I'm running DESeq2 for my expression analysis and have a question on the design formula. I have 2 cell lines, 2 time points, and 4 treatments in 3 batches. I tried the following 2 designs to compare the same 2 treatments and the results are slightly different in terms of the number of significant genes and their ranks. Why is that? Are these two models not the same?

1. dds$group <- factor(paste0(dds$cell, ".", dds$time, ".", dds$treat))
design(dds)<- ~ group + batch
dds<-DESeq(dds)

2. design(dds)<- ~ batch + cell*time*treat
dds<-DESeq(dds)

Thanks for the help!

Best,

Wei

First: How have you called "results", and how was the comparison described in the first line of the output from "results"? This might already give you a hint what is going on.

Second: No, the two models are not the same. 

About the first model: By default, DESeq2 performs a Wald test, checking for significance of the last model coefficient. This would be a comparison between two specific levels of the concatenated factor; something like: differences between cell line 3, treatment 2, time point 3 and cell line 1, treatment 1, time point 1. You can pass extra arguments to "results" to select which levels you want to compare.

In the second model, the comparison is for the double interaction between the three factors, which might be rather hard to interpret.

It might be easier to advice you if you told us more about what comparison you actually want to do.