Hello all,
I am analyzing a RNAseq experiment constituted of 5 different groups of alcohol drinking mice: non drinkers, low drinkers, high-resistant, high-sensitive and water control. My ultimate goal is to identify genes changed specifically in one of the conditions (genes changed only in non drinkers, only in low drinkers, only in high-resistant and only in high-sensitive).
I've tried doing it in different ways and wanted your opinion about which one would be the best or most reasonable statistical approach.
1) DESe2 + LRT 2) DESeq Wald test of each one of the 11 pairwise combinations 3) DESeq Wald test comparing one group against all the others (i.e., Non vs (low +high resistant + high sensitive + water) and so forth - 4 independent tests)
Also, I wanted to discuss some questions about each one of the above methods:
1) When I run a LRT test, may I follow it up with a Tukey or a pairwise.t.test correcting it for multiple tests (groups)? I've found some people recommending proceeding with the Wald test using contrasts. However, the number of genes that I find with Wald compared to LRT drops a lot. Instead, is it ok to do Wald test only in the subset of DEG found by LRT?
2) With the second approach, what method do you recommend to identify the genes differentially expressed in each condition? I am inputting each one o pairwise comparisons in a Venn diagram and calculating the overlaps using Reduce() function, which is a bit laborious.
3) Is that a fare approach? I fell like I might be inflating my statistical power somehow when I do that. If that is indeed a valid approach, what should I do with genes that are found differentially expressed in two different combinations, for example Low vs all and Non vs all?
I know there are a few posts about how to deal with multiple groups in RNAseq experiment, but after going through every single one, I still am not totally confident about what I should do.
Thank you so much! Daniel
