[BioC] DESeq2 vs limma/voom: PCA completely different. How to proceed?

[Simon Anders]

Hi

On 23/04/13 19:10, Stephen Turner wrote:
> Short story: I ran a PCA on a matrix of counts from an RNA-seq
> experiment and then using the voom-tramsformed data, and they're
> completely different. I imagine my results will be very different
> using DESeq2 vs limma, and I am wondering how to proceed. Here are the
> two PCA biplots:

Steve has already answered the question, but just for the record: 
Running PCA on a count matrix is never a good idea.

PCA expects data to be homoscedastic and counts are not. There are 
various ways of transforming count data to a homoscedastic scale, and 
DESeq2's rlog and limma's voom are two of them.

   Simon