Compare expression vs. FindMarkers

[dagarfield]

Hello,

Is there any guidance you can provide regarding the differences between sctransform::compare_expression, as described in your sctransform tutorial, and Seurat::FindMarkers (whose more general framework includes a LRT and t-test)?

I'm asking this in the context of trying to work through a badly unbalanced dataset with some pretty wicked batch effects. For clustering, an initial treatment with SCTransform (regressing on batch or including 'batch' in batch_var) followed by fastMNN seems to do nicely (Seurat's own CCA seems to over integrate in this case). However, I, like many am struggling with DE analysis following SCTransform and batch correction.

As far as I can tell, going through Seurat::SCTransform typically leaves us with only log_umi corrected data with batch-like variables to be dealt with using latent variables at the DE step**, while compare_expression seems to be more general (if requiring the complications of working with the vst output)?

--DG

**the exception being batch_var, which seems to affect the final count, but in sort of hard to understand ways and that may not work so well with, say, cell-cycle.