Hi,

I have a simple case, design(dds) = ~ sex + batch + condition, where condition is a factor with three levels (A, B, and C). I tried two different ways of testing differential expressed genes between condition B and C:

Method 1: Run DEseq on the whole dataset, and extract comparison between B and C as results(dds, contrast = c("condition","B","C")). 

Method 2: Subsetting the dataset by extracting samples only with condition B and C, relevel the condition, then run DEseq on the subset, get result via results(dds_subset, contrast = c("condition","B","C"))

I expected the two methods have same results, but actually they don't. I got different number of DE genes with the same FDR cutoff. 

Anything I misunderstand? Please advise. 

Thanks,

-Xianjun

You should expect different results. If you fit a linear model and then make comparisons, the statistic you use will incorporate an estimate of variance based on the data in the model. If you change what data are used in the model, you should expect to get different statistics as well.