of Wu et al. (2013) combines matrix-blank subtraction, screening by family-specific diagnostic fragment ions and constant neutral losses, and family-specific quantitative structure–retention relationship (QSRR) models that predict retention times to pick among isomeric database candidates. The demonstration leaves the unresolved core: QSRR-guided assignment can be only as reliable as the training standards' chemical space is representative of the analytes (the study reported successful differentiation of most, but not all, isomers), and the whole strategy was tied to one LC–MS setup, two compound families, and empirically chosen diagnostic ions. Later work (2015–2026) largely solved cross-chromographic-condition retention prediction for large aggregated datasets, but no published framework calibrates when retention-based isomer discrimination can be trusted as a function of departure from the standards' chemical space, sets the confidence attainable without standards, or replaces per-family empirical diagnostic-ion selection with validated systematic criteria. The problem — making retention-validated analogous-family identification calibrated, transferable, and standards-tolerant — remains open.

Background

In nontarget screening by liquid chromatography coupled to high-resolution mass spectrometry (LC–HRMS), analysts detect thousands of features in a complex sample and must assign structures without knowing in advance which compounds are present. A recurring practical setting, exemplified by herbal-medicine injections studied by Wu et al. (2013, J. Chromatogr. A 1303:39–47), is that most detected components belong to a few families of structurally analogous compounds (there: organic acids and ginsenosides), so identification can proceed family by family. Their integral strategy had three steps: (i) subtract a matrix-matched blank to remove excipient and solvent background; (ii) screen for family members using diagnostic fragment ions (ions recurring across the family's tandem-MS spectra) and constant neutral losses (fragments lost as a neutral molecule, e.g. CO₂ for organic acids), linked into a family network; and (iii) build a quantitative structure–retention relationship (QSRR) model — a regression from computed molecular descriptors to chromatographic retention time, the approach reviewed by Kaliszan (2007, Chem. Rev. 107:3212–3246) — from a small set of family standards, then use predicted retention times to choose among multiple isomeric database candidates for each detected peak, since mass spectra alone cannot distinguish positional isomers.

This reliance on QSRR as the deciding vote among isomers carries a systematic risk. The study reported that the QSRR approach enabled successful differentiation of most, but not all, isomers, and any such failure concentrates where an analyte's chemistry departs from the chemical space spanned by the training standards. In cheminformatics terms this is an applicability-domain problem: a regression model trained on a finite set of compounds degrades in a largely uncharacterized way when applied to analytes outside that domain, and here the model's output is used not merely as an aid but as the deciding vote among isomers. The workflow was also demonstrated on only two families under one set of chromatographic conditions, depended on authentic standards for building the models and for confirming assignments, and required the diagnostic ions and neutral losses to be picked empirically for each new family without systematic criteria or counter-screens against unrelated matrix components.

Subsequent literature transformed parts of this landscape. Community tools such as PredRet (Stanstrup et al. 2015, Anal. Chem. 87:9421–9428) map retention times directly between chromatographic systems; machine-learning retention prediction was evaluated systematically by Bouwmeester et al. (2019, Anal. Chem. 91:3694–3703); and transfer learning and multitask models — Retip (Bonini et al. 2020, Anal. Chem. 92:7515–7522), DNNpwa-TL (Ju et al. 2021, Anal. Chem. 93:15651–15658), MDL-TL with vectorized chromatographic parameters (Li et al. 2025, Anal. Chem. 97:16788–16795), FastRet (Fadil et al. 2026, J. Chem. Inf. Model.), Uni-RT (Xiong et al. 2026, Anal. Chem. 98:9500–9507), and the uncertainty-aware MC-Retention (Beck et al. 2026, Anal. Chem. 98:4777–4790) — now predict retention across columns, buffers, and conditions, while Borase et al. (2026, J. Chromatogr. A 1786:467335) review how experimental variables drive QSRR transferability. In parallel, Schymanski et al. (2014, Environ. Sci. Technol. 48:2097–2098) fixed a community scale of identification confidence in which only comparison with authentic reference material yields the top, confirmed level — making explicit that standards-free isomer-level claims are inherently capped. What this large body of work does not supply is the calibrated coupling: how much an isomer assignment made by retention prediction should be trusted as the analyte departs from the training standards' chemical space, how much such a capability transfers to a new family with few or no standards, and by what systematic rule family-specific diagnostic ions and neutral losses should be selected with known false-positive and false-negative behavior.

