omicverse-single-cell-clustering-backends

To run single-cell clustering on an AnnData object, you select a specific backend like Leiden, Louvain, GMM, or scICE, which then validates your graph inputs and generates corresponding cluster labels in the observation columns.

Leiden and Louvain are graph-based clustering algorithms, while GMM applies embedding-based mixture clustering. Comparing these backends helps determine which method provides deterministic labels or mixture modeling best suited for your single-cell dataset.

Yes, you need a precomputed neighbor graph or an existing embedding in your AnnData object before applying clustering backends. The process validates these required graph or embedding inputs before executing the clustering algorithm.

Cluster labels are validated by checking the observation columns added to the AnnData object. The system verifies that specific labels such as leiden, mclust, gmm_cluster, or scICE_k outputs are correctly generated after backend execution.

Yes, you can use the scICE backend for consistency-guided cluster search in single-cell analysis. It operates alongside Leiden, Louvain, and GMM to compare different clustering behaviors on a prepared AnnData object.

Explicit backend selection is required to ensure a repeatable workflow that targets specific clustering algorithms like Leiden or GMM. This approach prevents manually reconstructing notebook logic each time you need deterministic cluster labels.