Design for a two-gene mechanism in DESeq2

Question

[mev]

Hi, I have a question regarding the DESeq2 design of an analysis I've been wanting to try.

Say in our experiment we have 4 conditions:

-control

-sample with transfected gene A

-sample with transfected gene B

-sample with both transfected genes A + B together

The assumption is that when the 2 genes are expressed in parallel they activate a unique mechanism, which is not normally activated when each gene is expressed individually.

Thus, if the research aim is to find the differential gene expression induced by ONLY the combination of transfected genes A + B without including differentially expressed genes induced by either gene A or B individually, what would be the best way to go about it?

The simplest and most obvious way to me would be to do the following contrasts:

-transfected gene A vs control

-transfected gene B vs control

-transfected genes A + B vs control

Then I keep only the differential genes in the "transfected genes A + B vs control" contrast that are not also found in the results of the other two contrasts.

The issue is that I would be compromising the FDR control of the DESeq2 gene results by arbitrarily removing genes from the list because they are found in another list. Is there another way of doing this analysis with a different design in DESeq2 or is this the only way?

Thank you in advance.

[GenoMax]

Do you have biological replicates for all conditions?

[mev]

Yes, 3 each. Not ideal but it's all that's available.

Answer 1

[e.rempel]

Hi,

rather a comment than a full answer.

I think, you could use an interaction term in your design formula. Something like

design(dds) <- ~ geneA + geneB + geneA:geneB

Also, have a look here

[Devon Ryan]

Moved to an answer because this really is the appropriate approach :)

[mev]

Hi,

Thank you for the reply and to Devon for making it an answer. Still new to this and will be looking into how interaction terms work. But I'm not really sure how I would make that design, right now the design matrix has 1 condition column, containing the 4 sample groups mentioned in the OP, as shown below:

#condition
control
geneA
geneB
geneAB
#(each of the above rows is repeated thrice in the matrix for each of the replicate samples)

So I don't have multiple factors to input to the design formula as you showed, only factor levels, unless I'm misunderstanding something about how the design formula works. Thanks again.

[e.rempel]

Hi,

it is possible to specify your design with one condition column, but it is more meaningful to use two conditions 'gene A' and 'gene B' (or factors in the lingo of DESeq) with two factor levels each ('transfected' or 'not transfected'). Then it is possible to use the design formula in my original answer.

[mev]

Hi,

I see, that makes sense. Thank you for all your help with this, I'll go read up on doing contrasts with interaction terms then.