Hello! When performing differential expression analysis with RNA-Seq data using DeSeq2, and after data normalisation, there is the option to extract a count table with the transformed abundance levels using the variance stabilisation transformation (vst) method.

Line of code vst counts extraction: vst_dds <- vst(res)

The DeSeq2 manual itself indicates that these are the counts that are used to make clustering plots and heatmaps. Are these, therefore, the counts transformed to the negative binomial and with which the logFC, p-values and p-adj (FDR) are calculated (through a Wald test, for example)? In this count table, there are no longer values equal to zero (which allows you to calculate fold changes even if for a group (treatment or control) all the abundance values were, in the raw counts and after normalisation, equal to zero).

If what I have said is true, and with the DeSeq2 method I can extract the counts transformed to the negative binomial (equivalent to those transformed by the vst method), can I extract an equivalent table (with the transformed values) using the edgeR method? The edgeR manual allows you to extract the normalised table converted to logCPM but this is not the negative binomial transformed table, is it? Is it not possible, therefore, to extract a binomial transformed table from which to calculate (externally if desired) the FC and p-values? And not inside a cryptic function, as it seems to be.

Line of code logCMP counts extraction: logcpm <- cpm(y, log=TRUE)

Thank you very much

You have misunderstood what vst() does. There is actually no such thing as a "negative binomial transformed table", rather the only negative binomial quantities are the original counts. When people talk about "normalized counts" they actually mean quantities analogous to edgeR's logCPM values.

You might find these posts helpful:

• https://www.biostars.org/p/9475236/
• https://www.biostars.org/p/491767/