publications on logCPM transformation of MS intensities

[Ivo Kwee]

Sorry for the late question. Is there any peer-reviewed publications explaining the logic behind the logCPM transformation of mass spectrometry intensities ?

[originally from Felipe da Veiga Leprevost @ github 27july22]

[Ivo Kwee]

In our online documentation we have some explanation and references https://omicsplayground.readthedocs.io/en/latest/faq.html#can-i-use-limma-edger-and-deseq2-for-my-proteomics-data

[BigOmics Analytics Team]

Felipe: "I'm familiar with most of those publications. However, I don't recall reading about applying statistical transformations designed for discrete distributions on the relative ion abundances. Please correct me if I'm wrong. I would understand if you apply them to spectral counting or peptide counting, but the intensities originate from completely different technologies with different proprieties."

[BigOmics Analytics Team]

OK. So, you are making a difference between spectral counting and intensities (incl LFQ)? Up to now, we mostly handled only proteomics data from LS MS/MS as intensities (or LFQ intensities). 

Yes,  for discrete distributions, the logCPM normalizes the sum of the counts to a million, then takes the logarithm. On continuous signals, like the intensities, "logCPM" would scale the intensities to "million intensity units", then take the logarithm, which is equivalent to taking the logarithm and shifting all values with some constant. Maybe it would be better to call it logPMI  ("log per million intensity") but the calculation would still be the same. 

Are you having doubt on the "CPM" step, or the "logarithm" step for continous intensities? 

Ivo

[Felipe da Veiga Leprevost]

Doubt would be a strong word. I have been working with proteomics data analysis for some year now, and I never heard about this approach. As customary in the scientific field, when a new analysis method is introduced, a white paper or a peer-reviewed paper usually follows it. I understand that this might be a well-known method to handle RNA-Seq (count) data, but the proteomics community has been using different approaches, and there some reports of methods borrowed from the Genomics field under-representing.