search

mobiusklein / ms_deisotope

A library for deisotoping and charge state deconvolution of complex mass spectra
https://mobiusklein.github.io/ms_deisotope
Apache License 2.0
33 stars 14 forks source link

Expose classes from peak_retention_strategy #25

Closed armonyG closed 2 years ago

armonyG commented 2 years ago

After importing ms_deisotope the classes from ms_deisotope.deconvolution.peak_retention_strategy is missing. These classes are needed for the retention_strategy argument of deconvolute_peaks()

mobiusklein commented 2 years ago

Thank you. This feature was written to test an idea but I found I didn't use it in practice. Other people might.

armonyG commented 2 years ago

We are working with glycopeptides (on timsTOF pro) and we found that keeping the top 150 non deconvoluted peaks (in Bruker's DataAnalysis) yields better identifications.

mobiusklein commented 2 years ago

Are you changing the other parameters of the TopNRetentionStrategy from the defaults?

They're currently set to something pretty rough based upon the point where the intensity of the second peak of the isotopic pattern starts to exceed the monoisotopic peak calculated with the glycopeptide averagine (in a noise-free context), which would make this method report the wrong mass.

When you use this feature, do you also have a high isotopic fit score threshold? If so, which scoring method do you use?

armonyG commented 2 years ago

This is literally my first time working with the library so I haven't changed any settings (yet). Eventually I have a need to deconvolute isolated fragmentation spectra before I process them further.

I'll start with setting the filters off: base_peak_coefficient=0 and max_mass=spectrum max but I do plan to see if they are useful. I am taking a pragmatic approach and comparing database search performance. The wrong isotope mass might still be useful if the database search can handle it (I am using PeoLuCID which I know can handle this as MS1, not sure about MS2). I plan to compare the different algorithms to see which one provide good results while still not taking too long.

If you are interested, I can share my conclusions.

mobiusklein commented 2 years ago

I would be interested to hear how that turns out. I don't think timsTOF spectra have a dense noise baseline, but it might depend upon how you read them out, and how much energy you're using when you're fragmenting the glycopeptides (or are they deglycosylated?).

I know some SEQUEST-derivative search engines can search for isotopic peaks of product ions, but that tends to be tacked on rather than a core part of the search and scoring model. I know that is the case with Comet, at least. The off-by-neutron error is fairly common on the precursor.