I've read several posts about applying interactions to an RNA-Seq design where there are continuous or categorical variables (such as time trials).
But I a slightly different issue that I'm struggling to find a solution to.
I have 2 treatment groups: treatment + control. The treatment group has 8 samples with a gradient of chemical exposures (1ul to 8ul exposure). If the covariate (1ul - 8ul) is applied to both treatment and control I could maybe use design = ~ group + covariate + group:covariate.
In a DESeq2 design, how can I apply the covariate (categorical variable) only to the treatment group and not the control?
Or is there a better way to manage the data - fit the data using a binomial GLM model (and any suggestion as to the best way to do this would be wonderful)? Is there another RNA-Seq package that can handle categorical variables in one comparison group?
Thanks in advance.
I have 8 samples for each condition. 8 controls (with 0ul treatment) and then 1 'treatment' group of 8 samples- 1ul, 2ul, 3ul, 4ul, 5ul, 6ul, 7ul, 8ul samples. In terms of DE analysis with DESeq2, can I even compare the control group with the treatment group or is the treatment group effectively 8 separate treatment groups of 1 sample?
Having a single replicate per level of ul means that you need to make some modeling choices about how you think expression will increase with ul of treatment. I'd recommend using a single variable "dose" which goes from 0 to 8 ul, and then consulting with a local statistician who can help you pick an appropriate smooth function (linear, quadratic, splines, etc) for modeling. Note that what you put in the design is the relationship for the log counts. So if you say ~dose (where dose is in ul), this means you think that for each 1 ul increase in dose you should see a constant multiplicative change in gene expression.