This group tested some and the moral of the story might be to use a bunch of them.
My setup uses MSFragger, MSAmanda 2.0, MSPepSearch and SequestHT and Percolator to sort it all out at the end, so maybe my solo quest to do meaningful SCP here is on the right track in at least one regard.
I like that this group digs deep into the PTMs they can detect in single cells. It wasn't all that long ago that was a super controversial thing to say you were seeing. Are they the most convenient PTMs possible on the highest copy number proteins that are there? Yes and yes. Still there, though!
Nice little paper, and I always appreciate a good clear UpSetR plot.
ReplyDelete>Finally, we note that the different sets of proteins and peptides that are identified by the different algorithms suggests important questions about the reliability of the results.
Key point: unless you’re harmonizing the protein inference method and/or using an entrapment database, these comparisons aren’t very meaningful.
>In contrast, the proportion of peptide deamidation (N and Q) differs greatly across the different programs. For example, in the MaxQuant results for this dataset, only 6.2% of identified peptides were deamidated, while in the MSFragger results, 24.3% of peptides showed deamidation, and in the remaining programs, the percentage ranged from 9.5 to 15.6%.
If there’s this much variability, that indicates to me that there’s an issue with calibration somewhere that’s causing deamidation to be misclassifies as an isotopic peak or vice versa. And realistically, this PTM is generally an artifact, so identifying the peptide with a C13 instead of a deamidation is rarely meaningful.
Unfortunately, the search parameters applied in this paper are flawed leading to poor analysis/conclusions. The five search engines run through ProHits applied a low-res 0.6 Da fragment bin tolerance while SEQUEST HT (and presumably Andromeda in MaxQuant) had high-res MS2 settings applied. The three analyzed datasets were acquired in high-res MS2.
ReplyDelete