Dear Mete,

No, you cannot use non-integer counts with edgeR.

If you must use non-integer counts, please use the voom() function in the limma package instead. This will do an analysis that is not too different from edgeR, just a little less powerful, and is not bothered by non-integer values.

Best wishes
Gordon

Edit: more recent versions of edgeR do now allow non-integer counts. However the values must be still be on the same scale as the counts, e.g., the values should add up approximately to the correct library sizes. So it is still not suitable for your weighted counts.

> Date: Mon, 21 May 2012 12:24:43 -0700
> From: Mete Civelek <mcivelek@mednet.ucla.edu>
> To: <bioconductor@r-project.org>
> Subject: [BioC] non-integer counts for edgeR
>
> Dear All,
>
> I am analyzing microRNAseq data with edgeR. Because some of the reads map
> to multiple locations, I weighed the counts based on the number of genomic
> locations that a read maps. For example, if a read maps to 3 locations,
> each location gets a count of 1/3. Of course, this means that the read counts
> for some miRNAs in some samples are not integers. Is it possible to use these
> counts to do differential expression analysis with a quantitative trait in
> edgeR? My understanding is "no" but I thought someone can suggest a
> solution.
>
> Thank you for your help.
>
> Best Regards,
> Mete
 

Hi Dr. Smyth,

Will you be writing / have you already written a paper summarizing what voom() does and how?

My understanding of voom() comes from

http://statsandgenomes.wordpress.com/2011/11/23/bioinformatics-seminar-professor-gordon-smyth-variance-models-for-rna-seq/

wherein it appears to be the current best general-purpose tool for treating RNAseq data as if it were microarray data. This would seem to bring up issues with rare transcripts and the estimation of their abundance, which is why I ask about a writeup.

Thanks as always for your patient explanations and high standards of work.

--t