Deriving p values for gene correlations by permutation analysis

Question

[chris86]

Hi

I am calculating gene correlations for all patients gene expression data against clinical parameters. Then I want to derive p values by permutation. Is it OK to randomly permute the sample columns and then re calculate the correlation for every gene N times. Then ask how many times are correlations observed equal to or greater than a genes correlation by chance?

E.g. If my correlation is 0.3

Then the p value would be, n obs > 0.3/ total obs

So I have one distribution per gene?

Thanks,

Robert

[dariober]

It sounds about right to me. Have a look also at the boot package in R (and this FAQ http://www.ats.ucla.edu/stat/r/faq/boot.htm). Once you have the p-values, I guess you should correct them for the multiple tests you performed (i.e. one test per gene).

[chris86]

Yeah I hadn't done multiple testing correction, that is a good point