3 group differential gene analysis

Question

[tryseq]

Hi guys, I can't seem to find a straightforward answer to this, so I apologize if it has been posted before!

I have RNAseq data for 3 different cell types, cell A, cell B, and cell C. Historically, I like to use DESeq2 for my differential gene analysis.

For differential gene analysis, I understand it as a pair-wise comparison. I can only compare A-B, A-C, and B-C. This works fine, so no issues there.

However, I want to know if there is a way to get differential gene data for these three groups in one go, A-B-C.

The analogy I am thinking of is instead of performing a bunch of T-tests, arrange the data in a way that I can use an ANOVA.

Does DESeq2, or a different package, have the power to compare 3 different groups in one operation? Or does it start to get way too complicated way too fast?

Answer 1

[ATpoint]

Use the LRT test of DESeq2, see the manual and e.g. https://support.bioconductor.org/p/73172/#78824

Edit: As mentioned below, I did not stress that this tests for any differences, it is not cell-specific. For specifics I would really do the pairwise comparisons with the standard Wald test.

Answer 2

[LauferVA]

It depends on the exact hypothesis you are trying to test. Do you mean:

H0: No cell type (A, B, or C) differs from any other.
H1: There is at least one difference between A,B, and C.

If this is indeed what you want to implement, then, as ATpoint has already indicated, the LRT will test this, and it is already implemented in DESeq2.

However, this (the LRT) will NOT give you information on which group(s) are driving the difference, it will only tell you that a difference exists. In other words, from there you'll still have to do an additional test to determine which groups were sufficiently different to have resulted in a statistically significant LRT if you want to know the exact nature of the difference detected ...

---

Alternatively, you could mean something like:

I want to text all pairwise differences between each 2 pairs of groups, then I want to know which two groups differ and what the direction of effect is.

This is after all fairly similar to the information an ANOVA gives you, too... For this, you can just implement a Wald test (as you have likely been doing so far), then change the reference group. To be clear, run the Wald test, with cell A to reference. Now you can see A vs. B and A vs. C. What about B vs. C? For B vs. C, now set cell B as the reference group and look at B vs. C.