OmicVerse Reference Label Transfer

Goal

Map labels from a reference AnnData onto a query AnnData by building a shared integrated space, training a weighted kNN transfer model, and writing predicted labels plus uncertainty scores back to the query object.

Treat harmony, scVI, and scanorama as alternative integration branches. They share the same query/reference contract, so keep them in one skill rather than splitting into thin wrappers.

Quick Workflow

1. Start from a query AnnData and a reference AnnData that share gene names.
2. Make sure the reference has a cell-type column such as celltype in adata_ref.obs.
3. If the data are already log-normalized, recover or restore counts before concatenating.
4. Construct ov.single.AnnotationRef(adata_query, adata_ref, celltype_key=...).
5. Run preprocess(mode=...) on the combined object.
6. Choose one integration backend with train(method=...).
7. Transfer labels with predict(method=...).
8. Validate the method-specific prediction keys before plotting or writing outputs.