Hi,

Currently I'm trying to analyze a three-color experiment from http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE38522 

I have no problem to read such file, but I don't know how to analyze this kind of chip. After checking their chip design, ctrl is not always on each chip, which suggest that probably differential expression require comparison  between different channel from different chips. I think after they have done normexp background corrected and quantile normalized using limma, ComBat function was used for removing batch effect.(http://onlinelibrary.wiley.com/doi/10.1111/nph.12378/full)

So my question is,

if I treat the different channel from each chip as separated-single-channel, how to proceed normalization, with which package.

If not, how to normalize three-color-chip, but I think currently there is no suitable package can directly do that.

And I will be very grateful for all your answers and help.

Enjun 

I've never analyzed three color arrays as they never became popular. I think the way to do it though would be to generalize the idea of a separate channel analysis, something like this. All steps use the limma package.

1. Use read.maimages to read each channel separately from each gpr file, then cbind the expression values back into an EListRaw object.

2. backgroundCorrect(method="normexp", offset=16)

3. Run cyclic loess normalization separately for each physical array.

4. Either Amean or quantile normalization between arrays. Amean would be slightly better, but it needs some special coding in this case, so quantile would be easier.

5. Filter out unexpressed probes.

6. Use duplicateCorrelation to estimate the intra-array correlation.

7. Run a regular linear model and DE analysis treating each physical array as a block.