I am trying to find differentially expressed (DE) genes using EdgeR in a complicated experimental setting. There are two factors (each has different level of variation among replicates - the reason why the experimental setting is complicated to interpret) and each factor has a control and a treated group with two replicates for each.
I fit the GLM using a design matrix that describes my experimental setting. I get an unusual looking mean-variance plot (attached - mean-var-Plot.png). Here is my interpretation of the graph. The tagwise dispersion (light blue dots) is fit very well to the common dispersion (light blue line). However, since the raw variance (grey dots) is pretty far from the fitted tagwise dispersion, I am worried that the model might be overfitting my data and it could be bad for the analysis. Do you think this is the correct interpretation of this plot? I haven't seen this type of mean-variance plot before and therefore would like to know how would you interpret it.
Note: When I use each factor alone in the design matrix, the tagwise dispersion sits in the middle of the raw variance dots, which is how the mean-variance graphs typically look (attached -fac1.png, fac2.png).
Also, as I mentioned earlier that the level of variance (or dispersion) is different between my two factors (as the RNA levels were measured using different sequencing assays for both factors), I am wondering if EdgeR is the right method for these data, since EdgeR doesn't calculate the covariances separately for different factors.
I would highly appreciate any insights. Thanks in advance for your help.