Negbinom vs. logistic regression for Seurat FindMarkers()

Question

[Lucy]

Hi,

I was wondering if anyone has thoughts on whether the negative binomial or logistic regression approach implemented in Seurat is preferable for comparing between clusters (including "donor" as a latent variable). From what I have seen the negative binomial approach tends to be more popular, however I have some data which suggests to me than the logistic regression is better at identifying marker genes.

Is there any statistical reason why the LR is not a good choice here?

Many thanks,

Lucy

[Lucy]

Anyone have any thoughts on this?