2
$\begingroup$

I am new to proteomics analysis, so apologies if my question is silly! For context, I am working with proteome samples from the postmortem human brain for a case-control study.

Our lab generally receives an output file of Proteome Discoverer* (v2.5) from the proteomics center that we work with. From what I understand, PD perform t-tests on the normalized data, but it is a background-based t-test, so there is a weighting factor is applied that takes into account the distribution of ratios for all other proteins and/or the protein abundances themselves? As such, I am able to see an "Abundance Ratio Adj. P-Value".

My problem is that my data does not meet the assumptions for a parametric test and it has a small sample size, so the t-test is not valid (as far as I can tell, unless there is some sort of log transformation that PD does to the data before the t-test). Therefore, I would like to run a Mann-Whitney U Test. Can this be done in the PD software?

If not, how can I do this background adjustment manually such that I can run a Mann-Whitney U Test but still take into protein abundance like in PD? I do most of my stats in R in case that is relevant.

Thank you so much in advance for your help!

Kelly

Package distributed by ThermoFisher

Improve this question
$\endgroup$
3
  • $\begingroup$ Overall, this is a good call and the right thing to do. It is worth contact ThermoFisher tech. support to confirm the absence of non-parametric in the packages. Thats what the labs paying for. $\endgroup$
    – M__ ♦
    Nov 5, 2022 at 22:33
  • $\begingroup$ Thanks for your reply! I actually did contact Thermo, it doesn't seem as though they have non-parametric options, but I am trying to get information on the specific calculations used for controlling for protein abundance/abundance ratio adjusted p-value. Unfortunately, I am not getting anything helpful thus far :( $\endgroup$
    – Kelly
    Nov 7, 2022 at 3:12
  • $\begingroup$ Abundance Ratio Adj. P-Value - How is it calculated - the formula and procedure?what do you mean by normalized data? $\endgroup$
    – Subhash C. Davar
    Feb 9 at 12:57

1 Answer 1

Reset to default
1
$\begingroup$

Small samples size and parametric tests are not mutually exclusive. Specialized software has been developed to leverage sharing information across measurements to boost power, that is called Empirical Bayes. You can use the limma package for this sort of analysis, contrasting groups. Input is expected to be normalized and on log2-scale. It's an R package hosted at Bioconductor. Don't try a MWU, with small sample size that is massively underpowered.

Improve this answer
$\endgroup$
2
  • $\begingroup$ I appreciate the contribution and yes as a basic principle non-parametric tests have lower statistical power than their parametric alternatives, of course. However a significant result with a non-parametric tests is robust because it is assumption free. As we have seen with DSeq2 recently the problem start when a test is frequently deployed outside its original context, sooner or later someone identifies discrepancies and the whole circus falls over. $\endgroup$
    – M__ ♦
    Nov 6, 2022 at 14:05
  • $\begingroup$ Thank you so much for your reply! The reason that I wanted to use a non-parametric test is because the normalized data I receive from PD does not meet the assumptions for a t-test. I am able to do a Mann-Whitney U test, but the problem is that it does not take into account the protein abundance as described above. Do you know if the limma package you proposed can do this? Thanks again so much for your help. $\endgroup$
    – Kelly
    Nov 7, 2022 at 3:10

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.