Problem Statement

Establish when retention-validated identification of analogous nontarget compounds in complex LC–MS mixtures can be trusted and transferred. The source work (Wu et al. 2013) validates proposed structures by family-specific QSRR retention prediction and screens families by empirical diagnostic ions and constant neutral losses; the dependencies its demonstration exhibits — a small family-specific training set, empirically chosen diagnostic ions, one LC–MS setup and two compound families, and reliance on authentic standards — define the open question, whose components jointly support one objective — calibrated, transferable analogous-family identification. Specifically: (i) characterize quantitatively how the reliability of QSRR-based retention prediction — and of the isomer assignments it decides — degrades as the analyte departs from the chemical space spanned by the training standards (an applicability-domain or calibrated-uncertainty account); (ii) establish whether the resulting identification capability transfers across chromatographic systems, stationary phases, gradients, MS platforms, matrices, and structurally distinct families without full empirical re-derivation for each family; (iii) determine what identification confidence — in particular for positional isomers — is attainable when representative standards are unavailable, and how the strategy degrades or must be modified in that regime; and (iv) replace purely empirical per-family selection of diagnostic fragment ions and neutral losses with systematic selection criteria that address both false positives (non-family matrix components sharing the diagnostic ions or losses) and false negatives (family members with atypical fragmentation). The four components are coupled rather than redundant: component (iii) is the standards-free limiting case of the chemical-space account in (i), extended by the demand to state whether and how the strategy must be modified when no representative standards are available. Each component must be resolved in refutable form: (i) by naming a quantitative retention-error or uncertainty measure and the form of its dependence on departure from the training standards' chemical space; (ii) by an evidence-supported determination of transfer over the scope described by the listed factors, which delimit the scope of the claim rather than six separately demonstrated requirements; (iii) by fixing the attainable confidence level on the Schymanski et al. (2014) scale and stating the required strategy modification; and (iv) by stating the selection criteria together with the false-positive and false-negative behavior they are designed to control.

