Closed ubyndr closed 4 months ago
I'm not sure we have any examples of this yet, but we have discussed solutions. We need to fulfill the stated aim of having the most precise ontology term available mapped to a user annotation. I think the simplest would be to scan for any cell_type annotations that have no mappings to author cell sets after the current mapping algos run. For these, we should map to the smallest number of cell mapping cell sets.
Here's an attempt at laying out the logic:
If Cell set defined by cell_type annotation (E) subsumes the cell sets defined by user annotations e1, e2, e3. AND e1, e2, e3 don't subsume each other And e1+e2+e3 = E (?) then annotate e1, e2 and e3 with E.
Related issue:
If you examine the co-annotation analysis results below, our situation appears to be the opposite of what you've described, doesn't it
Indeed. In these cases there is no way to associate an ontology term with the cell sets based on what's in the AnnData file, and that's fine.
We are encountering a challenge in cases where cell_type annotations do not have direct one-to-one mappings with any author-defined cell sets, which I refer to as the edge case. This issue primarily affects the assignment of cell type ontology IDs in scenarios involving one-to-many mappings. The lack of straightforward mappings complicates the ontology ID assignment process, leading to potential inaccuracies or the inability to assign an ID altogether.
cc @dosumis @hkir-dev