[BioC] interaction effects for DEXSeq

[Simon Anders]

On 06/06/13 16:22, Mallon, Eamonn B. (Dr.) wrote:
> Dear all
> I have 11 samples for a cross infection experiment where there are two colonies (the hosts K or Q) and 2 strains (6 or 8).
>
> sample colony strain type
> K61 k six single-read
> K62 k six single-read
> K63 k six single-read
> K81 k eight single-read
> K82 k eight single-read
> K83 k eight single-read
> Q61 q six single-read
> Q62 q six single-read
> Q63 q six single-read
> Q81 q eight single-read
> Q82 q eight single-read
>
> I am most interested in finding exon usage differences related to the interaction of the colony and strain factors. Following the vignette I put the following code together
>
>
> formuladispersion<-count~sample+(colony:strain)+exon
> ecs<-estimateDispersions(ecs, formula=formuladispersion)
> ecs<-fitDispersionFunction(ecs)
>
> formula0<-count~sample+exon+(colony:strain)
> formula1<-count~sample+exon+(colony:strain)*I(exon==exonID)
> ecs<-testForDEU(ecs, formula0=formula0, formula1=formula1)
>
> Does this make sense?

Not quite. This tests for main effects and interaction in one go, so it 
returns as hits all exons whose usage differs between colonies _and/or_ 
between strains. You are looking for interactions, i.e., for exons whose 
usage is different between strains _and_ where this difference itself 
differs between colonies. So, you need

   formuladispersion<-count~sample+(colony*strain)+exon

   formula0<-count~sample+exon+(colony+strain)*exon
 
formula1<-count~sample+exon+(colony+strain)*exon+(colony:strain)*I(exon==exonID)

Things look a bit simpler if you use the new "TRT" functions:

   formuladispersion<-count~sample+(colony*strain)+exon

   formula0<-count~sample+exon+(colony+strain)*exon
   formula1<-count~sample+exon+(colony*strain)*exon

   Simon