[BioC] RMA with biological replicates

[Dipl.-Ing. Johannes Rainer]

hi,

i have again a question about RMA and how it works with a different number of
chips.
we are looking at the response of patients to a specific treatment, so we got
biological replicates and no technical ones (as we have not enough material
from the patients). at the moment i got 9 chips, 3 patients, 3 time points for
each patient.
What is now the best way to normalize them? normalize them all together with
RMA, or normalize the chips from each patient separatly (that means normalize
the 3 chips from patient one together, then those from patient 2...)?
As i think, RMA gives better results if i have more chips to normalize together
(more chips with roughly the same expression result in better fitted model
parameters i guess).
has anyone a conclusion what's the best was?

thanks, jo