[BioC] permutation/resampling FDR correction
Hannah at mpimp-golm.mpg.de
Thu Jul 14 10:32:29 CEST 2005
Rather than use the linear step-up FDR procedure I'd like to use a
permutation method to correct p-values from limma.
I use linear regression of 8 groups with 3 replicates against a
numerical measure for the eight groups.
x <- c(1,1,1,3,3,3,6,6,6,4,4,4,8,8,8,7,7,7,5.5,5.5,5.5,6,6,6)
design <- model.matrix(~x)
I want to run 1000 permutations to get p-values I can then use to
estimate the null distribution to do an FDR correction of the p-values.
I guess I can quantile normalise the p-values to combine them into a
single distribution to estimate the null - but is this ok??
If I want to control at a specific FDR then I could look at say p=0.001
and find 1000 significant in the test set and 10 in the null and
conclude I'm controlling at an FDR of 0.01. However, I'm looking for a
continuous correction like for the step-up FDR where the FDR corrected
p-values are the output and you can select whatever level you choose.
Has anyone got some practical advice/code/link to relevant function that
would allow this.
Thanks in advance,
More information about the Bioconductor