DESeq2 : foldchange calculation with multi-factor design

Question

[Nicolas Rosewick]

Hi,

I've a question concerning DESeq2. I've a multi-factor design as follow :

samples      condition      type
Sample_A1    untreated      X
Sample_A2    untreated      X
Sample_A3    untreated      Y
Sample_A4    untreated      Y
Sample_B1    treated        X
Sample_B2    treated        X
Sample_B3    treated        Y
Sample_B4    treated        Y

so I do

dds <- DESeqDataSetFromHTSeqCount(sampleTable=samples,directory="./",design= ~ type+condition)
dds <- DESeq(dds)
res <- results(dds)

In the results, I wonder how the log2foldchange is computed ? How it takes into account this two-factor design ?

Thanks

Answer 1

[Michael Love]

The two factor design is taken into account as with normal linear regression: we have a term which represents the changes due to type, and a term which represents the change due to condition. Then when you call results(), it is by default testing the condition treated vs untreated effect. But you could also test the type Y vs X effect (more on this in the vignette).

[Nicolas Rosewick]

So in my code the foldchange in the results will only take the factor "condition" into account?

[Devon Ryan]

It's "log2 fold change of the condition while controlling for type".

[Michael Love]

No. It is testing the condition term, but the design includes type, so the type effect is "taken into account" or "controlled for" in the model. You might want to check a reference on linear models for further reference on how this work.