[BioC] FW: DESeq Design (val1 vs mean (val2, val3, val4))

[Simon Anders]

> I want to find the DE for miRNAs. I have a factor type and  multiple values
> for that factor.  I want to calculate the DE in a design in which I can
> compare one value to the mean of the other values. It is possible in edgeR
> if you give a coefficient to your group but I don't know how to simulate it
> in DESeq.

To compare one factor level against the average of all other factor 
levels, the easiest might be to redefine the factor to have only two 
levels (the level of interest and "other"), and test for this. As you 
would then have a higher dispersion in the "other" level, you may also 
want to use the "per-condition" dispersion method.

For DESeq2, you could construct a design matrix with "sum" contrasts, 
using your level of interest as the last one, and this should also give 
the desired test.

However, are you sure you must compare to the average of the other 
conditions, rather than compare to a control condition? After all, even 
if only one condition is very different to the others, you may find not 
only this condition to differ significantly from the averages of the 
others, but also all other conditions, too, because the outlying 
condition changes the average you compare against.

   S