Edger Differential Expression

Question

[Arun]

Hi,

I have RNA-Seq data from time point experiments: T0 and T1, each with 3 replicates for no-treatment (condition1) and with-treatment (condition2). That is, 3 replicates, each, under T0 and T1 for no-treatment and 3 replicates, each, under T0 and T1 for with-treatment.

Type/TimePoints     T0       T1
Control             3 Rep    3 Rep  
Treatment           3 Rep    3 Rep

Question: Is the ratio of gene expression between T0 and T1 in control (no-treatment) significantly different from ratio of gene expression between T0 and T1 in treatment? To rephrase, Is the differential expression I observe between T0 and T1 in treatment solely due to treatment (which can be concluded by taking into consideration the expression at T0 and T1 in control)?

Now, normally in edgeR, one could find out differential expression between control and treatment with replicates. Meaning there are no 2 different time points amongst this comparison. So is this possible using edgeR (or DESeq or BaySeq for that matter)?

Thank you.

[Jeremy Leipzig]

there is a section in the DESeq vignette for breaking down multi-factor designs. (probably one in edgeR as well)

[Arun]

thank you Jeremy, I'll check it out.

[Arun]

The pre tag works in preview mode but not after I post. What other way can I format the table?

[Neilfws]

Indent lines 4 spaces; done for you.

Answer 1

[Mikael Huss]

I'm pretty sure it is possible in edgeR, although I can't find a specific example in the vignette at the moment. It definitely is possible in the limma package, which is of course usually applied to microarray data, but there is a new method, voom(), in the limma package for preprocessing RNA-seq count data. The limma user guide contains a wealth of information about constructing different contrasts and "contrasts of contrasts" such as the one you are proposing. E g "Factorial designs", chapter 8.7 in http://bioconductor.org/packages/2.6/bioc/vignettes/limma/inst/doc/usersguide.pdf.

edgeR and limma share a lot of the underlying philosphy so the same is probably possible in edgeR.

[Arun]

Thank you Mikael. I prefer to use edgeR if possible as I already use their TMM normalization. I'll try to find out in the edgeR manual. I already have a working model I managed to code using MASS package. However, I'd love to try and compare the results from edgeR.