hi Jon, I reply and CC the Bioconductor list. On Thu, Apr 4, 2013 at 10:56 PM, Jon Lees <jonglees@gmail.com> wrote: > Dear Prof. Love > > Im using the deseq2 package, its giving us really nice results. > and seems to fix some issues over deseq2 with our data. > > I just had quick question, after I scale the counts by the scale factors, > and calculate the log2 fold change of case versus control, I get a > slightly different value from the log2 fold change I get by > running "results(dse)". Is there a simple answer for why there is a slight > difference here. > > The log2 fold change provided by default is now a maximum of the posterior, incorporating a prior distribution on the log2 fold changes (betas). The DESeq() and nbinomWaldTest() functions have an argument betaPrior, and details on the estimation of the prior is in the man page for nbinomWaldTest. This is also mentioned in Figure 1 of the vignette and Section A where we describe the changes since DESeq. The effect of the prior is to shrink log fold changes more or less depending on shape of the likelihood. Mike Thanks for advice/ info > > and Best wishes > > Dr Jonathan Lees > UCL > [[alternative HTML version deleted]]