Mass-spectrum deconvolution uses a model to interpret overlapping mass-to-charge signals in terms of underlying masses and charge states. It can turn a complicated electrospray spectrum into a much simpler mass plot. The simpler output is an interpretation of the data, so its assumptions and supporting signals remain part of the result.
Understand which dimensions are being separated
Marty and colleagues' original UniDec paper describes a Bayesian approach that separates the mass and charge dimensions of electrospray and ion-mobility spectra. Their framework uses information about charge-state distributions to interpret signals that can overlap in m/z. It is a model-based reconstruction, not a second direct weighing of neutral molecules.Marty and colleagues — Bayesian deconvolution of mass and ion-mobility spectra (opens in a new tab)
The phrase zero-charge spectrum refers to a reconstructed mass-axis representation. It does not mean that the mass spectrometer directly detected uncharged peptides. Charge was essential to the ion measurement that supplied the input data.
| Record | What it contains |
|---|---|
| Measured spectrum | Response across observed m/z |
| Processing model | Assumptions connecting mass, charge and signal shape |
| Reconstructed mass distribution | Mass features inferred under that model |
| Reconstructed m/z fit | The model's predicted explanation of the input signal |
A simple case shows the bookkeeping
Take an invented pair of protonated signals at m/z 1,201.0073 and 801.0073. Assigning charges 2+ and 3+, respectively, gives a common neutral mass of 2,400.0000 Da with the rounded proton mass 1.0073 Da. The two input features can therefore support one mass feature.
If the first signal were incorrectly assigned 3+ instead, it would suggest 3,600 Da. An isolated peak can support multiple candidate mass-and-charge explanations. Additional charge states, isotope information and other constraints help distinguish them.
Real deconvolution handles more than this two-number example, especially when multiple molecules, adducts or unresolved distributions contribute. The arithmetic explains the inference problem; it does not reproduce a particular software algorithm or guarantee a unique answer.
Record assumptions that can affect the output
A processing record should identify the input interval, mass and charge ranges, assumed charge carrier, signal treatment and relevant model settings. A search limited to a chosen mass range cannot establish that no signal outside that range matters. Similarly, a proton-only interpretation needs scrutiny when substantial adducts are present.
In an original comparison, processing only the centre of a chromatographic peak and processing a broader interval that includes a neighbouring feature provide different inputs. A change in the reconstructed distribution might therefore originate in data selection rather than in the molecules changing between analyses.
Check how well the interpretation explains the data
Marty's later primary study specifically addressed deconvolution artefacts and demonstrated a SoftMax-based approach to reducing them in tested datasets. This establishes that algorithmic artefacts are a real interpretation issue; it does not imply that every small reconstructed peak is false or that one processing option removes every ambiguity.Marty — Reducing electrospray-deconvolution artefacts (opens in a new tab)
Review whether a proposed mass is supported by coherent input features and whether the reconstructed m/z fit leaves important unexplained signal. A favourable fit is useful, but it should be assessed alongside the model's assumptions and possible competing explanations.
Finally, a mass feature is still a mass feature. Deconvolution does not by itself determine sequence order, prove a molecular identity against all isomers or quantify the original vial. It simplifies the spectral interpretation so those further questions can be addressed with appropriately matched evidence.
Sources and further detail
- Marty and colleagues — Bayesian deconvolution of mass and ion-mobility spectra (opens in a new tab)
Analytical Chemistry 87, 4370–4376 (2015), DOI 10.1021/acs.analchem.5b00140. Original UniDec framework; no software performance guarantee inferred.
- Marty — Reducing electrospray-deconvolution artefacts (opens in a new tab)
J Am Soc Mass Spectrom 30, 2174–2177 (2019), DOI 10.1007/s13361-019-02286-4. Primary SoftMax study with specified datasets.
- NIST — 2022 CODATA constants (opens in a new tab)
Proton mass in unified atomic mass units. Examples round it to 1.007276 Da or 1.0073 Da as stated; exact ion assignments require the appropriate mass convention.
Sources checked 19 September 2026. Worked examples are illustrative unless a supplied report is explicitly identified. This article has not undergone independent scientific peer review.