How To Normalization Multiple Groups At Once: Edger Or Deseq

Question

[Rm]

I have three groups (A, B, C) to compare each with many samples: My goal is to normalize all groups as a whole and then get differential expression (DE) list for each pair.

My attempts so far with edgeR:

if i define my groups [ group <- c(rep("A",68),rep("B",65),rep("C",70)) ] before calcNormFactors and common estimateCommonDisp and exporting normalized values. It looks like normalization is occurring within each group (which i am not looking for)

what I tried is define entire samples as one group [ group <- c(rep("A",203)) ] . I could get the normalized values: but my difficulty now it is not able to perform exactTest as group has only one entity.

After normalization I redefined the groups as in my first case still unable to perform the exactTest.

I really appreciate if you can suggest to over come this.... I am open to using other tools if they support this.

[Mikael Huss]

Can you clarify a bit what you mean by this sentence: "After normalization I redefined the groups as in my first case still unable to perform the exactTest."? Did you normalize considering everything as one group, and then do something like d$samples$group <- c(rep("A",68),rep("B",65),rep("C",70)) (assuming your DGEList object is called d)? And in what way were you unable to perform the exactTest? Was there an error message?

[Steve Lianoglou]

As Mikael's prompt for further information is alluding to, it would be much more helpful if you pasted in the code you used for this analysis -- it won't be much. You can then inline in comments (or break code into sections) and ask specific questions about parts of it. This would help us help you a lot better.

Also, what makes you say that "it looks like normalization is occurring within each group"

[Rm]

Thanks @Mikael; it worked after following your suggestion "d$samples$group" . Can you post it as answer i will accept it..

Answer 1

[Mikael Huss]

Did you normalize considering everything as one group, and then do something like d$samples$group <- c(rep("A",68),rep("B",65),rep("C",70))?

I suppose it would also be possible to normalize everything as one group (again) and then set up a design matrix and contrasts that would describe the comparison you are looking for.