Scanpy compatibility

Supports Scanpy 1.11.5 through 1.12.x.

Scanpy 1.12.4 cannot compute PAGA tree layouts (rt or rt_circular). Supply positions with pos=... or use another layout. Other supported PAGA calls work.

Available plots

See the guides for examples and the API reference for arguments and defaults.

Plot family	Available calls	Guide
Embeddings	embedding, umap, tsne, diffmap, draw_graph, pca, pca_scatter	UMAP and embeddings
Cell or gene scatter	scatter	UMAP and embeddings
PCA diagnostics	pca_loadings, pca_variance_ratio, pca_overview	PCA diagnostics
Grouped expression	dotplot, matrixplot, DotPlot, MatrixPlot	Expression plots
Individual-cell expression	heatmap, clustermap, tracksplot	Expression plots, tracks
Matrices and trees	matrix, correlation_matrix, dendrogram	Expression plots
Ranked genes	ranking, rank_genes_groups, rank_genes_groups_dotplot, rank_genes_groups_matrixplot, rank_genes_groups_heatmap	Ranked genes
Embedding density	embedding_density	UMAP and embeddings
Quality checks	highest_expr_genes, highly_variable_genes, filter_genes_dispersion, scrublet_score_distribution	Data checks
Profiles and trajectories	timeseries, timeseries_subplot, timeseries_as_heatmap, sim, dpt_groups_pseudotime, dpt_timeseries	Profiles and simulations
PAGA	paga, paga_compare, paga_path, paga_adjacency	PAGA graphs and paths
Brush-linked summaries	EmbeddingSummary	Linked expression summaries

EmbeddingSummary is a GenomeSpy addition, rather than a Scanpy class. It links an embedding brush to group means, histograms, composition bars or gene tracks.

Not yet available: violin, stacked_violin, StackedViolin, their ranked violin variants, rank_genes_groups_tracksplot and spatial. Keep using Scanpy for those plots and its global style-setting helpers.

Shared differences from Scanpy

• Analysis stays in Scanpy. Compute embeddings, gene rankings, density and scores before plotting. Requested missing dendrograms may be computed through Scanpy; clustermap can compute cell/gene linkage, and PAGA can compute layout positions. These operations do not rerun the underlying embedding or ranking.

• Results are native GenomeSpy charts. Use show=False to receive a chart for configuration, composition or saving. show=True displays it and ordinarily returns None; show=None follows Scanpy’s autoshow setting. Builders return a chart through .to_chart().

• Matplotlib objects are not part of this API. ax, return_fig, figure/artist access and Matplotlib styling objects are unsupported. Unsupported options raise named errors rather than being silently ignored.

• Sizing and colors use GenomeSpy. width and height are plot-area sizes in logical pixels; axes, titles and legends add space. Native schemes and CSS colors are supported. Matplotlib colormap and normalization objects are not. Fonts, palette interpolation and layout can differ from Scanpy’s rendering.

• Saving uses the returned chart. chart.save("plot.html", inline=True) creates interactive HTML; JSON is also supported. Scanpy’s automatic image-saving options are unsupported. Use save=False if Scanpy autosave is enabled.

• Browser interactions work in exported HTML. Reading selected cells back into Python requires a live notebook kernel. The guides document each plot’s hover, navigation and selection controls.

Known limitations

• Storage support varies by plot. Cell-expression views require resident expression. They have no fixed cell or value cap. Preparation reads the chosen genes; its memory use depends on cells × genes. Clustermap computes pairwise distances unless linkage is supplied.

• Linked selections have a boundary mismatch. In EmbeddingSummary, cells exactly on a brush’s upper or right edge can appear in summaries but be omitted from the displayed count and Python selection results.

• Changing linked heatmap genes can misplace rows. With bundled GenomeSpy Core 1.2.0, a gene-menu change can leave tiles or value labels at old positions. Choose genes in Python when reliable row placement is required.

• Some legends can clip. Large symbols or vertical legends with multiple columns can exceed their allocated space in the bundled runtime. Inspect the final layout, especially with many categories.

• Large live counts can slow brushing. Exact counting currently scans plotted cells on brush updates. The faster native path awaits an adopted published Core release.