search

vdemichev / DiaNN

DIA-NN - a universal automated software suite for DIA proteomics data analysis.
Other
279 stars 54 forks source link

Excessive ram usage during quantification on DIA-NN 1.8.2 beta 27 #938

Open Cllozano opened 9 months ago

Cllozano commented 9 months ago

Hi Vadim,

I have been testing the last beta and have several questions:

Spectral library generation is a lot faster with 1.8.2. However, I have attempted interpreting raw files from an Astral instrument with a database including almost all modifications (Fasta is 200Mo and SL is 35go), and it did not run through. I suspect it needed more RAM than available on the computer (256 go). It seems that the QuantUMS algorithm is more demanding for the quantification step ?

Also, when comparing peptide quantifications (1.8.1 v 1.8.2 b27) on an Hela, we have the following correlation: image

Interestingly, we see a better correlation in the low abundance compared to more abundant peptides. Is that something you noticed already ?

Thanks for your help, Best, Clem

vdemichev commented 8 months ago

Hi Clem,

Sorry for the late reply. Considering all possible modifications is not a good idea indeed. Regardless of hardware requirements and not even for DDA, it's been shown in several open search papers that DDA has substantial peptidoform error rate, unless open search is used. RAM needed I think is roughly ~1kb / precursor in the library, so if it's 100M precursors you'll need 100Gb to just keep the library in memory.

Absolute quantities are per se not comparable between methods or between different peptide species, that's OK.

Best, Vadim