Hey Tibor, It is probably a good idea to convert first. ComBat handles different variance across batches, but if you have different variances across samples within the same batch, then you should probably account for this before using ComBat. Evan On Apr 5, 2013, at 6:00 AM, <bioconductor-request at="" r-project.org=""> <bioconductor-request at="" r-project.org=""> wrote: > Message: 6 > Date: Thu, 4 Apr 2013 07:31:16 -0700 (PDT) > From: "Tibor Pakozdi [guest]" <guest at="" bioconductor.org=""> > To: bioconductor at r-project.org, tibor.pakozdi at embl.de > Cc: sva Maintainer <jleek at="" jhsph.edu=""> > Subject: [BioC] Homoscedasticity > Message-ID: <20130404143116.8270113238B at mamba.fhcrc.org> > > > Hello, > > Since ComBat was originally developed for microarray expression data where homoscedasticity was assumed, do we first need to convert heteroscedastic data such as the ones from NGS experiments into homoscedastic (e.g. by DESeq's VST) or the method can deal with samples of unequal variance? > > Thanks. > > -- output of sessionInfo(): > > # > > -- > Sent via the guest posting facility at bioconductor.org.