[BioC] affy, how to define the normalization method that works with my data

[Dipl.-Ing. Johannes Rainer]

hi to all bioconductor users!

our goal is to find differentially expressed genes in cell line and patient
probes with and without a given treatment. so usually we have two chips (we
have not enough money to do replicates, i think that's fairly common with
affymetrix ;) ), one chip has RNA with treatment, the other one without.
i currently try to find out what normalization method gives us the best results,
i am using mainly the mas5 method implemented in the affy package, the rma
method and a "hybrid" version i call rma/mas, because bg correction is done
with the rma method, quantiles is used as the normalization step pm signals are
not corrected and finaly as the summary method i use mas5.

is there any possibility to find out what for a method normalizes my data best?

currently i am looking at the shape of the MA plots that i create from the
normalized data. there the rma method looks best, no big variance in the lower
expression level, but i have currently a problem with rma, because i am not
shure how well the model parameter are fitted into the data, when i have only
the data from two chips to calculate them. it looks good, better then mas5 and
rma/mas.
i am glad to have also a set of positive control genes, that i found by a
literature search. so i can check if these genes are regulated in all of the
methods, and they are. so from this point of view all methods work well.

so i repeat my question: has anyone experience in analyzing affy chip with only
two or four chips available? what methods for normalizing do you use, why? is
there a way to find out (on my data without replicates) what method performs
best?

sorry for this very long description of my work, but i need some discussion,
because i am the only bioinformatician in this lab :(

thanks, jo