Sunday, February 14, 2016

SUPERQuant test number 1

So I actually read some of the SuperQuant paper and then set up the method to run like this. After multiple failures that ended up just being me running out of C:\ space, I got a couple of runs in.

At first approach, I think the darned thing totally works. I also think that it needs some further evaluation.

I'll queue up a bunch of stuff to run now that I have some space, but here are a couple of screenshots.

Experiment here:
HeLa 1ug 120min run on 25cm EasySpray columns running in High/High mode (this is the normal HeLa file that I have used for just about every experiment you see on this blog that used to be publicly available via my old FTP site, so I know a lot of you out there have it.
Ran just like above for the SUPERQuant runs and without the 2 new nodes in the nonSuperQuant comparator.



Interesting!  In the end, what do we get? 39 new proteins. Okay. Worth noting that this took approximately 50% more total processing time than what I had with the regular method.

Check out how many MS/MS spectra it thinks it looked at!!!  It thinks it saw over 30,000 more MS/MS spectra after the processing. Through that Percolator actually ended up with FEWER PSMs. One thing we know about Percolator, though is that is works better the more MS/MS spectra it sees, right?

Does that mean that the matches I'm looking at after SUPERQuant are better than those before? I'm trying to come up with a good way to assess. I know for sure that the worst scoring peptide (and totally not a good match) disappears after SUPERQuanting. But the output here is just a little dense without plotting. What I do know is that I have more Protein groups ID'ed on a small quick dataset after SuperQuanting, and it didn't take long to install OR run and a really bad peptide that slipped through all my filters disappears after using these cool free nodes.

Sounds like a darned win to me. I'll feed it some more stuff and see if I can make sense out of it, but I definitely recommend you check this out!

EDIT: Another couple runs finished up!  Bigger dataset (55K MS/MS scans OT/OT on a Lumos)
SuperQuant gives me 100 new protein groups from over 1,000 new PSMs and its still hard to tell, honestly if the data coming out is better, but I feel pretty confident that it isn't markedly worse. Here is a quick screenshot. PSM #s vs. XCorr (I know...if you come up with a smarter metric, let me know...)

 I let the binning occur automatically, so it isn't identical, but its not too far off. Both runs come back with some peptides with XCorrs under 1.5, but the numbers are similar.

The 1,000 new PSMs look as solid to me as the rest of the data. More data for free? Consider me all signed up!