Multiprocessing on Windows

[blackjack6666]

Hi,

Thanks for your contribution in this program to extract MS1 features as I was looking for some tool to build a MS analysis pipeline. I have one question though, it looks like the number of processor is 1 on Windows system no matter what the setting is. I have only tried it on Windows, does the multiprocessing work only on Linux?

Thanks,

Tony

[markmipt]

Hi Tony!

We have a lot of problems with using Python multiprocessing module on Windows. In short, it consumes a lot of memory and can even slow down calculations. Some explanation can be found here: https://pythonforthelab.com/blog/differences-between-multiprocessing-windows-and-linux/

This can probably be fixed by changing the biosaur2 architecture, but that's not in our immediate plans.

By the way, only a few time-consuming functions inside the algorithm were parallelized, so there is a non-linear speedup due to the use of multiple cores. For example, on a random file I have 139 seconds (1 core) vs. 68 s (5 cores). If your computations are very slow, there may be other issues. For example, when using TOF spectra, you will have to play with the "-mini" (minimum intensity) option to cut off noise peaks. Or if you have data from long gradients with a lot of MS1 spectra, you should increase the "-minlh" (minimum number of MS1 scans per hill/peptide isotope) option. The default parameters are generally optimal for 5-10 min of LC MS1-only data acquisition using Orbitrap, which only contains a few thousand MS1 scans.

Thanks for your interest,
Regards,
Mark

[blackjack6666]

Hi Mark,

Thanks for your reply and suggestions. It is indeed faster after tunning those parameters. I do have another question regarding quantifying deamidated peptides. Currently we have a target deamidated peptide among a list of target peptides that we'd like to do some relative quantification based on MS1 intensities. The issue is that the deamidated species almost has the same retention time as unmodified species, I think I managed to make the algorithm separate the deamidated/unmodified m/z peaks into two hills by tunning some parameters as they have close m/z values, but still the relative quant of the deamidated peptide looks really off (from the output feature file it looks like the intensity assigned on m/z of deamidated peptide is much more than how it should be).
And I tried on another raw file where gradient is much longer so separation is much better, it did a decent job to calculating the total intensity of the deamidated/unmodified peptide.
So I'm wondering if you have any suggestions on how to use Biosaur2 to get accurate quant results on peptide with close m/z and retention times.

Thanks again for developing this beautiful tool,

Tony

[markmipt]

Dear Tony,

1. Are you sure you're looking at deamidated peptides and not confusing them with something else (like N-15 isotopes, for example)? According to this study (10.1002/elps.201000507](https://doi.org/10.1002/elps.201000507), deamidation should significantly alter retention time and these hills should be separated in RT space. What MS1 resolution do you use in your research?
2. Do you compare intensities Apex (max) or XICs (sum) values?
3. There is an option in biosaur2 called "-hvf", which is threshold to split hills into multiple if local minimum intensity multiplied by hvf is less than both surrounding local maximums (default: 1.3). You can try to play with this option and use values ​​of 1.1, 1.5 or 2.0, for example.
4. We have a software called IQMMA that maps the MS/MS identifications and MS1 features generated by the biosaur. Perhaps this will be useful in your studies. (https://github.com/PostoenkoVI/IQMMA, doi: 10.1021/acs.jproteome.3c00075).

Regards,
Mark

[blackjack6666]

Dear Mark,

Thanks for your prompt response. To answer your question:

1. Yes it's deamidated peptides, we have precursor mass and MS/MS match to confirm. Yes you are right normally deamidation can alter retention time and that's what we saw from the 90 min gradient, our current LC gradient is 30 min with microflow and the retention time difference is only 0.1 min between unmodified and deamidated species for that peptide. Our MS1 resolution is 60K.
2. I used column 'intensitySum' from the output feature file and summed up all charge states for one target peptides, and calculate the relative abundance of deamidation using formula:
   XIC% = total intensity of modified peptide/(total intensity of modified peptide + total intensity of unmodified peptide) *100%
   I can try using intensityApex and see how that goes. To my understanding, intensityApex returns the intensity of most abundant isotopic peak?
3. I tried tunning the 'hvf' parameter already, and it seems lowering it below 1.0 helps identifying the MS1 feature for deamidated peptide, increasing it up lead to the loss of such MS1 feature.
4. Thanks for the suggestion, I'll look into and try IQMMA!

Best regards,

Tony

[markmipt]

Hi Tony,

2. The apex is simply the maximum intensity (among MS1 scans) of the monoisotopic peak (we gave it the same name as in the Dinosaur software, which was the baseline for biosaur2). You can also use option "-iuse" which is the number of isotopes used for intensity calucation. 0 - only mono, -1 - use all, 1 - use mono and first isotope, etc. (default: 0). Thus, the Apex and Sum columns can represent not only a monoisotopic peak, but also the sum of several isotopes.
3. This actually blows my mind (lowering hvf below 1.0). I have never done this and I don't have a clear understanding of what this will lead to. If I understand correctly what is happening, you are splitting the normal (no local minimum, but long enough to spit) RT hills into several. Please look at the pictures in this issue to better understand what hvf is actually doing: Question of detailed algorithm #7

Regards,
Mark