Matrix-assisted laser desorption/ionization mass spectrometry (MALDI-MS) requires a matched matrix–analyte pair for sensitive detection, yet which metabolites ionize with a given matrix, and how efficiently, has remained largely empirical. Yukihira et al. (2014) built the first quantitative structure–property relationship (QSPR) models of metabolite ionizability (91% classification accuracy) and ionization efficiency (rank correlation ρ = 0.78) for 200 metabolites in negative-ion mode with 9-aminoacridine matrix, and posed the follow-on questions: does the approach generalize to other matrices (the descriptors identified were interpreted as specific to 9-AA and/or negative mode, leaving transferability to positive ion mode and quantitative accuracy beyond rank correlation as natural further extensions), what physical mechanism makes structurally similar compounds (e.g., alanine vs β-alanine, leucine vs isoleucine) ionize so differently; and how do preparation conditions (pH, solvent, crystal phase) modulate the structure–ionization relationship? The objective is to establish a general, validated structure-based predictor of MALDI ionization behavior, together with a mechanistic account of how molecular structure and preparation conditions govern it.

Background

MALDI-MS is a soft ionization technique in which a laser-absorbing crystalline 'matrix' compound is co-crystallized with the analyte; the matrix absorbs laser energy, desorbs the mixture into the gas phase, and participates in proton-transfer or electron-transfer reactions that convert analytes into gas-phase ions. In negative-ion mode, analytes are observed mainly as deprotonated molecules [M–H]⁻, and 9-aminoacridine (9-AA), a basic dye matrix, is widely used for small metabolites. 'Ionizability' denotes whether a compound yields a detectable peak at all, while 'ionization efficiency' denotes how strongly it ionizes, commonly operationalized by the limit of detection (LOD), the lowest concentration at which the peak is reliably detected. Choosing the matrix and anticipating detection limits is still largely a matter of trial and error: a 2026 Chemical Reviews survey of 467 known matrices concludes that most were discovered by chance and that matrix selection remains 'often empirically determined' (Wu et al. 2026).

The QSPR (quantitative structure–property relationship) approach attempts to predict a measured property from computed molecular descriptors — numerical encodings of structure such as electrotopological state (E-state) indices (which combine electronic and topological environment of each atom), WHIM descriptors (weighted holistic invariant molecular descriptors summarizing molecular size, shape, symmetry, and atom distribution in 3D), or substructure fingerprints — using statistical or machine-learning models such as Random Forests (an ensemble of decision trees whose averaged vote gives a prediction). Yukihira et al. (2014) applied this to 200 standard metabolites measured with 9-AA in negative mode: a Random Forest classification model predicted ionizability with up to 91% accuracy (3D descriptors), and regression models predicted ionization efficiency (LOD) with rank correlation ρ up to 0.78 (2D descriptors). Descriptor-importance analysis indicated that proton affinity (via hydrogen-bonding strength and negatively charged surface area) dominates ionizability, while an amino-acid-specific model selected WHIM shape descriptors, implying that molecular shape or flexibility itself modulates ionization — consistent with the strikingly different ionization of isomeric or near-isomeric pairs such as alanine/β-alanine and leucine/isoleucine. Independent work has since extended structure–ionization modeling beyond this 9-AA/negative-mode baseline: a multi-matrix ionization-efficiency relationship for 234 isotopically labeled metabolites in negative-ion mode across three matrices (Nikolopoulou et al. 2025), and a machine-learning predictor of ionization behavior for ~1200 drug-like compounds in both positive and negative ion modes (Koerfer et al. 2025; Soltwisch et al. 2024); no validated predictor yet spans chemically diverse metabolites across multiple matrix chemistries, both ion modes, and absolute detection limits. Analogous structure–ionization modeling is mature for electrospray ionization (ESI), where an ionization-efficiency scale and prediction models for organic compounds exist (Oss et al. 2010), but MALDI ionization couples additional processes — co-crystallization, desorption, and matrix-mediated chemistry — that have resisted a single predictive framework. Mechanistic models of MALDI ion formation (e.g., the cluster-ionization and coupled chemical/physical dynamics models reviewed by Knochenmuss 2006, 2016) describe classes of pathways without yielding compound-specific efficiency predictions, and experimental work has long shown that preparation conditions such as pH and solvent composition alter which species are observed (Sun et al. 2008).

