[BioC] Using edgeR with no replicates, estimating dispersion and the pseudo.alt slot

Gordon K Smyth smyth at wehi.EDU.AU
Sat Oct 27 01:30:58 CEST 2012

Dear Mick,

When you choose your own value for the dispersion, the intention is that 
you simply input the pre-determined value into exactTest(), like so:

   y <- matrix(rnbinom(40,size=1/0.4,mu=10),nrow=20,ncol=2)
   d <- DGEList(counts=y,group=1:2)
   et <- exactTest(d,dispersion=0.4)

There is no need for the pseudo.counts to be computed.  And you can't 
possibly expect estimateTagwiseDisp() to run, because it needs replicates. 
If there isn't enough data to estimate a common dispersion, how could 
there be enough data to estimate tagwise dispersions?

If you want to compute the pseudo.counts for another purpose, then the 
function equalizeLibSizes() is provided:

   d2 <- equalizeLibSizes(d,dispersion=0.4)

But be aware that the edgeR authors have not recommended the pseudo.counts 
as a measure of normalized expression.

Finally, let me mention that the 'pseudo.alt' component has been renamed 
'pseudo.counts' in the latest release of edgeR.  It is generally advisable 
to install and work with the current release version of edgeR and 

Best wishes

> Date: Thu, 25 Oct 2012 16:37:48 +0100
> From: WATSON Mick <mick.watson at roslin.ed.ac.uk>
> To: "'bioconductor at r-project.org'" <bioconductor at r-project.org>
> Subject: [BioC] Using edgeR with no replicates, estimating dispersion
> 	and the pseudo.alt slot
> Hi
> We have a collaborator with RNA-Seq data with no replicates - and yes, I 
> understand the limitations of this.
> In the egdeR manual, one solution suggested to estimating common 
> dispersion is choosing a value based on experience e.g. 
> "d$common.dispersion <- 0.4".  However, this has the disadvantage of not 
> filling up the pseudo.alt slot of the object, which are useful measures 
> of normalised expression.  The estimateTagWiseDisp() function then 
> fails.
> It's fairly trivial to edit the estimateCommonDisp() function to accept 
> a user provided common dispersion value, but I am not sure this is 
> entirely valid.
> The question being how to populate the pseudo.alt slot when you are 
> estimating the common dispersion parameter yourself, and whether it is 
> even valid to do so.
> Cheers
> Mick
> -- 
> The University of Edinburgh is a charitable body, registered in
> Scotland, with registration number SC005336.

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

More information about the Bioconductor mailing list