mProphet report and redundant library

mProphet report and redundant library heyang  2021-09-22

Morning Skyline team,

I have some questions want to get help.

  1. How can I export mProphet p value and q value in report? I tried, but can't find p value. Only find detect Q value, is it the same with q value in mProphet ?
  2. what redundant library have? only contain redundant peptide info? could we delete them?


heyang responded:  2021-09-22

Forgot to mention version of Skyline, it is

Nick Shulman responded:  2021-09-22
If you train an mProphet model, and use the model to pick peaks ("Refine > Reintegrate"), then you can find the "Detection Z-Score" and Detection Q Value" in the document grid.

There are other scenarios where you might get a q-value in there without using an mProphet model. For instance, if you have an EncyclopeDIA library, Skyline will read the score from the .elib file and display that value in the "Detection Q Value" column.

The Z-score has something to do with number of standard deviations. For this reason, it is a very simple formula to go from the Z-score to the p-value (or maybe it's a simple formula to go from Z-score to q-value-- I can't remember). Like a z-score of 2 corresponds to the 95% confidence interval which means a p-value of 0.95. (I might have even gotten those example numbers wrong).

Yes, you can delete the redundant library. Skyline looks in the redundant library in order to give you a choice of spectra to look at in the Library Match window. Skyline only ever looks at the "best" spectrum when deciding which transitions to add to the document, and the best spectrum is stored in the regular .blib file. If you delete the .redundant.blib file, the only place you will notice a difference is the Library Match window.
-- Nick
Brendan MacLean responded:  2021-09-22
Yes, you can easily calculate an unadjusted p value from a z score. In R, the function to use is pnorm(). So, as described by Nick, a range of +/- 2 SDs is roughly a 95% confidence interval with 2.5% of the distribution below -2 and above 2.

You can see that with pnorm() as follows:
> pnorm(2)
[1] 0.9772499
> pnorm(-2)
[1] 0.02275013

Though, +/- 1.96 is closer to the true value:
> pnorm(1.96)
[1] 0.9750021
> pnorm(-1.96)
[1] 0.0249979

However, with multiple hypothesis testing like mProphet it is not generally acceptable to mention the unadjusted p value. Instead, you want the q value, which is an "adjusted p value" meaning that it refers to the false-discovery rate of the entire set, while the unadjusted p value is only acceptable to use in a controled experiment with a single well-defined test.

So, you really want to use the "Detection Q Value" and "Detection Z Score" and avoid talking about the p value you can easily get from pnorm(Detection Z Score).

Hope this makes sense and helps to clarify Nick's response.


P.S. - The redundant library can also give you the ability to add to an existing spectral library. Whereas, if you only have the non-redundant library that is a fixed artifact to which you cannot add new data. I agree with Nick, though, that in many cases you don't really care about the uses of the redundant library and you can delete it without harmful impact.
heyang responded:  2021-09-22
Thank you, Nick.

I also have another question about retention time calibration. In attached Skyline menu file which suggest use ms/ms ID time from DDA data (page 15). Is this good enough or we need to do rt calibration for every time?

heyang responded:  2021-09-22
Thank you very much, Brendan for your explanation. Heyi
heyang responded:  2021-09-24
Hi Brendon and Nick,

More question about Detection Q value and Z score. Do you have detailed instruction about these two score for beginner? For a beginner to quick check their data, what are threshold of values for 2 scores? For Z score, -3 and 3 are the same significant? Also I processed 30 samples in a batch mode. Looks same peptide from different sample have same Q value, but different Z score. Is this right?