300,000 cells – lasso to explore

Charts
Two linked jupyter-scatter plots of 300,000 synthetic single cells, mirroring selected() between them, with Python-driven selection and camera presets.

Two jupyter-scatter plots of a synthetic 300,000-cell single-cell atlas: a 2D UMAP embedding, and a marker-gene scatter (CD3E vs. CD14). Dragging a lasso on either plot updates the other one’s selection too – true two-way linked brushing, kept in sync by mirroring selected() from each plot onto the other’s .widget.selection trait. “Color by” restyles both plots in place (the 300k points cross the wire exactly once, at first render); two preset buttons drive the selection and camera from Python.

300,000 cells – lasso to explore

Run it

git clone --depth 1 https://github.com/posit-dev/py-shinywidgets
cd py-shinywidgets/website/gallery/showcases/single-cell-atlas
uv run shiny run app.py

Source: app.py · Open in Shinylive

Try this

  • Drag a lasso on either plot (hold Cmd/Meta to add to the selection, Alt to remove, scroll to zoom): the other plot highlights the same cells, and the composition panel recomputes cell-type percentages and median marker expression for exactly that selection.
  • Click Find the top 1% CD3E cells: Python computes the 99th-percentile threshold, selects those cells on both plots, and zooms the embedding to fit them.
  • Switch Color by between cell type and the two marker genes: both plots restyle without losing the current selection or zoom.

Accessors demonstrated

  • umap.selected() / genes.selected() – the array of currently-selected point indices on each plot; read on both sides of the linking effect (np.array_equal guards against an infinite mirror loop) and by the composition panel.

Verification

No extra packages needed (jupyter-scatter is already in the shared venv). A real Playwright click on Find the top 1% CD3E cells changed the selection count from “All 300,000 cells” to “3,000 cells selected (1.0%)” – exactly 1% of 300,000, confirming the server-side np.quantile threshold and the client-side selection trait agree – and the embedding visibly zoomed to the selected cluster while the marker-gene plot mirrored the same selection. See img/hero.webp and img/after_top.webp.

Source: promoted from docs/widget-guides/jscatter/gallery/showcase_single_cell_atlas.py.