OmicVerse Single-Cell Annotation

Goal

Annotate one clustered AnnData object with one OmicVerse backend at a time. Treat celltypist, gpt4celltype, and scsa as alternative branches that share the same input/output contract, not as sequential stages.

Quick Workflow

1. If the input is raw counts, hand it off to the OmicVerse preprocessing path first; this skill starts from annotation-ready AnnData with a cluster column such as leiden.
2. Create ov.single.Annotation(adata) once.
3. Choose exactly one backend:
   • celltypist when you already have a CellTypist model path or want to resolve one with query_reference(source="celltypist").
   • gpt4celltype when you want LLM-based cluster labels and can provide AGI_API_KEY.
   • scsa when you want marker-database scoring from clustered cells and have or can point to a local SCSA database.
4. Pass only the backend-specific kwargs for the selected branch.
5. Validate the expected adata.obs label column and any branch-specific side products before plotting or downstream analysis.

Interface Summary