Skip to content

Changelog ​

1.0.0 (2026-11-18) ​

Format 1, final. The bundle format states what stays fixed, what is required and what is optional, with validate as its conformance check; python -m spatialscape.fidelity measures a built bundle against its source, and the browser's region test against scipy. The correctness fixes from the outside review of 0.6.0: a test compares a gene only among the cells that measured it, a measured gene with no counts draws as zero instead of grey, and fold changes follow the scale the bundle stores. A ↻ tool turns a section 90° on screen and the link keeps it. An optional layer whose file is missing (an image, boundaries, transcripts) is left out with a note instead of stopping the build, and SpatialExperiment images stored as URLs are fetched. The genome panel keeps the coverage it has read, and a cold gene switch takes a third fewer requests. Eleven public datasets across technologies are live and reproducible from their scripts; benchmark scripts time a build and the viewer on a bundle of your own. With transcripts on, selected genes draws only those genes, and all genes draws every gene in its own colour. A section or embedding of up to two million points draws every one once you zoom in, on any window. A gallery can put its cards under headings with group:, and a card shows its tags.

0.6.0 (2026-11-03) ​

The release before 1.0. Two windows: pin any two sections side by side, each with its own camera, from a pin tool that says what a click does. A test in the browser: draw regions, run a Wilcoxon rank-sum test on the cells themselves, per platform, and read the result in the Compare panel, fed by a per-cell row store every bundle now carries. Every input whole: SpatialData stores (zarr v2 and v3) with their shapes, labels, points and images; Seurat objects with their Visium slices; spatialproteomics objects; alongside SpatialExperiment, Seurat and AnnData as before. Comparison tables carry statistics and the matrix shows one at a time; the ROIs group has a switch; multiplexed protein imaging; any six of many image channels; images written band by band and OME-Zarr pyramids passed through; and the review fixes listed in the repository changelog. 1.0 will be this release with the format stated as stable, the documentation finished and every app rebuilt.

0.5.0 (2026-10-29) ​

Compare, by matrix. Differential expression tables of every common shape go in as they are: marker tables (Seurat FindAllMarkers, scanpy rank_genes_groups, presto, scran, or a bare cluster-and-gene list from a paper) become one level per cluster against the rest, and a case/control result fit per cell type is one comparison with a facet; levels link to an annotation's categories, aliases included. The Compare panel is a genes-by-levels matrix with the effect in tinted chips and the level's legend colour in its column head, beside violins of the gene per category (from the per-cell values the maps draw) or the per-replicate dots. A Comparisons group in the side panel, a markers button on every legend row, show().add_markers(adata). Docs by platform family (sequencing-based, imaging-based, single-cell), boundary and molecule tables from CSV as well as Parquet with NanoString and Vizgen conventions recognised, init proposing a Xenium folder's boundaries and transcripts, a segmentation overlay for comparing boundary sets, plainer names in the panel.

0.4.0 (2026-10-23) ​

Xenium at every scale. Cell boundaries are cut into spatial tiles, coloured like the cells (gene, annotation, blend, filter) and drawn filled or as edges once zoomed in, at about 1 ms per frame where whole-section outlines took 30. Transcripts (molecules: from transcripts.parquet) are tiled at 3.5 bytes each and drawn once a view holds about one per screen pixel, following the gene on screen, with pinned genes in fixed colours or every gene in its own hue, and a size slider. Fluorescence images get channel names, colours and fitted windows from the file, a channel list to switch and solo, and a bounded read (max_size holds for files without pyramids; Xenium morphology at 0.85 µm per pixel), and the viewer fetches each image tile once. The side panel is five groups: Image, Cells, Transcripts, Selections, Samples. Comparisons: differential expression results in the bundle, a Compare panel, and the canvas split into one pane per group with per-pane point size.

0.3.0 (2026-10-21) ​

ATAC, multiome and notebooks. Genome tracks from fragment files or existing bigWigs with a genome panel under the maps; spatialscape.show(adata) views an object from the session it is in, with lasso selections and regions flowing back; regions drawn in the viewer as groups of lassos; Seurat objects read directly; a faster, lighter expression writer; review fixes across the build, the viewer and the docs.

0.2.0 (2026-10-14) ​

One yaml per dataset, one command. init reads an R object (.rds / .rda) or AnnData files and writes a dataset yaml that is a menu of the data; build converts R objects once, rebuilds only what changed and applies every edit; site.yaml describes the gallery (site init, site build). Removed convert, plan, refresh and add-sample. Per-sample subset, fixed grid columns, cluster names on embeddings, a shared gene color scale, per-view point sizes, the viewer shipped inside the package (serve, site build), and the S3 + CloudFront hosting recipe.

0.1.0 (2026-10-06) ​

First public version.

Viewer. One-canvas mosaic or strip of all samples with focus and keyboard stepping; gene, annotation, feature-group and two-set blend coloring; legend filtering; independent filter channel (gene, continuous, categorical); hover with all annotations; lasso selection with composition, gene statistics and CSV export; expression-by-annotation dot plot; domain outlines; cell polygons at high zoom; H&E and fluorescence images; platform bar; linked snRNA-seq embedding split view; scale bar; registry gallery with real thumbnails and an optional site bar; full URL state.

CLI. init, plan, inspect, build, add-sample, refresh, outlines, thumbnails, validate, serve, synth. Globs and templates, per-platform defaults, automatic field and image discovery, sharded Zarr v3 uint8 expression, dataset-level vocabularies, OME-Zarr images, outline tracing, polygon packing, thumbnails.

The repository's CHANGELOG.md is the authoritative list.

MIT licensed.