mProphet scoring of removed peaks Tobi  2019-12-11
 

Dear all,

I wanted to report some strange behavior of Q-values and z-scores as shown in the attached screenshot and the uploaded file ,,mProphet scoring on removed peaks". (I expect it to show up pls let me know if not)

When working with DIA and peptide identification in skyline, when removing peaks manually (right-click, remove peak) and via Avant-Garde DIA, those peptides yield Q-values and Z-scores as if nothing happened. I even checked with a different scoring model and see values changing aka refreshing, but peptides where peaks were removed (see RT as well as dotp = 0) still show values from prev. existing peak boundaries. I would expect something like #N/A.

Could you please take a look at this? I guess a workaround would be to filter peptides where dotp > 0.00 or where RT fails.

Can you please also think of extending the document grid for exporting all intermediate values of the scoring? Or is there already a way to score Skyline curated data with Percolator? (When decoys might not be normally distributed etc.).

Looing forward getting in touch with you.

Best,
tobi

 
 
Nick Shulman responded:  2019-12-11
When you reintegrate, Skyline ignores all of the peptides where the peak was manually changed by the user. It sounds like Skyline really should null out the Q-value and Z-score, but for now, it sounds like Skyline just skips over those peptides.

You should probably ignore the Q-value and Z-scores in all rows where "User Set Total" is "True", since this indicates that the peak was manually changed by the user.

You can see the intermediate values for scoring by going to:
File > Export > mProphet Features
That's not quite as convenient as having them in the Document Grid, but it does have all the information you would need to calculate the final scores yourself if you know the weightings.

I suppose we could add those scores to the Document Grid. It would be tricky since the scores apply to candidate peaks, and there are multiple candidate peaks per Precursor Result, and multiple scores per candidate peak, so that would result in a grid with a lot of columns unless we did something clever.
-- Nick
 
Tobi responded:  2019-12-12
Dear Nick,

thank you very much for the fast and wholesome response.

The "User Set Total" filtering works great for manually curated data but not for Avant-Garde since everything is annotated as "IMPORTED". Its still really good to know this feature exists.

The MProphet export looks very useful, sry I missed it by looking just at export templates and doc grid before. Maybe adding the p-value to the already existing q and z scores in doc grid would be a nice,small addition.

For the scoring, is there a reason to ignore manually adjusted peaks? If I would curate my data manually or nowadays with AVG-DIA I would like to rescore z,p, and q values without having the peak boundary adjusted another time by "overwrite manual intergration" (Because AVG peak picking seems quite decent and less likely to pick good second best peaks instead of the true peak but thats just a feeling). I guess the current work around would be to only do any kind of scoring after peak picking and curation is finalized.

So far thank you very much for the help, its highly appreciated.

Best,
tobi