The metabolite fraction library (mFL) method removes the main obstacle to quantitative 1^1H NMR metabolomics — severe overlap of metabolite signals — by chromatographically fractionating a pooled sample into a library of simplified spectra from which a sample-derived metabolite basis set (mBS) is extracted and fit to unfractionated profiling spectra. Published in 2026 with strong accuracy on ground-truth mixtures, the method still requires an operator at three points: interactive correction of automatically determined mBS elements, tolerance of matrix-dependent chemical-shift changes for pH-sensitive metabolites, and construction of a new fraction library and basis set for every distinct biological matrix. The open problem is to make this analysis stage fully automated (no interactive adjustment, at equal accuracy) and transferable (mBS elements reusable across matrices and studies), turning a per-study interactive technique into an unattended, matrix-general profiling platform.

Background

Metabolomics aims to quantify the small-molecule content of biological mixtures such as biofluids and tissue extracts. One-dimensional proton nuclear magnetic resonance (1D 1^1H NMR) spectroscopy is a standard, reproducible, and inherently quantitative tool for this purpose, but in an unfractionated extract the resonances of dozens to hundreds of metabolites overlap so heavily that individual contributions cannot be separated by simple integration. Existing analysis strategies trade off two limitations. Library-fitting methods model the mixture spectrum as a sum of reference spectra of known metabolites: BATMAN (Bayesian deconvolution and quantification of metabolites in complex 1D NMR spectra; Hao et al., 2014) fits a user-chosen set of reference spectra and allows chemical-shift adjustment, Bayesil (Ravanbakhsh et al., 2015) performs fully automated profiling but only for specific biofluids under tightly standardized conditions, and commercial suites such as Chenomx depend on subjective interactive steps — the source paper of this problem explicitly notes that interactive, operator-dependent workflows make results statistically fragile and prone to bias. These approaches also cannot account for metabolites absent from their libraries, and predicted-spectra libraries remain too inaccurate in chemical shift to substitute for measured ones. Library-free deconvolution methods decompose spectra into generic peak lists without metabolite identity; the recent metabodeconplus R package (Schmidt et al., 2026) shows this line is advancing quickly but stops at aligned signal integrals rather than metabolite-resolved quantification. The metabolite fraction library (mFL) approach (Esselman et al., 2026) attacks the overlap problem experimentally rather than computationally: a pooled study sample is separated by HPLC (high-performance liquid chromatography) into ~140 fractions, each measured by 1D 1^1H NMR, so that co-eluting metabolites are largely resolved in the chromatographic dimension. Each fraction spectrum is decomposed by SAND (spectral automated NMR decomposition; Wu et al., 2024), a time-domain modeling method that represents a spectrum as a sum of decaying sinusoids and outputs a quantitative table of every signal's frequency, amplitude, decay, and phase. The FLAnalyzer software then traces signals across fractions and groups peaks whose intensities co-vary across fractions — the same correlation idea as STOCSY (statistical total correlation spectroscopy), here applied along the chromatographic axis — into metabolite basis set (mBS) elements, each representing one metabolite's complete resonance pattern including unknowns; elements are annotated by matching reference libraries such as GISSMO (a database of experimentally measured small-molecule NMR spectra) and BMRB (the Biological Magnetic Resonance Data Bank). Finally the whole mBS is fit to conventional unfractionated profiling spectra with BATMAN. On ground-truth mixtures of 53 metabolites, the interactive workflow accurately quantified 50 metabolites and described 91–96% of total spectral intensity; applied to the fungus Neurospora crassa it identified 45 metabolites with high and 45 with medium confidence, describing 94% of spectral intensity — without any external spectral database. Three bottlenecks keep this out of routine unattended use. First, residual co-elution and overlap still confuse the automatic correlation grouping, so FLAnalyzer provides interactive tools to add or remove signals from mBS elements; the authors state this step adds significant operator time and the subjectivity of an individual operator's interpretation, and that they are working to refine fully automated analysis. Second, pH-sensitive metabolites such as histidine and ascorbate shift substantially between the fractionated matrix and each mixture matrix; BATMAN's per-sample shift adjustment is reported to become impractical and bias-introducing in large studies, and published peak-alignment techniques — correlation optimized warping (COW; Nielsen et al., 1998) and icoshift (Savorani et al., 2010) — have not been adapted to this workflow. Third, the mBS is sample-specific: fitting the N. crassa mBS to the ground-truth mixtures (and vice versa) described only 73–88% of spectral intensity versus the near-complete native fits, so a new fraction library and basis set is currently needed for every distinct biological matrix, while the authors only anticipate that mBS elements could be reused across studies on the same sample type.

Problem Statement

In the metabolite fraction library (mFL) workflow of Esselman et al. (2026), make the analysis stage fully automated and transferable. Specifically: (i) construct the metabolite basis set (mBS) from fraction-library spectra algorithmically, with no interactive operator adjustment of automatically determined elements, while matching the published interactive-workflow accuracy on the 53-metabolite ground-truth mixtures (50 of 53 metabolites accurately quantified; 91–96% of total spectral intensity described); (ii) quantify pH-sensitive metabolites such as histidine and ascorbate robustly under the chemical-shift displacement that occurs between the fractionated library matrix and each mixture matrix, without per-sample manual shift adjustment; and (iii) demonstrate that mBS elements constructed for one biological matrix or study can be reused to quantify spectra of other matrices or studies, approaching native-matrix fit performance (describing at least ~90% of spectral intensity, versus the 73–88% currently described in cross-matrix fits), so that a new fraction library need not be constructed for every distinct matrix. These are linked components of one objective — removing the operator and the per-matrix construction cost from mFL-based quantitative NMR profiling — and a solution must address all three on public, ground-truthed data.

