I use Limma to determine differential expression between classes. My question concerns how to find class specific coefficients of variation (CV) for one or all probes.
Three questions:
1) can i extract CV values "automatically" when using the Limma package to find differential expression?
2) if not, at what stage would I be able to use the standard deviations and mean expression/intensity to calculate CV?
3) wouldnt i want to calculate CV after applying the eBayes method to shrink the probe-wise sample variances?
Best,
Nils Rudqvist
Postdoctoral Research Scientist
Center for Radiological Research
Columbia University
It isn't appropriate to compute CV as part of a limma analysis.
CV is only appropriate for strictly positive variables (like raw expression values). limma works with log-expression values for which variances and standard deviations are more appropriate. The standard deviation of log-expression is in fact approximately the same as the CV of unlogged expression (except for a scaling factor because limma uses log2).
Your next question might be, can you get class-specific standard deviations from limma, but limma doesn't automatically compute that. You could get class specific SD by running limma on the classes separately, but why would you want that?
Thank you Gordon for both answering me so quickly and for making Limma so excellent to use. I know that many people out there - including me - rely on and is thankful for your work.
Let us say I have two groups of five animals each: group A (irradiated) and B (non-irradiated controls). To visualize/compare the distribution of samples within each group we discussed calculating CV values (or SD as you propose) for each probe (for group A and B respectively). For this, I must need class specific SD for each probe? My idea now is to plot the variance or SD probe-wise for group A vs group B.
I do understand that I might attack this using the wrong approach.
