Hello,

I'm using DESeq2 on a non-standard dataset where I wish to supply my own normalisation factors. I'm getting an error message from varianceStabilizingTransformation() which made me question the workflow I'm using:

# This works as expected, no warnings or errors
dds <- DESeqDataSetFromMatrix(countData = cnts, colData = smpls, design = ~ Batch + Condition_Treatment_Time)
normalizationFactors(dds) <- normfacs
dds <- DESeq(dds, test="Wald", parallel=TRUE)
using pre-existing normalization factors
estimating dispersions
gene-wise dispersion estimates: 16 workers
mean-dispersion relationship
final dispersion estimates, MLE betas: 16 workers
fitting model and testing: 16 workers

# However I get an error here
vsd <- varianceStabilizingTransformation(dds)
-- note: fitType='parametric', but the dispersion trend was not well captured by the
   function: y = a/x + b, and a local regression fit was automatically substituted.
   specify fitType='local' or 'mean' to avoid this message next time.
Error in getVarianceStabilizedData(object) :
  call estimateSizeFactors before calling getVarianceStabilizedData if using local dispersion fit

# same error using fitType="local"
> vsd <- vst(dds, fitType="local")
Error in getVarianceStabilizedData(object) :
  call estimateSizeFactors before calling getVarianceStabilizedData if using local dispersion fit


# but it works fine with fitType="mean"
vsd <- varianceStabilizingTransformation(dds, fitType="mean")

The error is about not having called estimateSizeFactors(), which I don't want to do because I supplied my own normalisation factors. fitType="mean" solves the issue but I'm wondering if I should also pass the same argument when building the dds object in the first place (even though I don't get any warnings or error at that stage).

I.e., change this:

dds <- DESeq(dds, test="Wald", parallel=TRUE)

to this:

dds <- DESeq(dds, test="Wald", fitType="mean", parallel=TRUE)

Thanks

Enrico

hi Enrico,

Thanks for reporting this. This is just a small bug from some old code I didn't update. It is just a test which should also succeed if there are normalization factors, but it's looking only for size factors. I can write a fix soon and push it to the release branch.