References

  1. Wu, Liang; Gong, Ping; Wu, Yuzheng; Liao, Ke; Shen, Hanyuan; Qi, Qu; Liu, Huiying; Wang, Guangji; Hao, Haiping (2013). An integral strategy toward the rapid identification of analogous nontarget compounds from complex mixtures. Journal of Chromatography A 1303:39–47. doi:10.1016/j.chroma.2013.06.041. https://doi.org/10.1016/j.chroma.2013.06.041
  2. Kaliszan, Roman (2007). QSRR: Quantitative Structure-(Chromatographic) Retention Relationships. Chemical Reviews 107(7):3212–3246. doi:10.1021/cr068412z. https://doi.org/10.1021/cr068412z
  3. Stanstrup, Jan; Neumann, Steffen; Vrhovšek, Urška (2015). PredRet: Prediction of Retention Time by Direct Mapping between Multiple Chromatographic Systems. Analytical Chemistry 87(18):9421–9428. doi:10.1021/acs.analchem.5b02287. https://doi.org/10.1021/acs.analchem.5b02287
  4. Wolfer, Arnaud M.; Lozano, Sylvain; Umbdenstock, Thierry; Croixmarie, Vincent; Arrault, Alban; Vayer, Philippe (2016). UPLC–MS retention time prediction: a machine learning approach to metabolite identification in untargeted profiling. Metabolomics 12(1):8. doi:10.1007/s11306-015-0888-2. https://doi.org/10.1007/s11306-015-0888-2
  5. Bouwmeester, Robbin; Martens, Lennart; Degroeve, Sven (2019). Comprehensive and Empirical Evaluation of Machine Learning Algorithms for Small Molecule LC Retention Time Prediction. Analytical Chemistry 91(5):3694–3703. doi:10.1021/acs.analchem.8b05820. https://doi.org/10.1021/acs.analchem.8b05820
  6. Bonini, Paolo; Kind, Tobias; Tsugawa, Hiroshi; Barupal, Dinesh Kumar; Fiehn, Oliver (2020). Retip: Retention Time Prediction for Compound Annotation in Untargeted Metabolomics. Analytical Chemistry 92(11):7515–7522. doi:10.1021/acs.analchem.9b05765. https://doi.org/10.1021/acs.analchem.9b05765
  7. Ju, Ran; Liu, Xinyu; Zheng, Fujian; Lu, Xin; Xu, Guowang; Lin, Xiaohui (2021). Deep Neural Network Pretrained by Weighted Autoencoders and Transfer Learning for Retention Time Prediction of Small Molecules. Analytical Chemistry 93(47):15651–15658. doi:10.1021/acs.analchem.1c03250. https://doi.org/10.1021/acs.analchem.1c03250
  8. Tynes, Michael; Gao, Wenhao; Burrill, Daniel J.; Batista, Enrique R.; Perez, Danny; Yang, Ping; Lubbers, Nicholas (2021). Pairwise Difference Regression: A Machine Learning Meta-algorithm for Improved Prediction and Uncertainty Quantification in Chemical Search. Journal of Chemical Information and Modeling 61(8):3846–3857. doi:10.1021/acs.jcim.1c00670. https://doi.org/10.1021/acs.jcim.1c00670
  9. Zhang, Qingqing; Huo, Mengqi; Zhang, Yanling; Qiao, Yanjiang; Gao, Xiaoyan (2018). A strategy to improve the identification reliability of the chemical constituents by high-resolution mass spectrometry-based isomer structure prediction combined with a quantitative structure retention relationship analysis: Phthalide compounds in Chuanxiong as a test case. Journal of Chromatography A 1552:17–28. doi:10.1016/j.chroma.2018.03.055. https://doi.org/10.1016/j.chroma.2018.03.055
  10. Zou, Ye-Ting; Long, Fang; Wu, Cheng-Ying; Zhou, Jing; Zhang, Wei; Xu, Jin-Di; Zhang, Ye-Qing; Li, Song-Lin (2019). A dereplication strategy for identifying triterpene acid analogues in Poria cocos by comparing predicted and acquired UPLC-ESI-QTOF-MS/MS data. Phytochemical Analysis 30(3):292–310. doi:10.1002/pca.2813. https://doi.org/10.1002/pca.2813
  11. Yan, Pan; Wang, Lu; Li, Sen; Liu, Xuechen; Sun, Yi; Tao, Jianmei; Ouyang, Hui; Zhang, Jianqing; Du, Zhifeng; Jiang, Hongliang (2021). Improved structural annotation of triterpene metabolites of traditional Chinese medicine in vivo based on quantitative structure–retention relationships combined with characteristic ions: Alismatis Rhizoma as an example. Journal of Chromatography B 1187:123012. doi:10.1016/j.jchromb.2021.123012. https://doi.org/10.1016/j.jchromb.2021.123012
