---

name: visualization-and-results description: "Handle CellTypist AnnotationResult exports, AnnData insertion, UMAP plots, and dotplots safely." disable-model-invocation: true metadata: disco-role: operating license: MIT

Visualization and Results

Use this sub-skill after a CellTypist annotation already produced an AnnotationResult and the user needs summaries, exported tables, AnnData metadata, UMAP figures, or prediction-vs-reference dotplots.

Route Here When

• The user asks what is inside predictions.predicted_labels, predictions.decision_matrix, predictions.probability_matrix, or predictions.adata.
• The task is to export CSV or Excel tables with AnnotationResult.to_table().
• The task is to insert labels, confidence scores, decision scores, or probabilities into AnnData with AnnotationResult.to_adata().
• The task is to create CellTypist UMAP plots with AnnotationResult.to_plots() or compare predictions against manual labels/clusters with celltypist.dotplot().
• The task is to debug missing majority_voting, missing conf_score, dotplot reference mismatches, output folder errors, or slow UMAP computation.

Route Elsewhere

• Generate the AnnotationResult with celltypist.annotate() in annotation-workflows.
• Download, inspect, subset, convert, or extract markers from models in model-management.
• Interpret training artifacts or custom model outputs in training-and-custom-models.

Working Pattern

1. Inspect the result shape first: predictions, predictions.predicted_labels.head(), and predictions.summary_frequency(by="predicted_labels").
2. Choose the output form: to_table() for files, to_adata() for downstream Scanpy workflows, to_plots() for quick UMAP files, or celltypist.dotplot() for comparing predictions to manual labels/clusters.
3. Decide whether the workflow uses raw labels (predicted_labels) or majority-voted labels (majority_voting); do not request majority-voting confidence or dotplots unless that column exists.
4. Validate exported artifacts with scripts/result_shape_check.py before plotting or handing files downstream.
5. Use references/troubleshooting.md when an output folder, majority-voting column, reference labels, order lists, or UMAP step fails.

References

• Result objects and exports: AnnotationResult attributes, summary_frequency(), to_table(), and to_adata().
• Plotting and dotplots: to_plots() UMAP behavior, celltypist.dotplot() semantics, order/filter behavior, and interpretation.
• Troubleshooting: focused fixes for output folders, majority voting, UMAP compute, dotplot label mismatches, and shape validation.