Report 086 · Lab Science
A mass spectrometer does not recognise compounds. It measures mass-to-charge ratios, and sometimes it breaks a molecule and measures the pieces. Everything between that measurement and a chemical name on a report is inference. Analytical chemistry has a published scale for grading how much inference, and it is worth knowing which rung your result is standing on.
"We identified it by LC-MS" is one of the most load-bearing sentences in modern science, and it is doing at least five different jobs. It appears in environmental screening, in metabolomics, in food safety, in doping control, in forensic toxicology, and in the certificate of analysis attached to whatever supplement is on your counter. In every one of those settings it can mean anything from "we ran an authentic reference standard alongside the sample and they matched" to "the exact mass is consistent with a formula and a database returned a plausible structure."
Those are not close to the same claim. Analytical chemists know this, which is why they built a scale, and the useful thing about the scale is that it is short enough to remember and specific enough to ask about.
The scale most widely used for high-resolution mass spectrometry was proposed in 2014 by Emma Schymanski and colleagues at Eawag, the Swiss Federal Institute of Aquatic Science and Technology, in a two-page Environmental Science & Technology viewpoint. Schymanski presented the same framework, with its minimum data requirements, at the NORMAN network workshop on non-target screening later that year. Here are the levels with the evidence each one requires, as she set them out:
Level 1, confirmed structure, by reference standard. Minimum data: MS, MS², retention time, and a reference standard.
Level 2, probable structure, in two flavours. Level 2a is by library spectrum match, requiring MS, MS² and a library MS² spectrum. Level 2b is by diagnostic evidence, requiring MS, MS² and experimental data, and it covers the case where, in her framing, only one structure fits the experimental and spectral information available.
Level 3, tentative candidates: structure, substituent, or class. Minimum data: MS, MS² and experimental data. This is the grey zone where there is evidence for possible structures but not enough to pin one, positional isomers being the classic example.
Level 4, unequivocal molecular formula. Minimum data: MS isotope and adduct pattern. You know the atoms. You do not know how they are arranged.
Level 5, exact mass of interest. Minimum data: MS. A number, and a reason to care about it.
Notice what the ladder is actually measuring. It is not measuring how good the instrument is. Every level here can come off the same excellent instrument on the same excellent day. It is measuring how much external information was brought to bear, and in particular whether anyone ever put the real, purchased, known compound through the same system and compared.
Level 1 is separated from everything below it by a single physical requirement: somebody has to have the authentic reference standard in hand and run it. Not a literature spectrum, not a database entry, not a predicted fragmentation. The actual compound, on your instrument, under your conditions.
That is a purchasing and logistics problem more than an analytical one, and it is why so much real work lives at level 2. Standards are expensive, many compounds have no commercial standard at all, and a non-target screening run can flag hundreds of features at once. Nobody is buying hundreds of standards to chase a screening result.
The consequence shows up plainly in practice. A 2022 paper in Exposome proposing an annotation scoring framework for gas chromatography high-resolution mass spectrometry validated its approach against 80 spiked standards, and only those with in-house reference standards could reach level 1. The same authors note that "Level-1 assignment (using standards) is still necessary in most work with direct regulatory and policy implications." That is the honest split. Regulatory work buys the standards. Exploratory work generally cannot, and reports at level 2 or 3, and that is fine as long as the label survives the trip to the summary slide.
This is the part that surprises people who have just learned the scale, and it is the part I find most useful.
In her own worked examples from real screening data, Schymanski lists compounds assigned level 1 with question marks beside them. One entry is 4- and 5-methylbenzotriazole at m/z 134.0712, marked level 1, with the annotation that it involves co-eluting isomers, and the note that it "fulfils all identification criteria, but not a unique structure." Two isomers came off the column together. Every criterion for level 1 was met. The answer is still not a single molecule.
Another entry, pyrimidinol, is level 1 but with MS/MS from data-independent acquisition only, meaning no isolated precursor. Another, 4-dimethylaminopyridine, is level 1 with high-resolution MS only and no MS/MS available in that analysis, though it was a target with a reference standard. Her framing of the open question is direct: when is there sufficient evidence for a target, and she suggests that a level alone is not enough, that "The combination of 'level' and 'score' could help represent structure and evidence."
