afer reading some posts about different DE analysis provided by edgeR, I found that the QL framework it's a the better choice, as it provides more accurate type I error rate control. However, when the dispersions are very large and the counts are very small it's better to switch to the LRT framework. But what do you mean for "very large" and "very small"? Are there reasonable thresholds?
For instance, I'm dealing with a dataset of 12 samples (1 variable/4 factors), and I'm not sure about the best approach to be adopted. Based on plotMDS, plotBVC, and plotQLDisp posted above, what's your feeling?
Thank you in advance for your time and suggestions! Best Marianna