6

There are quite a few approaches (ranging from empirical additive rules based on presence of functional groups to full blown in silico QM calculations) to predicting molecular properties such as UVmax absorption or IR vibrational frequencies. While not perfect, they give a general ballpark range of what to look for.

I’ve recently been doing some quantitative MS work which has involved a lot of dilutions and calibrations.

It got me curious if there are approaches to predict peak intensity as it’s not really something I’ve ever read about (or even included in software similar to basic IR/NMR prediction). It obviously wouldn’t be useful for quantification, but could give a vague sense of how strongly something will be observed (if at all).

To give an example, imagine mixing 1 mmol of paracetamol and 1 mmol of aspirin in DMSO and recording an LCMS of this—although you have the same amount of both the intensity in the MS TIC will vary significantly.

It seems to me that the parameters are quite well controlled (voltage, temperature, $\mathrm{p}K_\mathrm{a}$ etc.)—is it possible?

andselisk
  • 37,604
  • 14
  • 131
  • 217
NotEvans.
  • 17,137
  • 4
  • 69
  • 137
  • There is the Competitive Fragmentation Modelling method, which attempts to predict MS via machine learning: https://link.springer.com/article/10.1007/s11306-014-0676-4 – S R Maiti Mar 05 '21 at 22:26
  • 1
    Looks like an interesting paper, but looks like it’s focus is predicting MS/MS fragmentation – NotEvans. Mar 05 '21 at 23:09
  • 1
    I'm not sure how uniform LC-MS results are across platforms (unlike say NMR). How much variation can you expect if you perform the experiment with a different instrument? Will a few calibrations suffice to make intensities comparable? If the answer is "yes" then there is no good reason to suppose simulation (prediction) should be very difficult (or impossible). – Buck Thorn Mar 06 '21 at 04:59
  • For QM calculations of EI-MS this article is an interesting read: https://pubs.acs.org/doi/10.1021/acsomega.9b02011. The outlook also mentions potential for CID-MS. – awvwgk Mar 09 '21 at 18:23

1 Answers1

1

The absolute response coefficient (i.e. peak intensity in counts per mM or similar units) will be way too strong a function of instrument-specific parameters like solvent composition, pH, one or two or three temperature variables specific to the ion source, one or or two or three more voltage parameters particular to the ion source, whether a given emitter is dirty or fresh, how big it is, etc.

However, it may be possible to predict relative response coefficients across a series of compounds. Most approaches I've seen rely on prediction of gas-phase basicity values, or alternately proton affinity of gas-phase neutral species (the assumption being that you are predicting the ionization efficiency of singly-charged ions in positive ionization mode arising from neutral molecules).

Folks have been working on "machine learning" approaches in this area. Here is one example...

Curt F.
  • 21,884
  • 2
  • 60
  • 115