Hi Michael or DESEQ2 community,

First, sorry, I'm not sure why the images don't load, but if you right click on them they will.

When I use the DESEQ2 results() function I get a variable called stat, which is useful for GSEA.

When I run lfcshrink on the DESEQ2 dataset the stat column isn't there. It also isn't there in the screen shot in the vignette

When I use the code from this post to generate an example dataset and run lfcshrink on it the stat column is there: dds <- makeExampleDESeqDataSet(betaSD=1) dds <- DESeq(dds, betaPrior = FALSE) res <- results(dds) res.shr <- lfcShrink(dds=dds, coef=2, res=res)

What determines if the stat column is present in the results of lfcshrink or not and how to I make sure it is present? I'm interested in this because, from what I have read, the stat column is useful for GSEA analysis (fGSEA package) and from what I have found and is posted in the vignette I get the impression it is best to use shrinkage for downstream analysis. Thanks for your help!

Cory

Log fold change shrinkage is for posterior estimation of the effect size. You get the posterior mean or mode and the posterior SD (although it is labelled lfcSE to not break downstream workflows, its metadata description is "posterior SD").

If you run with type="normal" (which now generates a note to switch to another shrinkage type), you will get a stat column for backward compatibility.

If you run type="apeglm" or "ashr" you do not get a stat column because these don't really make sense in the context of the posterior quantities that are returned.

For GSEA, you could use the LFC in my opinion. The reason why many downstream software recommend using Wald statistics is for shrinkage of noisy LFC estimates, but we have taken care of that directly in DESeq2.