[BioC] p-value/B-statistic interpretation

[Mike Schaffer]

I know this is probably beaten to death, but I can't seem to find a 
satisfactory answer.

How can the p-values/B-statistics from limma be properly interpreted?

With assumptions satisfied, an FDR corrected p-value cutoff should 
produce a list of induced/repressed genes that includes a given 
percentage of false positives.  However, we all know that we cannot 
assume independence with arrays.  So, how does one rationalize a 
p-value or B-statistic cutoff to get beyond just a list of the top X 
genes?  Does this dependence render the p-values completely 
meaningless?

My "problem" is that my FDR corrected p-values are incredibly low 
(<1x10-4) and a moderate p-value cutoff produces a list of over 10% of 
the genes on my array.  Now, I can obviously lower the cutoff, but how 
does one decide where to draw the line?  Is it just empirically by how 
many genes I *do* expect to see induced/repressed.  Since I don't know 
this answer, this places me in a untoward position of justifying the 
rationale.

Thanks in advance.

--
Mike