search

MannLabs / alphapeptdeep

Deep learning framework for proteomics
Apache License 2.0
102 stars 20 forks source link

Is it possible to get 1/K0 mobility instead of CCS from AlpahPept CCS predictions? #116

Closed KarlClauser closed 8 months ago

KarlClauser commented 9 months ago

I am using the AlphaPept_ccs predictor through the Koina server. https://koina.proteomicsdb.org/docs/#post-/AlphaPept_ccs_generic/infer

The only described output is 'ccs'. I am seeking to compare the predictions with measured ion mobilities in units of 1/K0 from our Bruker timsTOF SCP instrument. Is there an undocumented option to get the output directly as 1/K0? Or do I need to do that myself?

The following sentence is in the methods section of your paper Zeng et al NatComm_2022: The predicted CCS values are converted to mobilities of Bruker timsTOF using the Mason Schamp equation.

Thanks, --Karl

GeorgWa commented 8 months ago

Unfortunately, I don't know about the Koina interface to alphaPeptDeep. Maybe my collaegue @jalew188 knows more about this. When using the library or the gui, alphaPeptDeep provides mobility 1/K0 values by default. See this tutorial: https://alphapeptdeep.readthedocs.io/en/latest/nbs/tutorial_speclib_from_fasta.html

LLautenbacher commented 8 months ago

Hi, I'm the main developer of Koina at the moment. @jalew188 helped me beginning of the year at EUBIC to integrate some AlphaPept models into Koina.

I had a quick look around the AlphaPept codebase. And to me it looks like this code here is converting the predicted CCS values to mobility values.

If you can confirm that this transformation generates the mobility values that @KarlClauser expects I can implement this in Koina as an alternative output.

GeorgWa commented 8 months ago

Hi @LLautenbacher , Yes that's right. Thank you for implementing this in Koina. I wasn't aware of Koina yet but it looks really cool! I'm looking forward to try it out.

If this solves the issue I will close it later this week.