omicverse-cross-modal-celltype-transfer

What problem does it solve?

Transfers cell-type annotations from a reference AnnData to a query AnnData using a shared embedding, enabling cross-modal reuse of curated labels across datasets.

Core Features & Use Cases

• Weighted KNN transfer core: train with ov.utils.weighted_knn_trainer on the reference embedding and apply with ov.utils.weighted_knn_transfer to the query, obtaining predicted labels and uncertainties.
• Embedding flexibility: works with any shared embedding in .obsm (e.g., X_glue) and supports label_keys prefix matching to select the target annotation column.
• Writeback and visualization: write transf_celltype and transf_celltype_unc back to query.obs; optional visualization via embedding() for quick validation.
• Use cases: label an ATAC query from an RNA reference or convert OmicVerse cross-modal notebooks into reusable annotation workflows.

Quick Start

Train the KNN model on the reference embedding and transfer labels to the query using the trained model, then write back transf_celltype and transf_celltype_unc to the query AnnData.