Design Matrix For Pre-Subtracted Arrays

Question

[Dave Bridges]

I am using limma to look at a GSE microarray (GSE10918 to be specific). I am trying to figure out which design matrix to use before fitting the linear model and am having a hard time figuring it out. It seems from the phenoData, the GSM's are in this format (so they are already subtracted and log2 transformed):

gene1 - control
gene1 - control
gene1 - control
gene2 - control
gene2 - control
gene2 - control

What I want is to find the effects of gene1 and gene2 separately. I am wondering what the best design matrix is for this.

[Cheng Zhongshan]

Would do please explain it in more details? are you going to transform the raw dataset of GSE for R package limma?

[Cheng Zhongshan]

As descripted in the raw dataset of GSE10918 GSE10918_series_matrix.txt (ftp://ftp.ncbi.nih.gov/pub/geo/DATA/SeriesMatrix/GSE10918/), I think the value is log2(cy5/cy2).

[Cheng Zhongshan]

You can find some background infromation about microarray analysis here (http://soybeangenomics.cropsci.illinois.edu/files/NSF_Web_Microarrayresults.pdf).

Answer 1

[Jan Oosting]

You could read the raw data (the GPR files) to obtain separate values for the Cy3 and Cy5 channel.

However the experiment has a common reference design without dye swaps. Even after dye normalization you will not be able to distinguish dye effects from biological differences between renilla and the two genes studied.