Does xiSearch consider substitutions by default?

[michabirklbauer]

Hi,
I am currently looking through a result from xiSearch (non-validated, so the direct output without xiFDR) and I am seeing some very weird CSMs that feature peptides with various amino acid substitutions (see picture below). I did not configure amino acid substitutions in the config, does xiSearch consider them by default? Is there a way to turn them off?
This was with xiSearch version 1.7.6.7, so not the most recent one (sorry).
I have also attached the config file and output file directly.
Thanks a lot in advance!

Config and results: xiSearch.zip

[grandrea]

Hi,
Notice that that PSM is between decoy peptides. It is in fact a decoy-decoy match (protein1Decoy=1 and protein2Decoy=1) so that is not a target CSM. I would wait for @lutzfischer to be sure, but I am guessing that that keeps track of the reverse decoy generation to still generate a tryptic sequence for the decoy peoptides. Hence the shuffling of S to K and K to S to keep the same mass relative to the original peptide but generate a tryptic decoy

[grandrea]

Also- you are currently defining modifications twice in your config both inside the crosslinker as fully variable (searched on everything) and in separate lines as linear only. So you are searching them twice with different names. The definition inside the crosslinker is sufficient and you can remove the linear modification lines.

[grandrea]

if you want to include STY only as linear modifications, remove K from the linearmodifications line as the K modifications are searched already from inside the crosslinker definition

[michabirklbauer]

Oh yeah, that makes sense! Thanks for the explanation, I will correct it!

[lutzfischer]

The reason you see this is, that xiSEARCH does it's best to keep a equivalent target and decoy database.

The basic "problem" is that a peptide should be either target or decoy (just makes FDR so much easier). However when generating decoy peptides (by digesting decoy proteins) it does happen that a decoy peptide sequence does already exist as a target peptide. Now the simplest resolution would be to drop the decoy - however that means your decoy DB gets smaller.

To address that xiSEARCH actually generates permutations of these peptides and if it finds one not seen as target yet, accepts that one.

Mainly happens for short peptides and if you database size increases the smaller becomes the chance to find a novel permutation. Hence for very large databases the target decoy equivalence for short peptides becomes problematic.

This is by the way one reason why xiFDR excludes short peptides from results before filtering.
But for anything longer then 5 amino acids you either need to have a very large database for the permutation not to result in something usable or be very unlucky with your peptides.

[lutzfischer]

Actually the observation that it is no longer tryptic is kinda important. But at this point the question would be: is having a non-tryptic decoy peptide worse than one less decoy peptide.

I opted for accepting non-tryptic looking peptides. Admittedly they probably will be somewhat disfavoured in matching. Meaning the equivalence for these is also not 100%, but I think it is still better then not having them at all.

[michabirklbauer]

Ah okay, I understand! Good to know, thanks a lot for the comprehensive explanation!