So the scale is a vocabulary, not a guarantee. It tells you what kind of evidence exists. It does not promise the evidence was sufficient, and the level's own author was saying so in public within months of proposing it.
There is a second scale in circulation, and mixing them up produces real confusion because the numbers run in a different direction and there are only four of them.
The Metabolomics Standards Initiative's Chemical Analysis Working Group published minimum reporting standards in 2007, in a paper led by Lloyd Sumner with a long author list, in Metabolomics. Its levels, as reproduced in Schymanski's workshop material: level 1, identified compounds, requiring "At least two independent and orthogonal data relative to an authentic compound under identical conditions." Level 2, putatively annotated compounds, "Similar to level 1, but based on literature values reported for authentic samples by other laboratories." Level 3, putatively characterised compound class, "Based upon characteristic physicochemical properties of a chemical class of compounds, or by spectral similarity to known compounds of a chemical class." Level 4, unknown compounds, which "can still be differentiated and quantified based upon spectral data."
Both scales put "we ran the standard" at level 1, which is the good news. After that they diverge, and an MSI level 2 is not a Schymanski level 2. If a method section says "level 2 identification" without naming the scale, you cannot resolve it from the number alone. Ask which framework.
One more caution, from the people who build these frameworks. The Exposome authors, describing why they were proposing something new for GC-HRMS rather than transplanting the LC framework, note that "the parameters proposed for assigning confidence levels are potentially subject to high false-positive and false-negative rates."
Assigning a level is a judgement call executed by a human or a script, with thresholds someone chose. A library match above some cosine similarity gets called level 2. Change the threshold and the same spectrum changes level. The scale disciplines the reporting language, which is a genuine and large improvement over the alternative, but it does not convert a soft inference into a hard one.
I could not open the 2014 Environmental Science & Technology viewpoint itself; it sits behind the ACS paywall, and the ACS site returned 403 to me. I verified its full citation independently through PubMed: Schymanski, Jeon, Gulde, Fenner, Ruff, Singer and Hollender, Environ. Sci. Technol. 48(4), 2097-2098, 2014, DOI 10.1021/es5002105, PMID 24476540, published 18 February 2014 with an epub of 29 January 2014. The level definitions and minimum data requirements I have quoted come from Schymanski's own NORMAN workshop presentation, which reproduces them and cites that paper. That is a first-author source for the framework, but it is a conference presentation and not the peer-reviewed article, and I am flagging the distinction rather than papering over it.
The same applies to the MSI levels. I verified the Sumner et al. citation through PubMed, but the four level definitions I quoted are as reproduced in Schymanski's presentation, not read from the 2007 paper itself.
I did not survey how the scale is used across the literature, so when I say most reported identifications are not level 1, that is a claim supported by the Exposome validation result and by the structural economics of reference standards, and it is not a number I measured. I have also not assessed the more recent scoring implementations built on this framework; a 2026 Analytical Chemistry paper implements Schymanski-based annotation confidence scoring in a commercial workflow, and I could not open it, so it is absent from this report.
My own peer-reviewed research is in microwave spectroscopy rather than mass spectrometry. The confidence framework here is not from my field, and I am reporting it, not speaking from inside it.
Three things to carry.
First, "identified" is a range, not a verdict. When a paper, a lab report, or a certificate of analysis names a compound, the question that separates a strong claim from a weak one is whether an authentic reference standard was run on that instrument, in that method, alongside that sample. If the answer is no, you are somewhere on levels 2 through 5 and the honest write-up will say so.
Second, ask which scale before you interpret a level number. Schymanski's five levels and the MSI's four levels agree at the top and disagree below it, and a bare "level 2" is ambiguous until the framework is named.
Third, even level 1 is a statement about evidence rather than a statement about truth. Co-eluting isomers can satisfy every criterion and still leave you with two candidate molecules. The scale exists to make people say what they did, which is a real achievement in scientific communication and a modest one in metaphysics. What it buys you is a specific question to ask. That is usually worth more than a confident answer.
U.S. Army combat veteran (Counter-IED / Electronic Warfare), peer-reviewed researcher in microwave spectroscopy, and founder & CEO of Shroombiosis. Consults on laboratory operations, AI, and supplement formulation.