Problem Statement

Establish, validate, and mechanistically ground the prediction of MALDI-MS ionization behavior of small molecules from their molecular structure. Concretely, relative to the 2014 baseline (QSPR models for 200 metabolites with 9-AA in negative mode: 91% ionizability classification accuracy, ρ = 0.78 efficiency regression): (1) construct and externally validate quantitative models of ionizability and ionization efficiency for chemically diverse metabolites that generalize across MALDI matrix chemistries beyond the 9-aminoacridine negative-ion-mode baseline, using available structure–ionization data across matrix chemistries and holding out compounds and matrices for external validation; (2) explain the physical mechanism by which molecular structure modulates ionization — i.e., how molecular structure, including its shape and flexibility, governs matrix–analyte co-crystallization, proton (or electron) transfer, and hence the divergent ionization of near-isomeric compounds such as alanine/β-alanine and leucine/isoleucine; and (3) characterize how experimental preparation conditions (pH, solvent composition, and the formation of specific co-crystal phases) modulate the structure–ionization relationships, so that the predictive models hold under realistic sample-preparation variation. All three strands are jointly required for completion. Strand (1) requires models whose predictive accuracy is established on compounds and matrices not used in fitting; strand (2) requires a mechanistic account that makes a falsifiable claim about the structural origin of the near-isomeric ionization contrast; strand (3) requires evidence that the named preparation conditions shift the descriptor–ionization relationship. Together they serve one objective: replacing empirical matrix selection and detection-limit guessing in MALDI-MS of small molecules with structure-based, condition-aware prediction.

References

  1. Yukihira D., Miura D., Fujimura Y., Umemura Y., Yamaguchi S., Funatsu S., Yamazaki M., Ohta T., Inoue H., Shindo M., Wariishi H. (2014). MALDI Efficiency of Metabolites Quantitatively Associated with their Structural Properties: A Quantitative Structure–Property Relationship (QSPR) Approach. Journal of the American Society for Mass Spectrometry 25(1):1–5. DOI 10.1007/s13361-013-0772-0. https://doi.org/10.1007/s13361-013-0772-0
  2. Nikolopoulou V., Stem A., Vasiliou V., Aalizadeh R. (2025). Investigating the Quantitative Structure–Ionization Efficiency Relationship of Small Molecules and Lipids in the Presence of Ammonium Fluoride in MALDI-TIMS-QTOF Mass Spectrometry Imaging. Journal of the American Society for Mass Spectrometry 36(11):2460–2469. DOI 10.1021/jasms.5c00211. https://doi.org/10.1021/jasms.5c00211
  3. Koerfer K., Oldopp J., Dreisewerd K., Palmer A., Soltwisch J., Marshall P. (2025). Predicting the Ionization Behavior of Drugs in Tissue in MALDI and MALDI-2 Mass Spectrometry Imaging Using Machine Learning. Analytical Chemistry 97(43):23879–23886. DOI 10.1021/acs.analchem.5c03250. https://doi.org/10.1021/acs.analchem.5c03250
  4. Soltwisch J., Palmer A., Hong H., Majer J., Dreisewerd K., Marshall P. (2024). Large-Scale Screening of Pharmaceutical Compounds to Explore the Application Space of On-Tissue MALDI and MALDI-2 Mass Spectrometry. Analytical Chemistry 96(25):10294–10301. DOI 10.1021/acs.analchem.4c01088. https://doi.org/10.1021/acs.analchem.4c01088
  5. Padilla C. A., Díaz-Sánchez L. M., Blanco-Tirado C., Combariza A. F., Combariza M. Y. (2024). AI-Guided Design of MALDI Matrices: Exploring the Electron Transfer Chemical Space for Mass Spectrometric Analysis of Low-Molecular-Weight Compounds. Journal of the American Society for Mass Spectrometry 35(12):2836–2848. DOI 10.1021/jasms.4c00186. https://doi.org/10.1021/jasms.4c00186
  6. Wu R., Liu R., Hu H., Qin L., Chen L., Bao Z., Fu J., Guo H., Wang L., Wang A., Wang Z., Yang C., Cheng X., Chen D., Liu H., Jing Y., Guo S., Fu Y., Wang X. (2026). MALDI Matrix: Origins, Innovations, and Frontiers. Chemical Reviews 126(5):3324–3528. DOI 10.1021/acs.chemrev.5c00786. https://doi.org/10.1021/acs.chemrev.5c00786
  7. Knochenmuss R. (2006). Ion formation mechanisms in UV-MALDI. The Analyst 131(9):966–986. DOI 10.1039/b605646f. https://doi.org/10.1039/b605646f
  8. Knochenmuss R. (2016). The Coupled Chemical and Physical Dynamics Model of MALDI. Annual Review of Analytical Chemistry 9:365–385. DOI 10.1146/annurev-anchem-071015-041750. https://doi.org/10.1146/annurev-anchem-071015-041750
  9. Oss M., Kruve A., Herodes K., Leito I. (2010). Electrospray Ionization Efficiency Scale of Organic Compounds. Analytical Chemistry 82(7):2865–2872. DOI 10.1021/ac902856t. https://doi.org/10.1021/ac902856t
  10. Sun G., Yang K., Zhao Z., Guan S., Han X., Gross R. W. (2008). Matrix-assisted laser desorption/ionization time-of-flight mass spectrometric analysis of cellular glycerophospholipids enabled by multiplexed solvent dependent analyte–matrix interactions. Analytical Chemistry 80(19):7576–7585. DOI 10.1021/ac801200w. https://doi.org/10.1021/ac801200w