Hi all,

Sorry if this question has already been addressed, I did not really found an answer on the site...

I have a dataset of microarray (agilent 1 color) with ~35 condition (corresponding to 3 time series, but with different time points, and where time cannot be consider as a continuous variable), 2 rep each. I have imported them with limma and normalized with vsn normalization.

I would like to identify gene that change in at some point in one of the time course only (corresponding to 7 conditions), like an anova or a likelihood ratio test, in order to perform clustering after.

From the vignette I understood that the F-test of limma would give me what I want, correct?

In the vignette example, to do it on 3 condition, a contrast is made with all pairwise comparisons ; do I really have to do that? if I just use lmfit() and eBayes():

> fit <- lmFit(arrayNorm, design)

> it <- eBayes(fit))

I got a F.p.Value in my fit object, does this correspond to the F-test for all the element in the design matrix?

If so, to perform the test on just a subset of my data, can I just give a design matrix with only the samples and factors that I am interested in?

I understood from few post online that Benjamini Hochberg correction is not optimal for F-test (am I wrong?), but I did not really got what should I use instead in this case. Could someone light me on that?

Thanks!

THIS!

"Just keep the same fit object and extract the contrast you want. That is faster, easier and better."