omicverse-reference-label-transfer

To transfer cell-type labels, map a reference AnnData object onto a query AnnData object using shared gene names and a reference cell-type column. This produces predicted labels and uncertainty values for single-cell datasets.

Integrating single-cell datasets requires selecting a transfer backend like harmony, scVI, or scanorama. These methods build a shared integrated space to map reference labels onto query cells for annotation.

Label transfer in OmicVerse provides uncertainty scores alongside predicted cell types. It validates branch-specific prediction and uncertainty keys before downstream plotting to ensure reproducible cell annotation.

Before running label transfer, you need paired query and reference AnnData objects with shared gene names and a reference cell-type column. AnnotationRef preprocessing and training are also required to validate branch-specific keys.

Single-cell label transfer fails with unmatched gene names because it requires shared gene names to build a shared integrated space. Preprocessing validation checks ensure reference and query AnnData objects are compatible before integration.

You can use scanorama for label transfer instead of harmony. The workflow supports harmony, scVI, and scanorama transfer branches to accommodate different single-cell dataset integration needs and produce prediction outputs.