Plots Window

Plots Window

Every figure the viewer produces — dotplots, UMAPs, neighbourhood-enrichment heatmaps, co-occurrence curves, ligand–receptor dotplots, CNV heatmaps, the provenance DAG — appears in one dockable Plots panel and is written to <dataset>/plots/.

Opening it

The dock is hidden until the first figure arrives, then reveals itself at the bottom of the window. View → Show Plots (Ctrl+Shift+P) toggles it, and closing it with its own close button unticks the menu item.

Closing the dock only hides it — the figures in it are kept, and the menu item brings them back. If you float the dock and lose track of it behind another window, the menu item raises it; if it has ended up somewhere no screen reaches, it is docked back into the main window instead.

The gallery

Figures are listed newest first. Each card shows a thumbnail, the figure's title, and the files it was written to.

Button	What it does
Open	Opens the figure in its own resizable window with matplotlib's pan / zoom / save toolbar.
Save as…	Writes an extra copy wherever you choose, in whatever format the extension names.
Remove	Drops the figure from the gallery. Files already written are untouched.
Clear all	Empties the gallery.

The gallery keeps the twenty most recent figures; beyond that the oldest is dropped. Matplotlib figures are not small, and a long session produces a lot of them.

Where the files go

Every plot is written to <dataset>/plots/, in each format selected under Preferences → Plot format:

Setting	Files written per plot
PNG + PDF (default)	a 300 dpi raster to look at, and a vector version to publish
PNG	one raster
PDF	one vector file
SVG	one vector file, for further editing

Names are keyed by what the figure is about — dotplot_leiden_r1.0.png, nhood_enrichment_graphclust.pdf, umap_EPCAM_KRT5.png — so a second run against a different clustering does not overwrite the first. A run against the same clustering does overwrite it, which is usually what you want.

Two kinds of output do not go through the gallery:

• Pairwise volcano batches (Rank Genes, ROI Analysis, ARMS Overlay) produce one figure per cluster pair — dozens at once, which would bury everything else. They still ask where to put them, defaulting to <dataset>/plots/volcano_<clustering>/, and honour the format setting.
• CopyKAT heatmaps are drawn by a detached background process in a separate conda environment, which cannot reach the GUI. They are written straight to <dataset>/plots/.

In the notebook

Each figure is recorded as a plot:* node in the provenance graph, and the recorded savefig call names the file that was actually written, relative to the dataset directory. A replayed notebook therefore reproduces the figure and puts it where the session put it.

Two figures are recorded as notes rather than code — the annotation neighbourhood heatmap and the annotation distance plot. Both are computed over shapes drawn in the viewer, which a notebook has no access to; the note records that the figure exists and where it went rather than pretending there is code to replay.

Deleting them

<dataset>/plots/ is viewer-created, so Tools → Dataset lists it as deletable. Nothing else depends on it.