Question about the DESeq2: log2foldchange contradictory

Question

[tiago211287]

A doubt that came to me when playing with the data made me very nervous about it, follow me:

Looking into the DESeq2 manual you can see:

it is important that "control" or "untreated" level as the first element ("reference level"), so that the log2 fold changes produced by default will be the expected comparison against the reference level, that is log2 (treated/untreated).

I checked and my Deseq2 object is fine, with untreated/control as the first element

So, when Experimental group has lower reads than Control, you expect negative log2foldchanges. And when experimental is bigger than Control you expect positive log2foldchanges. So I think it is fine to separate(TRUE/FALSE) them in my dataset like this:

Cresultado$up <- Cresultado$log2FoldChange > 1 & Cresultado$padj < 0.01
Cresultado$down <- Cresultado$log2FoldChange > -1 & Cresultado$padj < 0.01

The problem is, when looking in the raw read counts, every gene maked as upregulated has experimental group with lower reads than control and vice-versa. like this:

gene_id                                       log2foldchange
ENSMUSG00000068606    -9.316849            Downregulated

But is pretty up-regulated below

                                      Control_1         Experimental
ENSMUSG00000068606      15               18145
Control_2
ENSMUSG00000068606      19               15001
Control_3
ENSMUSG00000068606      9                 18767

The same occurs with other IDs

Can someone enlighten me?

Answer 1

[tiago211287]

Well guys, looking in the DESeq2 Manual, I found that when contrasting your pair of comparissions you need to do like:

results(dds, contrast=c("condition","Experimental","Control")) so DESeq2 will do log2(experimental/Control)

and I was doing the just the opposite like in

results(dds, contrast=c("condition","Control","Experimental")) and was seeing everything inverted

My results just got mirrored. Oh my god.

[Devon Ryan]

The part of the manual you were reading before only applies to when one simply extracts the coefficient, rather than making a contrast.