Best way to calculate p-value for protein regulations

Question

[alex]

Hey,

I'm currently working on label-free protein quantification from LC-MS data.

Does anyone of you guys know how to calculate p-values for protein regualtions?

Right now Im using a(n unpaired) t-Test, where the input is simply all measured peptide intensities.

Vulcano plots look quite nice but currently we are understimating the significance of regulations.

I'm sure there are more sofisticated ways how to do it.

Many thanks!

Answer 1

[Laurent]

You can use the standard tests for your proteomics data - classical t-test, limma, ... and then adjust the p-values. The biggest offender with label-free MS1-level (area under the curve) proteomics will be the missing values.

Update: you might also look at the Bioconductor package MSstats for dedicated statistical infrastructure.

[alex]

Missing values are not a problem, we have implemented a way to use artifical values in this case which provide a background intensity value.

Since these intensities are very low, i assume this is somewhat problematically for calculating p-values.

[Laurent]

Missing value imputation is a concern in general. In your case, I would check that the distribution of the values after imputation is still compatible with the assumptions of the test.

[alex]

Is it valid to calculate a paired test on matching peptides?

[Laurent]

It is difficult to answer that question without information about your experimental design, but I would say no. You would use a paired test when each replicate i of condition C1 matches replicate i of condition C2 - same samples at different time points, or same individuals before/after drug administration. You will always want to compare matching peptides (if that's what you are applying you test on) whether you use a paired test or not.