[BioC] RMA false positive rate

[Matthew Hannah]

Partly in response to Anthony's question and to further the low rep affy discussion
I'd like to get more views on this.

I have 3 paired biological reps of untreated (U) and treated (T), normalised by RMA 
and in this case looked at 2 fold changes. I made the U-T pairwise comparison for 
each replicate and counted the number of genes where fold change >2. To assess false 
positives I compared within the biological replica 1-2, 1-3 and 2-3 for genes >2.
I got the following results.

	1rep		2rep		3rep
U-T	1544		1201		1008	These are the changes due to treatment
U	31		13		1	False positives?
T	66		27		4		"

The biological replica seem to be very good, and the within replica comparisons 
suggest very low false positive rates. However the inclusion of an additional
replica loses c.200-300 genes in the U-T, whilst the within replica comparisons
suggest much lower false positives. I guess this is because no account of variance
has been taken and there will be many genes with distributions around the 2fold
change for U-T, but any other comments on this approach?

If you were a reviewer what would be your comments on assessing false positive rates
by comparing between biological replica versus comparing U-T?

Thanks
Matt