Excessive 'follicular dendritic cell' annotations

[luxeredias]

Hi,

I've been using scimilarity to annotate cells in multiple different datasets. As a sanity check, I ran scimilarity on cells for which I already had cell type annotations performed with another method. For the majority of cells, the correspondence is very good, even if cell subtypes/classes might differ. However, I've been finding a consistently high number of follicular dendritic cells, even when working with non-lymphoid tissue such as skin and tissues from solid organs. Looking at the original annotations for these cells, there is a large proportion that are fibroblasts, but there are also others that are just stromal cells in the original annotation. When I look at the second best hit from scimilarity, I find stromal cell to be the most common annotation.

Is there any potential reason that explains this behavior and any solution for the problem?

[javh]

We'll have to look closer at the FDC training data to get a sense of what specifically may be happening with FDCs.

But, for an immediate fix you could use either the cell_annotation.blocklist_celltypes (remove cell types) or cell_annotation.safelist_celltypes (limit to specific cell types) methods to constrain the annotation predictions based on what you expect to see in the tissue.

That's often quite helpful after a first attempt using unconstrained annotation.