Dear DESeq2 developer and colleagues,

DESeq2 generates an output table including "log2FoldChange". I have two questions about this value.

I. As far as I know, this is a moderated version, not a log2 fold change from the raw count value. But I want to have unmoderated value that can be calculated from count directly and my current way is: (1) get normalized count (2) log transformation (3) calculate averages per condition (4) divide two averages: "FC". Is this a right way? I am confusing whether it is fold change of log, or log of fold change.

II. Effect size is "log2FoldChange" divided by SD where SD is sqrt(1/mu + dispersion). Here, I think that mu is an average of "log" of normalized count and dispersion can be retrieved by a command "dispersions()". I assume that this command outputs in a same order with "results()" to match values between two output objects. Also, I can use moderated version or unmoderated version with the same formula, mu and dispersion. Are they correct ?

1) get normalized count (2) log transformation (3) calculate averages per condition (4) divide two averages: "FC". Is this a right way?

No. Take the averages of the normalized counts, divide one by the other, take the log base 2 of that.

DESeq() since version 1.15 (~2.5 years) has not performed shrinkage, but this was moved to lfcShrink(), see the current vignette.

The standard deviation of the LFC is not equal to sqrt(1/mu + dispersion). See the DESeq2 paper for details.