References

  1. Esselman, C.; Garrison, K.; Ponce, L.; Borges, R. M.; Delaglio, F.; Edison, A. S. 'Metabolite Fraction Libraries for Quantitative NMR Metabolomics.' Analytical Chemistry 98(27):20255–20264 (2026). DOI 10.1021/acs.analchem.6c01279, https://doi.org/10.1021/acs.analchem.6c01279 (version of record, open access; full text also at PubMed Central PMC13373914)
  2. Esselman, C.; Garrison, K.; Ponce, L.; Borges, R. M.; Delaglio, F.; Edison, A. S. 'Metabolite Fraction Libraries for Quantitative NMR Metabolomics.' Preprint posted 2025-12-31. DOI 10.64898/2025.12.30.696914, https://doi.org/10.64898/2025.12.30.696914
  3. Wu, Y.; Sanati, O.; Uchimiya, M.; Krishnamurthy, K.; Wedell, J.; Hoch, J. C.; Edison, A. S.; Delaglio, F. 'SAND: Automated Time-Domain Modeling of NMR Spectra Applied to Metabolite Quantification.' Analytical Chemistry 96(5):1843–1851 (2024). DOI 10.1021/acs.analchem.3c03078, https://doi.org/10.1021/acs.analchem.3c03078
  4. Hao, J.; Liebeke, M.; Astle, W.; De Iorio, M.; Bundy, J. G.; Ebbels, T. 'Bayesian deconvolution and quantification of metabolites in complex 1D NMR spectra using BATMAN.' Nature Protocols 9(6):1416–1427 (2014). DOI 10.1038/nprot.2014.090, https://doi.org/10.1038/nprot.2014.090
  5. Ravanbakhsh, S.; Liu, P.; Bjordahl, T. C.; Mandal, R.; Grant, J. R.; Wilson, M.; Eisner, R.; Sinelnikov, I.; Hu, X.; Luchinat, C.; Greiner, R.; Wishart, D. S. 'Accurate, Fully-Automated NMR Spectral Profiling for Metabolomics.' PLOS ONE 10(5):e0124219 (2015). DOI 10.1371/journal.pone.0124219, https://doi.org/10.1371/journal.pone.0124219
  6. Schmidt, T.; Sombke, M.; Zacharias, H. U.; Oefner, P. J.; Spang, R.; Gronwald, W. 'Metabodeconplus—An R Package for Automated Deconvolution and Alignment of 1D NMR Metabolomics Data.' Metabolites 16(9):604 (2026). DOI 10.3390/metabo16090604, https://doi.org/10.3390/metabo16090604
  7. Lee, B. L.; Nunes, A. T.; Kovur, P.; Gholami, M.; Firouzi, A.; Mandal, R.; Wishart, D. S. 'Automated, targeted, NMR spectral profiling of human breast milk.' Metabolomics 22(3):90 (2026). DOI 10.1007/s11306-026-02472-y, https://doi.org/10.1007/s11306-026-02472-y
  8. Wuyart, A.; Martinez, A.; Cantrelle, F.-X.; Lameiras, P.; Mahmoudi, R.; Millot, J.-M. 'An automated 1D 1H and 2D 1H-13C HSQC NMR platform for high-field (900 MHz) metabolite profiling of cerebrospinal fluid.' Talanta 312:130352 (2027). DOI 10.1016/j.talanta.2026.130352, https://doi.org/10.1016/j.talanta.2026.130352 (print issue dated 2027-02; published online 2026-07-24)
  9. Nielsen, N.-P. V.; Carstensen, J. M.; Smedsgaard, J. 'Aligning of single and multiple wavelength chromatographic profiles for chemometric data analysis using correlation optimised warping.' Journal of Chromatography A 805(1-2):17-35 (1998). DOI 10.1016/S0021-9673(98)00021-1, https://doi.org/10.1016/S0021-9673(98)00021-1
  10. Savorani, F.; Tomasi, G.; Engelsen, S. B. 'icoshift: A versatile tool for the rapid alignment of 1D NMR spectra.' Journal of Magnetic Resonance 202(1):190–202 (2010). DOI 10.1016/j.jmr.2009.11.012, https://doi.org/10.1016/j.jmr.2009.11.012
  11. Patel, K.; Esselman, C.; Croy, J.; Gillis, M.; Da Pos Rodrigues, P.; Simmons, A.; Borges, R. M.; Edison, A. S.; Snyder, W. E. 'Higher Lipid Saturation in Well-Irrigated Georgia Cotton Plants: A Field-Based NMR Metabolomics Study.' Preprint (v2), posted June 2026. DOI 10.64898/2026.06.08.730374, https://doi.org/10.64898/2026.06.08.730374