Interaction terms in DESEQ2 with 3 factors

Question

[denalitastic]

Hello, I have been trying to figure out how to deal with combining interaction terms in Deseq2. I have tried all posts and the tutorials including the manual but none try exactly what I am looking for.

My design is as follows

design = ~cell + drug + time + drug:time

Drug: V,L,T
Time: 0,1,2,6,12,24,48
cell: A ,B

I am interested in the difference of differences. Basically I am looking for what genes are regulated differently between 2 treatments or between 1 treatment vs the other 2 across time points as the condition effect. I would also like to interrogate the same question by pooling some time points. So for example for drug T i have the condition effect of early vs late (early being time points 1 and 6 combined) and late being (24 and 48 combined). And how does this compare to the case of drug L for the same condition efffect. All this while controlling for cell type. My solution is using the design above. Assume here that I have made L drug my reference.

this is from resultsNames

 [1] "Intercept"          "cell_A_vs_B" "drug_V_vs_L"     "drug_T_vs_L"   
 [5] "time_1_vs_0"        "time_6_vs_0"        "time_12_vs_0"     "time_24_vs_0  
 [9] "time_48_vs_0"       "drugV.time1"      "drugT.time1"     "drugV.time6"     
[13] "drugT.time6"     "drugV.time12"     "drugT.time12"    "drugV.time24"    
[17] "drugT.time24"    "drugV.time48"     "drugT.time48"

late vs early condition , T drug vs L drug (L is reference level)

contrast = list(c("drugT.time48", "drugT.time24"), c("drugT.time1","drugT.time6")), listValues = c(0.5,-0.5)

For 48hrs vs early for V vs L (L is reference)

contrast = list(c("drugV.time48"), c("drugV.time1","drugV.time6")), listValues = c(1,-0.5)

What if i wanted condition effect of Late vs Early and the difference and T & V combined vs L? My guess is this. (where L is reference)

c("drugT.time48", "drugT.time24", "drugV.time48", "drugV.time24"), c("drugT.time1", "drugT.time6", "drugV.time1", "drugV.time6"), listValues = c(0.25,-0.25)

Is this correct? Are there other methods of doing this. What if I made V my reference and asked the same questions. This seems more complicating but it might help explicate the DESeq syntax. I can add anything to this post that is necessary.

Thanks for any help you can provide!

[GenoMax]

Don't use [CODE] tags. Instead format code in your post by highlighting the relevant text and then clicking "1010010" button in the editor.

Answer 1

[dariober]

I didn't go through your design so I'm not exactly answering your question, sorry, but I have a couple of suggestions.

• Have a look at the edgeR manual section 3.3.1 (Defining each treatment combination as a group). There is a very nice general solution for dealing with more complex designs and contrasts. It might be directly applicable to DESeq, not sure though.

• I would suggest you add to your count matrix some "control" mock genes that behave like positive and negative controls. In other words, just add a fake gene that you expect to be picked up by your design as differentially expressed and a fake gene that is not expected to be detected as DE. This will give you some reassurance that the design is correct. I found this strategy quite useful and I try to use it quite often if I can. It is a sort of equivalent of unit tests and test driven development frequently used in software engineering.