CellTypist Annotation

CellTypist Annotation labels cells in a single-cell dataset using a local CellTypist AI Model.

Use it for

• first-pass cell-type annotation;
• checking whether a dataset roughly matches a known reference;
• preparing a single-cell dataset for later exploration.

Inputs

• AnnData .h5ad file.
• CellTypist model name, for example Immune_All_Low.pkl.
• Optional majority voting.

The .h5ad file should contain normalized expression values. Raw counts can fail or produce poor labels.

Outputs

• CSV with predicted labels.
• JSON summary.
• Cell count.
• Label count.
• Top label fraction.
• Provenance.

How to read the result

Start with the label distribution table. If one label dominates almost all cells, ask whether that makes sense for the dataset.

The top label fraction is the fraction of cells assigned to the most common label. A very high value can be correct for a purified dataset, but suspicious for a diverse tissue sample.

Good first pipeline

1. Add a small .h5ad file to Data.
2. Install CellTypist Local Annotation.
3. Add CellTypist Annotation to a pipeline.
4. Select the .h5ad file and model.
5. Run and inspect labels in Results.

Technical details

Tool ID: ai-celltypist-annotate

Compatible model:

• CellTypist Local Annotation

Outputs follow the shared Liatir tool output contract and include provenance with model ID, runtime, input file, selected CellTypist model, and majority voting setting.