OmicVerse Single-Cell Preprocessing

Goal

Turn raw or already-QC'd single-cell AnnData into a reusable execution spine: QC, preprocessing, PCA, graph construction, embeddings, Leiden clustering, and optional marker discovery. Keep marker discovery in the same skill because it only becomes useful after the same cluster-ready object exists.

Quick Workflow

1. Inspect the input AnnData shape, layers, and whether a cluster key already exists.
2. If the data is raw, run ov.pp.qc(...) first; if it is already filtered, jump to preprocessing.
3. Choose one preprocessing branch with ov.pp.preprocess(...), then keep the HVG output and build scaled, PCA, neighbors, and an embedding.
4. Run ov.pp.leiden(...) after the graph exists.
5. If you need marker genes, choose a method that matches the input representation, then call ov.single.find_markers(...) and ov.single.get_markers(...) or ov.pl.markers_dotplot(...).
6. Validate the expected obs, var, obsm, obsp, and uns keys before writing output.

Interface Summary