[BioC] BaseCounts & edgeR

[Gordon K Smyth]

Dear Marco,

It is amazing that CASAVA is unable to return read counts.  If that is 
really true, then perhaps you should switch to a better mapping/count 
tool.  My lab uses subread, which is better, faster and available as a 
Bioconductor package.

You cannot use edgeR (or DESeq) to analyse base counts.  The negative 
binomial assumptions of those packages would be grossly violated. 
However, the limma-voom pipeline for RNA-Seq data would remain valid for 
base counts, and is highly competitive with the negative binomial 
approach.  limma-voom is optimized for the same sort of quadratic 
mean-variance relationship as the negative binomial approach, but without 
making assumptions about the form of the distribution or the absolute size 
of the counts.  See the last case study in the limma User's Guide.

Best wishes
Gordon

> Date: Tue, 31 Jan 2012 10:29:19 +0100
> From: Marco Groth <mgroth at fli-leibniz.de>
> To: bioconductor at r-project.org
> Subject: [BioC] BaseCounts & edgeR
>
> Dear Bioconductor team,
>
> I am using DESeq and edgeR for analysis of count data and find
> differentially expressed genes and btw it worked very well. As input
> data I created read count. For that I used also Illumina's CASAVA and
> GenomeStudio. Unfortunately, Illumia changed the counting procedure. As
> of version 1.8 read counting is not possible anymore, they changed
> finally to base counting. Means all mappable bases which fulfil a given
> quality score will be counted. So by using high quality 50bp reads for
> mapping the counts will be around 50x higher compared to the read count
> method.
> By using the base counts in edgeR the number of differentially expressed
> genes is dramatically reduced. We normally compare DESeq and edgeR
> results and they fit around 80%. Using the base counts the results do
> not fit anymore.
> I divided all base counts by 50 to approximate the read count and the
> results look better, but not as good as before. Furthermore, I get
> uncertainties because of counting in splice site is more sophisticated.
> Nevertheless, my question is whether I can run edgeR using base counts
> and getting good results?
>
> Thanks, Marco
>
> -- 
>
> Marco Groth

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