[BioC] GLM common and trended dispersion EdgeR

Gordon K Smyth smyth at wehi.EDU.AU
Sat Jan 11 01:26:52 CET 2014

Dear Reema,

I am sorry that it is a little confusing.  Let me clarify.

First, we always recommend that you use the full pipeline including 
tagwise dispersions as summarized in Section 1.4 of the User's Guide. 
This is for any RNA-seq experiment with replicates.

Second, the estimateTagwiseDisp() and estimateGLMTagwiseDisp() functions 
always assume a dispersion trend unless you explicitly tell them not to. 
This is so whether or not you have called estimateGLMTrendedDisp() prior 
to estimateGLMTagwiseDisp().

The reason for calling estimateGLMCommonDisp() and 
estimateGLMTrendedDisp() prior to calling estimateGLMTagwiseDisp() is 
partly for diagnostic purposes and partly to give estimateGLMTagwiseDisp() 
a general idea of the magnitude of the dispersions before it estimates the 
tagwise values.  However neither the common or trended dispersions are 
explicitly used in the tagwise estimation scheme; they are used only to 
setup the resolution of the tagwise estimation grid.

The reason that there is no estimateTrendedDisp() step in Section 1.4 is 
just because historically this function didn't exist.  Remember that we 
created the classic estimateTagwiseDisp() pipeline way back in 2006. 
Omitting the trended step makes only a small difference to the final 
tagwise results.

We are gradually revising all the edgeR code to make the classic and glm 
pipelines consistent with one another.  In the latest versions of edgeR, 
you can simply call:

   y <- estimateDisp(y)


   y <- estimateDisp(y,design)

and it will do the whole disp estimation pipeline (classic or glm) for you 
at once.

Best wishes

> Date: Thu, 9 Jan 2014 14:42:53 +0000
> From: Reema Singh <reema28sep at gmail.com>
> To: bioconductor <bioconductor at r-project.org>, Bioconductor mailing
> 	list	<bioconductor at stat.math.ethz.ch>
> Subject: [BioC] GLM common and trended dispersion EdgeR.
> Dear All,
> I have a query regarding using estimateGLMCommonDisp and
> estimateGLMTrendedDisp.
> It has been mentioned in the EdgeR manual[*Section 2.8.2 :- Estimating
> Dispersion*] "Note that we need to estimate either common dispersion or
> trended dispersions prior
> to the estimation of tagwise dispersions" and "If both exist, the default
> is to use the trended dispersions". Whereas in case study 4.4[*Section
> 4.4.4- Estimating the dispersion*] dispersion has been estimated using both
> common and Trended before computing the tagwise dispersion.
> Here my queries are:-
> 1) Is it fine to use both common and trended dispersion?
> 2) is there any specific conditions to use common or trended dispersion
> estimation?
> Kind Regards
> -- 
> Reema Singh
> Postdoctoral Research Assistant
> College of Life Sciences
> University of Dundee,
> Dundee DD1 4HN, Scotland
> United Kingdom
> http://www.compbio.dundee.ac.uk/people.html

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

More information about the Bioconductor mailing list