# Predict NMR spectra from the Mass Spectrometry

Imagine, you have a tissue and the Mass Spectrometry for it. How would you get the corresponding NMR spectra knowing only the Mass Spec? Is there is a special formula or a well-known technique?

I have an idea to build an ML model based on http://hmdb.ca metabolites data where you have those pairs (MS and NMR). But I worry it is completelly wrong.

1. Do you think is it a stupid idea?
2. If there is an easy way of converting MS -> NMR which everybody knows?

I'm sorry for a probably silly question, I came from mathematics.

• Short of simply reverse-searching the mass spectrum in a database and finding out which molecule it corresponds to, I doubt it is possible to predict NMR data from MS data in any meaningful way. It's almost impossible to extract the structure of any random chemical from its MS. – orthocresol Nov 23 '18 at 23:36
• NMR data typically has far more information about the structure of a molecule than MS data. The main datum from (hi res) MS is the mass of the molecular ion, which characterises the molecule globally by providing a molecular formula. Meanwhile, NMR provides some global information (e.g. H and C counts) but also directly provides a lot of local information. The only way to make up the difference would be to perform a lot of ion fragmentation with MS-MS, and then try to build the molecule back together from the pieces. There will still be things you're unlikely to figure out (e.g. chirality). – Nicolau Saker Neto Nov 23 '18 at 23:43
• Sorry but I do think it "is a stupid idea" because those spectroscopies are based on entirely different aspects – SteffX Nov 24 '18 at 14:51
• @NicolauSakerNeto thank you for your comment! You said that NMR has far more information about the structure of a molecule than MS. Does it mean that the reverse transition NMR -> MS would be possible? At least some approximation? – Galina Alperovich Nov 24 '18 at 15:28
• @SteffX I understand that the principles are entirely different. But it is not obvious why necessary they don't have anything in common. Imagine, you have a camera and the painter - underlying principles of making a picture are very different, but it doesn't mean you will see the completely different output. – Galina Alperovich Nov 24 '18 at 15:32

Sorry, but in my opinion this endeavour is unlikely to be successful. The techniques are quite different by their underlying principles. This is, at least in part, why chemists working in (especially, but not limited to) organic sythesis use these and additional ones to get a sufficiently "rounded picture" about the substances analyzed, putting the information each technique provides together.

• mass spectroscopy typically relies on fragmenting the molecules in question, and then separates these then charged fragments by their mass/charge ratio. There are techniques keeping the molecules almost intact (softer ionization techniques, used for example in MS-mapping of tissues), and others (electon impact ionization) where the molecules are intentionally shattered into many differently sized fragments. Depending on the ionization technique applied, the eventually intensity of these individual signals will vary. And if the molecules of interest are embedded in a matrix (e.g., a serum, a tissue), identification of this molecules by MS becomes even more complex.
• NMR spectroscopy typically characterizes probes soluted, keeping the molecules of interest (smaller ones, or large proteins) basically intact, without fragmentation. You basically probe your sample for specific atoms of a kind (isotopes) and their nearest chemical environment. Yet the NMR spectrum of a given sample -- say of an amino acid (building block for proteins) -- already looks differently for scrutinizing the either peripheral hydrogens, the backbone of carbon atoms, or nitrogens.

I think a few pictures may provide a glimpse about the complexity here. For a fairly simple molecule like acetophenone, given the molecular structure is known, a sketcher like Chemdoodle may compute within a blink of an eye an estimate how the NMR spectrum either about the hydrogens, or about the carbon atoms will look like. This is possible because the underlying physical principles are known. In addition, there are very many similar compounds analyzed successfully by NMR, allowing (fine tuning) of details in the spectra simulated, e.g. where to expect the signals to appear. Regarding MS, a conservative approach is to predict only the peak(s) that by theory should appear without change of the molecule in terms of its mass. In other words, without fragmentation. Hence Chemdoodle's estimation is this:

There are equally projects to simulate MS spectra, like CFM-ID. Under the menu Utilities -> Spectra Simulation, you have the possibilty to try out different modi of ionization (shortcut). Acetophenone's structure, expressed in a machine-readable Smiles code is

C1=CC=C(C=C1)C(=O)C


Then, just choosing ESI mode about positive ions, without considering adducts, the three MS spectra simulated by CFM-ID of actephenone, in increasing hardness of ionization are these:

As you see, not only the presence / absence of the signals changes in function of the probability to record certain fragments of a specific mass/charge ratio (m/z), their relative intensity equally is affected by changing the parameters of ionization.

(Hovering the mouse over the streaks allows you to display the suggested fragments responsible for the signals in these simulations. Because of the special conditions in a mass spectrometer, unstable specis you wouldn't be able to store in a shelf may be seen, too.)

• Thank you for your answer! Now I understand, that the underlying principles are different. One question: you as an expert, would you ever "guess" how would NMR spectra look like given an MS? Some kind of an approximation or this is not possible too? – Galina Alperovich Nov 24 '18 at 15:25
• @GalinaAlperovich Indeed, there are MS fragments of certain mass/charge ratio typically observed again and again for certain key structures. A signal at m/z = 91 is typical for benzyl derivates, a m/z = 149 of derivatives of phthalic anhydride (often found in polymers), for example. But it were only a hint for the potential presence of such a sub-structure present. Other techniques (e.g., NMR, or IR, etc) then would be used to check if this is supported by their results, or not. Especially, since the prediction of NMR or IR spectra about a given (known) structure is easier, than for MS. – Buttonwood Nov 24 '18 at 21:51
• @GalinaAlperovich An addendum in the answer illustrates the problem in further detail. – Buttonwood Nov 24 '18 at 23:09
• thank you for your time and explanation! I will accurately go through it to understand these new things better. – Galina Alperovich Nov 26 '18 at 12:05