DESeq2 multifactorial design

[Erinyes]

Hi all.

I have a complex experiment that I try to analyse, and I am not sure how to go about it. I have performed RNA-seq experiments in my cells under two different conditions, starved and stimulated, and I wanted to see how gene expression changed. I repeated the experiment after knocking-down a transcription factor that might influence the response to starvation-stimulation. So, I designed shRNA for this factor, and I performed RNA-seq experiments again under starvation and after stimulation. On top of that, I prepared libraries of cells treated with scramble RNA under and after stimulation. All experiments were performed 3 times and analysed with DESeq2.

Each repeat of this experiment was prepared with each own scramble library. I noticed that I needed to "normalise" each treated sample to its own control. I did this by including a 'prep' group, grouping together all libraries prepared on the same day and sequenced together.

I analysed the data using this DESeq2 design:

Sample Treatment Prep Condition
shRNA-1 treated A Starved
shRNA-2 treated B Starved
shRNA-3 treated C Starved
scr-1 control A Starved
scr-2 control B Starved
scr-3 control C Starved
shRNA-1 treated A Stimulated
shRNA-2 treated B Stimulated
shRNA-3 treated C Stimulated
scr-1 control A Stimulated
scr-2 control B Stimulated
scr-3 control C Stimulated

design= ~ Prep + Treatment

This was done per condition (i.e. one analysis for starved samples, one for stimulated samples). This analysis gave me a few interesting things, but I would like now to see how the expression levels change from starvation to stimulation for my shRNA-treated cells, after some kind of normalisation to the scramble samples. I tried using the following design:

design= ~ Prep + Treatment + Condition

but I am not sure that this makes sense. At least the gene list I am getting does not make much sense. Is there something wrong with using both Prep and Treatment as blocking factors?

Any suggestion whatsoever will be much appreciated.

[dpryan]

You probably want:

design = ~prep+Treatment*Condition

At least I assume that you did the experiment because the shRNA knockdown should affect the response to the condition change (stimulated vs. starved), in which case the interaction is pretty interesting.

[Erinyes]

I didn't try that because I thought that I would be making the assumption that there will be a interaction (and I am not sure whether there is one, although I hope!). I'll give it a try now and see.

Any thoughts about how I could also incorporate the wild type cells (starved and stimulated) into the analysis? I know that I am asking too much now.

Thank you for the reply though!

[dpryan]

The design I gave already incorporates that. If it helps, another way to write the exact same design is:

design = ~prep + Treatment + Condition + Treatment:Condition

"*" is just an abbreviation for "as main effects and with an interaction".

[Erinyes]

Ok, thanks, I'm trying it as we speak.

What I meant was that apart from the 'control' (i.e. cells treated with scramble RNA), I also have wild type cells with no treatment at all. I was wondering whether it makes any sense to include them into the analysis as well.

[dpryan]

Ah, I follow you know. Yes, it probably makes sense to include those, since you're probably more interested in control vs. starved and control vs. stimulated than starved vs. stimulated.