  12. Ye, Hui; Zhu, Lin; Sun, Di; Luo, Xiaozhuo; Lu, Gaoyuan; Wang, Hong; Wang, Jing; Cao, Guoxiu; Xiao, Wei; Wang, Zhenzhong; Wang, Guangji; Hao, Haiping (2016). Nontargeted diagnostic ion network analysis (NINA): A software to streamline the analytical workflow for untargeted characterization of natural medicines. Journal of Pharmaceutical and Biomedical Analysis 131:40–47. doi:10.1016/j.jpba.2016.08.007. https://doi.org/10.1016/j.jpba.2016.08.007
  13. Dai, Chen; Wang, Chong; Zhang, Chunhua; Wang, Guoxiang; Wang, Jin; Chen, Jun; Guo, Bin; Yang, Tianshu; Cai, Bo (2016). A reference substance free diagnostic fragment ion-based approach for rapid identification of non-target components in Pudilan Xiaoyan oral liquid by high resolution mass spectrometry. Journal of Pharmaceutical and Biomedical Analysis 124:79–92. doi:10.1016/j.jpba.2016.02.020. https://doi.org/10.1016/j.jpba.2016.02.020
  14. Schymanski, Emma L.; Jeon, Junho; Gulde, Rebekka; Fenner, Kathrin; Ruff, Matthias; Singer, Heinz P.; Hollender, Juliane (2014). Identifying Small Molecules via High Resolution Mass Spectrometry: Communicating Confidence. Environmental Science & Technology 48(4):2097–2098. doi:10.1021/es5002105. https://doi.org/10.1021/es5002105
  15. Sun, Shiyuan; Cui, Biyue; Kong, Fanyu; Zhang, Zitong; Qiao, Youfu; Zhang, Shuting; Zhang, Xinran; Sun, Changhai (2024). Construction and application of a QSRR approach for identifying flavonoids. Journal of Pharmaceutical and Biomedical Analysis 240:115929. doi:10.1016/j.jpba.2023.115929. https://doi.org/10.1016/j.jpba.2023.115929
  16. Yuan, Hanwen; Su, Wei; Wang, Mengyun; Liu, Shi-qi; Yu, Huanghe; Li, Bin; Yang, Yupei; Wang, Wei (2025). KadsuraChem: A Machine Learning-Powered LC–MS Database for Kadsura Phytochemicals and Its Role in Compound Identification of Kadsura heteroclita. Journal of Agricultural and Food Chemistry 73(32):20488–20499. doi:10.1021/acs.jafc.5c05624. https://doi.org/10.1021/acs.jafc.5c05624
  17. Li, Yansong; Dong, Kunjie; Yu, Di; Huang, Dongdong; Liu, Xinyu; Xu, Guowang; Lin, Xiaohui (2025). Prediction of Retention Time by Combining Multiple Data Sets with Chromatographic Parameter Vectorization and Transfer Learning. Analytical Chemistry 97(31):16788–16795. doi:10.1021/acs.analchem.5c01703. https://doi.org/10.1021/acs.analchem.5c01703
  18. Beck, Armen G.; Jones, Gwenyth; Singh, Andrew; Fine, Jonathan; Shrestha, Rojan; Barrientos, Rodell C.; Sherer, Edward C.; Williams, Katharine B.; Regalado, Erik L.; Aggarwal, Pankaj (2026). Uncertainty-Aware Learning of Multiple Conditions as a Framework for Streamlined Retention Time Prediction to Accelerate Method Development. Analytical Chemistry 98(6):4777–4790. doi:10.1021/acs.analchem.5c06729. https://doi.org/10.1021/acs.analchem.5c06729
  19. Xiong, Ziyao; Wang, Xujie; Liu, Junqi; Cong, Shan; Yang, Qiong; Ji, Hongchao (2026). Unified Multitask Modeling for Retention Time Prediction Across Chromatographic Conditions. Analytical Chemistry 98(13):9500–9507. doi:10.1021/acs.analchem.5c07973. https://doi.org/10.1021/acs.analchem.5c07973
  20. Borase, Mayur; Patil, Kunal; Ghode, Piyush (2026). Quantitative structure–retention relationships (QSRR): Effect of experimental variables on QSRR model performance in HPLC - A critical review. Journal of Chromatography A 1786:467335. doi:10.1016/j.chroma.2026.467335. https://doi.org/10.1016/j.chroma.2026.467335
  21. Fadil, Fadi; Schmidt, Tobias; Amesoeder, Christian; Heckscher, Simon; Schoen, Marian; Gronwald, Wolfram; Oefner, Peter J.; Spang, Rainer; Dettmer, Katja (2026). FastRet: Fast and Simple Retention Time Prediction in Liquid Chromatography. Journal of Chemical Information and Modeling, doi:10.1021/acs.jcim.6c01344. https://doi.org/10.1021/acs.jcim.6c01344
  22. Bhandari, Dipendra; Paz, Henry A.; Henderson, Keith; Adepu, Kiran Kumar; Mani-Varnosfaderani, Ahmad; Assress, Hailemariam Abrha; Piccolo, Brian D.; Lan, Renny S.; Børsheim, Elisabet; Kay, Colin D.; Chintapalli, Sree V. (2026). From known chemical space to unannotated metabolites: a cluster-guided retention-time driven framework for biologically informed annotation. Metabolomics 22(5):146. doi:10.1007/s11306-026-02520-7. https://doi.org/10.1007/s11306-026-02520-7