Hi all. First things first, sorry for posting one more question about experimental design and time series in DESeq2.
We have performed RNA-seq with two different treatments (control and protein over-expression) in two different time points (t=0 and t=8h).
These two different time points correspond to a treatment of the cells with a transcriptional inhibitor. t=0 corresponds to the steady-state of the cells and in t=8 we would expect a general decay of RNA abundances (since transcription has been blocked). I attach the design table that we are using.
| Sample | Treatment | time |
|---|---|---|
| sample1 | control | t0 |
| sample 2 | control | t8 |
| sample 3 | overExp | t0 |
| sample 4 | overExp | t8 |
... (we have 4 replicates, so the above table x4)
Our hypothesis is that the RNA changes are different in control cells and cells over-expressing our protein of interest, since our protein of interest have been linked to RNA stability. We would like to know the different response of the control cells vs the ones with the over-expressed protein at t=8h.
My naive approach so far has been to treat independently control cells and cell over-expressing our protein of interest, and then compare the log2FoldChanges obtained independently. I was wondering whether there would be any way of getting this comparison directly with the design formula of DESeq2.
However, I have doubts regarding two different questions.
First, I don't know whether this is a real time-series experiment and should use the design: ~ treatment + time + treatment:time
Problem number two: since cells were treated with a transcriptional inhibitor, we shouldn't observe any gene increasing transcription. Should we discard genes that show increased RNA levels?
If the time series is the most suitable approach for this analysis, I have used the aforementioned formula, and the treatment~time reduced formula. Would this design allow us to differentiate the different log2FoldChanges for the two different conditions?
Thanks before hand for your help,
Jordi
Hi Michael, I've been reading and trying to understand about the interaction term, however, I'm not entirely sure that I completely understand it. As I stated in my question, I want to see the difference in the response of the treatment between normal cells and and cells over-expressing my protein of interest. Technically, with the design formula that I have written above, we should see this difference in response. To extract this results from the DESeq2 call, I have used then the following code:
results(dds, contrast = list(c("time_8h_vs_0h","treatment.time8h")))Is this the actual way of doing so? I have also looked at log2FoldChanges provided in the output. The sign of the log2FoldChange corresponds to how the genes change in t=8 vs t=0, however, I don't see where the interaction is included in the log2FoldChange. Could you please shed some light for me to better understand the analysis?
Thank you and sorry to bother again
This isn’t correct if you want to see the difference in the treatment effect. I’d strongly recommend to collaborate with a statistician or someone familiar with linear models in R. I can provide software support here but I unfortunately don’t have sufficient time to provide statistics consultation on analysis plans.
What would the contrast be then?
Would it be
results(dds, contrast = list(c("treatment_control_vs_overExp","treatment.time8h")))Thanks again