OmicVerse Single-Cell Clustering Backends

Goal

Turn the notebook's clustering comparison into one reusable job: choose and run a clustering backend on a preprocessed single-cell AnnData. Keep this skill focused on backend selection and output contracts after PCA and, when needed, graph construction.

Quick Workflow

1. Inspect whether the input already has PCA or another embedding, and whether a neighbor graph already exists.
2. Choose a backend explicitly: leiden, louvain, scICE, or GMM for the notebook-covered paths.
3. Build a neighbor graph first for leiden and louvain.
4. Pass use_rep explicitly for embedding-based methods such as GMM and scICE.
5. Validate which obs columns were added, and treat ARI or figure rendering as optional downstream reporting rather than the core contract.

Interface Summary