[BioC] GLM common and trended dispersion EdgeR.

Ryan C. Thompson rct at thompsonclan.org
Thu Jan 9 21:40:13 CET 2014

> 1) Is it fine to use both common and trended dispersion?

You can estimate both common and trended dispersions on the same 
DGEList object, but only one of them will be used. The other will be 
ignored. I believe that trended dispersion will always be preferred 
over common dispersion if both are available.

> 2) is there any specific conditions to use common or trended dispersion
> estimation?

Well, broadly speaking, if you believe that there is a dependency 
between mean and dispersion in your data, then you need trended 
dispersion. If not, you would use common dispersion. A number of 
publications have argued that when modeling RNA-seq data with the 
negative binomial distribution, such a dependency exists. For example, 
the (I think) original DESeq paper: 

So generally you would always want to use trended dispersion if you are 
analyzing RNA-seq. For other types of data, you would likely want to 
test that assumption.


More information about the Bioconductor mailing list