DESeq2 multiple contrasts no "control" condition

[aprilw]

I have an odd experimental design in that I have 4 conditions A, B, C, and D (all with replicates). However there is no "control" condition to compare to.

What I'm interested in is what genes are enriched in B compared to A, C, & D...ect. Is the best way to do this just using the "contrast" argument from results and doing all the comparisons pairwise, and then looking at the overlap of significant results? I'm a bit stumped as there is no reference condition.

[Skiaphrene]

Hi AprilW,


I'm not entirely sure but I think that Michael I. Love's response to my question (http://seqanswers.com/forums/showthread.php?t=58040) might help you too. If you check out section 3.3 from the latest version of DESeq2's vignette (http://bioconductor.org/packages/rel...doc/DESeq2.pdf), it mentions the use of "expanded model matrices" which do not use the concept of reference levels. With this (and the examples provided) you should then be able to write out contrasts to find DEGs that are specific to each condition. I'm still getting used to DESeq2 contrasts so I can't help you more at the moment!


Hope this helps,

-- Alex

[Michael Love]

It's fine to run DESeq() without a control condition.

Just use a design of ~ condition, and then you can make pairwise comparisons with 'contrast' argument of results().

"what genes are enriched in B compared to A, C, & D"

There are multiple ways to do this, which will give different results.

One is to do pairwise comparisons and combine the results somehow (intersection, etc).

Another approach is to compare B to the average of A,C,D. Note that this is not the same. For a very simple example, consider 5 vs 3,4,8. It is different than all three, but equal to their mean.

The average approach can be formulated like this:

results(dds, contrast=list("conditionB",
    c("conditionA","conditionC","conditionD")), 
  listValues=c(1, -1/3))