[BioC] GLM in EdgeR

[Gordon K Smyth]

> Date: Mon, 16 Jun 2014 09:39:22 +0000
> From: "Smith H." <Hazel.Smith at soton.ac.uk>
> To: "bioconductor at r-project.org" <bioconductor at r-project.org>
> Subject: [BioC] GLM in EdgeR
>
> Good morning all,
>
> I am just starting to analyse my first set of RNA-seq results and am 
> trying to use EdgeR to do this. I have spent time reading the vignette 
> and manual and working through it using our data. The design of our 
> experiment is that we have two treatments (drought and control), three 
> genotypes (clones of three trees) and each of the individual plants has 
> four leaves taken from it (each of which represents a different 
> developmental stage of the plant). There are between four and five 
> replicates for each genotype-treatment combination. The code we have 
> used so far is:
>
>> x <- read.delim("Poplar.counts.matrix",row.names="Symbol")
>> targets <- read.delim(file = "targets.txt", stringsAsFactors = FALSE)
>> Treatment <- factor(targets$Trt, levels=c("Con","Dr"))
>> Genotype <- factor(targets$Geno, levels=c("France","Italy","Spain"))
>> Leaf <- factor(targets$Lf, levels=c("Ap","L11","L7","Tag"))
>> data.frame(Treatment,Genotype,Leaf)
>> design <-model.matrix(~Genotype+Genotype:Treatment+Genotype:Leaf)
>
> Does this look correct? Using this model, we don't seem to be getting 
> the contrasts we expect (i.e. there is no "Ap" or "France") and I think 
> that this might be because EdgeR is using the first level of the factors 
> of Genotype and Leaf as references. Is this right?

Yes, that's right, although it is really R that is doing it rather than 
edgeR.  Every contrast has to compare something with something else, so 
something must be set as the reference.  One can't expect every level of 
each factor to appear as a coefficient in the model.

> Also is there a way to identify if there is an effect of genotype within 
> treatment (i.e. not just compare the genotypes in a pairwise fashion)?

Well, you have set up your model formula so as to compare treatments 
within genotype.  If you want genotype within treatment, just define your 
model the other way around.

Best wishes
Gordon

> Any help you could give us would be much appreciated.
>
> Many thanks,
>
> Hazel
>
>
> Research and Teaching Fellow
> Centre?for Biological Sciences
> University of Southampton
> United Kingdom
> Email: hazel.smith at soton.ac.uk

______________________________________________________________________
The information in this email is confidential and intend...{{dropped:4}}