Dear Michael,
I am having some doubts when trying to extract relevant info from my gene lists generated by LRT in DEseq2.
Now I have gene lists of differentially expressed genes explained by:
- the interaction of the 2 conditions that were applied to my cells,
- condition 1
- and condition 2
I have been trying to explore this data, by splitting the lists into Up and Down-reg gene lists, and entering these signif genes to GO in R, Panther and GOrilla, for example. Nevertheless, I have several levels per each condition, and I am not convinced that I am interpreting the results properly:
The reference condition in LRT is not the reference condition anymore, right? I am comparing the full design with the reduced design, not experimental conditions vs a reference condition. To do that I should do Wald, right? And from Wald, extract Up/Downs. But then I need to run many, many contrasts (more than 0).. and I don't feel this is how I should be doing this.
Is it correct to explore and visualize UP/DOWNs from the LRT gene lists? Or does it makes more sense to just say "these genes are affected by the interaction of condition1 and condition2 (LRT p.adj)", but then I have to run Wald to explain when are these genes up/down regulated?
I was also trying to plot my results in a heatmap where I could see the top diff. expressed genes for all the different conditions, but then is where I realize that the comparison is not against a reference anymore, but against the full model, so a heatmap would make no sense in this case. May I ask what would be the best way to visualize this result?
Thank you for your time, once again.
Laia
Hello James,
In my case, I have 3 timepoints that I want to compare timepoint 1 vs. timepoint 0 and timepoint 2 vs. timepoint 0.
I tried to extract the results from res_lrt and res_wald, and they are different.
Run DESeq2 with LRT
Run DEseq2 with Wald Test
You mentioned LRT test is an omnibus test, but I still can extract the results for one pair of comparison. I do not know which one is correct.
You need to pay attention to the output.
As I noted before, the LRT is an omnibus test where the p-values test that the model fits better with the term versus without. You can still extract logFC values for any comparison, but that is unrelated to the test. The Wald test is a direct test between two groups, and will differ depending on the groups you compare.