MSX data processing is prohibitively slow

support
MSX data processing is prohibitively slow Tim McCubbin  2019-03-31 19:18
 

Dear Skyline team,

We are presently attempting to assess the benefits of using multiplexing to assist in the label free quantification of complex samples, trialing this on a HeLa protein standard. The tutorial on how to do this was clear and simple to follow and we have acquired our MSX scans. However, the import of these result files into Skyline is incredibly slow. While we expected it to be slow given the tutorial said upfront that one user reported a 20 hour import time for a 200 MB file several years ago, a single result file is importing at a rate of ~2% per day (our files are just shy of 200 MB each). Because we expected it to be slow, we used slightly less windows than in the tutorial for our trial (4 windows of width 6m/z). Data is from a Q-Exactive HF-X and our library size is quite large as we are doing untargeted proteomics, so roughly 55,000 peptides and 680,000 transitions. We are currently running version 4.2.0.19009 of Skyline on a Windows 10 machine with 64GB RAM and a 6 core hyper-threaded Xeon processor. I thought our computer was of reasonable specs so was wondering if you had a sense of why the import is slow; are we simply not meant to be doing untargeted proteomics with MSX or is it likely we have messed up a setting? Please let me know if you need any files or more information.

Many thanks,

Tim

 
 
Nick Shulman responded:  2019-04-01 10:52
Hi, Tim,

We do not recommend using the "Deconvolution" options that are available on the "Edit Isolation Scheme" dialog in Skyline. We have been meaning to remove that feature from Skyline for a while, but have not gotten around to it.

The recommended way of doing deconvolution is to use MSConvert to create a .mzML file which has extra MS2 scans representing the deconvoluted data.
Here's a page that talks about how to do that in the MSConvert user interface:
https://skyline.ms/wiki/home/software/Skyline/page.view?name=DemultiplexingOverlappingDIAWindows

The demultiplexing in MSConvert (which we call "Prism") looks at the entire MS2 scans, and decides what the deconvoluted spectra should look like.
The algorithm in Skyline looks at all of the transitions in your target list, and only deconvolutes the part of the spectra that you are selecting for. This becomes prohibitively slow when the number of transitions you are monitoring becomes even a modest number.

We had been planning on exposing the Prism algorithm in Skyline, but it is still really slow, such that if there was any chance you might have to extract chromatograms more than once, you are better off saving the results of deconvolution as a .mzML.
-- Nick
 
Tim McCubbin responded:  2019-04-04 22:48
Hi Nick,

Thank you for the quick response. All works perfectly (and quite rapidly!) using MSConvert.

Many thanks,
Tim