branch point analysis about slingshot HOT 4 CLOSED

Hi @roryk!

Thanks for the question! This is something that I have been meaning to add to the Slingshot vignette for a little while now.

Generally, my hope is that once you have done trajectory inference, you won't need to go back to the original clusters. That said, I think the simplest way to compare two branching lineages would be to take the two endpoint clusters and do some sort of differential expression test between them (which is different from what you described, so apologies if I'm not understanding something!)

This approach has a few drawbacks, however. The two biggest (in my mind), are that it only looks for DE at the endpoints, not along the full lineage, and that it only uses a small subset of the data with basically no information gained from trajectory inference. So my actual recommendation is to check out our new package: tradeSeq. Specifically, I think the patternTest is a good way to identify genes that have different expression patterns between two lineages (although the earlyDETest may be a little bit closer to what you were describing).

from slingshot.

Thanks for the fast response! For looking at the branch points, we were more looking for genes that might be involved in the choosing of the lineage, not genes that mark the ends. For example, testing at the branch point might let us detect a hypothetical receptor that expressed at the branch point, and that is doing the sensing to send it down either one of the trajectories, but the receptor isn't actually expressed at either end state.

I'll definitely look at your tradeSeq package, that is exactly what I need to do for the other part of this project.

from slingshot.

Oh ok, I think I understand a little bit better now. It might also make sense in that case to test the branch-point cluster vs. previous clusters in the lineage (rather than all other clusters), since you want genes that come on near the bifurcation, but don't necessarily care about when and how quickly they get turned off?

from slingshot.

Hi Kelly, that is a super good idea, I'll give that test a go as well.

from slingshot.