questions about DESeq2

Question

[tony.fox2016]

Hi,

I have a couple of questions about DESeq2:

1. in the DESeq2 workflow, to test the dex treatment effect, the following command was used:

   dds <- DESeqDataSet(se, design = ~ cell + dex)
   

   And I guess the only factor here is "dex" and it might be more straightforward to use:

   dds <- DESeqDataSet(se, design = ~ dex)
   

2. If I am performing an ANOVA-like comparison of 3 groups A, B and C:

   design(dds) = ~ group
   dds = DESeq(dds, test = "LRT", reduced = ~ 1)
   results(dds)
   

To extract the DE genes between B-vs-A and C-vs-B. We should use the following commands and no post-hoc tests are needed. Is it correct?

res.B.vs.A <-results(dds, contrast=c("group","A","B"))
res.C.vs.B <-results(dds, contrast=c("group","B","C"))

Many thanks for Michael Love and colleagues for contributing such an excellent software and their answers in the forum!

[tony.fox2016]

Thank you very much for the answer. It is very helpful. I will post it in bioconductor forum and hope Michael could see it.

Answer 1

[mbio.kyle]

I'm by no means a DESeq2 expert, but I have been using in my last few projects. In regards to your first question I suppose it would depend on your experimental setup / matrix. If you truly have just one 'treatment' then I believe your design is correct. One way to verify this is to use:

resultsNames(dds)

Which will show you which contrasts have been formed.

As for part two, the results() function will return every single gene, so you will likely want to wrap the call with a call to subset, something like:

subset(results(dds, contrast=c("group", "A", "C"), padj < 0.05)

One final note, I have had much more luck posting questions like these (specific to bioconductor packages) to the bioconductor forum.