omicverse-single-cell-metatime-annotation

To annotate tumor microenvironment cells with MetaTiME, you over-cluster a batch-corrected AnnData object, score clusters against pretrained meta-components, and write both fine-grained and major labels back into the dataset for visualization.

MetaTiME cell-state annotation requires a human, log-normalized single-cell AnnData object containing a pre-computed batch-corrected embedding, such as X_scVI or X_harmony, to accurately execute the over-clustering and scoring workflow.

MetaTiME requires over-clustering at a high Leiden resolution to ensure homogeneous cell populations, which allows precise scoring against pretrained meta-components and accurate assignment of both fine-grained and major cell-state labels.

Yes, you can visualize MetaTiME annotations on UMAP or MDE coordinates using collision-aware cluster labels, allowing you to plot both fine-grained cell states and major category labels directly on the embedded single-cell data.

To diagnose unassigned cells after MetaTiME annotation, compare the generated meta-component scores against manual marker-based annotation to identify why specific clusters lack defined cell-state or major category labels.