[BioC] Technical replicates are really off - how to fix?

[Ken Termiso]

Hi all,

>From a recent experiment with affy 133v2 chips, I had four different cases, 
each with three controls and three experimental chips. The four cases are 
analyzed separately, since they all rec'd different treatments. Combining 
them doesn't work since they are all so different.

Due to the world being an imperfect place, I only have technical replicates, 
so all of my statistics are based on these technical replicates. (In the 
past, from similar experiments, I've been able to confirm ~90% of 
differential expression with RT-PCR based on this strategy).

The problem is, that for one of the cases, two of the experimental chips are 
much, much brighter than the other one. The six chips were normalized 
together using RMA, which for this tissue has been the superior 
normalization all along vs the others we've tried (GCRMA, MAS5.0, and 
li-wong). Here is a random selection of probesets to illustrate what I'm 
saying -

Control 1	Control 2	Control 3		Exp 1		Exp 2		Exp 3
3514.66		2917.78		1960.94		3915.28		13710.44	13385.72
2272.28		2004.44		1387.7		2751.73		8759.5		8596.44
568.71		396.12		335.61		614.27		1617.23		2374.7
1754.51		1250.99		1082.94		1781.94		7261.92		7679.01
2156.23		1709.61		1394.24		2793.15		8053.99		11112.1
468.04		452.48		382.7		605.1		1692.47		3202.71

Our (somewhat lousy) core processed these chips about a year ago, so I'm 
wondering if there's anything I can do to deal with the last two 
experimental chips, which are way, way off...I'd like to bring them down to 
the realm of the other exp chip without introducing any bias before doing 
stats to find significant, diff expressed genes.

Thanks in advance,
Ken