[BioC] Customize normalization on specific time series experiment?

[andrea.grilli at ior.it]

Hi to all,
I'm performing a time-series analysis on 2 cell lines with 4 time  
points, with 2 replicates at each time. I imported the data with  
"Affy" library, and I normalized with RMA.

QC parameters (average background, scale factors, % present calls,  
ratios 5'-3') are all in the suggested range (according to the book of  
Gentleman et al., chapter 3), but when I look at probes expressions,  
they have one behaviour in 2 time points (1st-3rd), and a different  
behaviour (wider expression) in the remaining 2 time points (2nd-4th),  
so probes seem to go up and down during time.
Also, I detected genes d.e. at each time point with linear models, and  
clustering analysis with these d.e. genes groups samples of time  
points 1-3 in one major group, and those of 2-4 in a second-one,  
regardless of the cell line type.

In fact, the arrays have been hybridized in different moments, one for
time-points 1-3, months later for time-points 2-4.
So, technically everything seems ok, but all the genes have this
behaviour of "up and down" across time and this seems to affect  
following analysis.

- Which could be a normalization that can reduce this (probably  
technical) bias?

- Could be a good idea normalizing e.g. on the control probes and is  
this approach reasonable for this case? How this can be done?

- Or you suggest that because QC is ok to leave things as they are with
RMA normalization?


Thanks for your time,
Andrea