motility_painting

Linking live-cell-imaging (LCI) motility trajectories to Cell Painting (CP) single-cell morphology, via deck.gl + Celldega widgets.

Python · Jupyter · anywidget deck.gl · WebGL Celldega clustergram Spearman + BH-FDR Experimental overview → Engineering notes →

What is this?

~4,700 cells were imaged live for 65 timepoints (LCI), tracked into motility trajectories, then fixed and imaged again with Cell Painting (CP). A per-FOV similarity-transform alignment links each CP cell to its own motility trajectory (shared link_id), so a cell's fixed morphology can be compared directly to how it moved while alive. This site is a browsable, kernel-free view of that pipeline's widgets and findings — no notebook required.

Live, in your browser

Pick a view. Each one is a self-contained widget with no kernel attached — all interactions (scrub/play, hover, click-to-link, pan/zoom) run purely in the browser.

open standalone ↗

The dropdown swaps between the three widget pages. Each is fully self-contained (no CDN calls at page load) via a small custom static host implementing anywidget's own AFM spec — see the engineering notes for why the classic ipywidgets embedding path doesn't work here.

All results

Motility scrubber
Scrub/play through all 65 LCI frames with tracked-cell paths; every other detected cell shown as a background dot.
Linked view (motility ↔ CP)
Synced final-frame motility | stitched Cell Painting panels; matched cells share a link_id, click to highlight a pair.
Clustergram & correlations
Celldega clustergram of CP features (colored by motility tertile) + the CP↔motility Spearman correlation heatmap.
NF01 · Displacement null model
No population-level excess net displacement vs. a real-steps-only null (p=0.40) — but short-range turning-angle persistence is real.
NF02 · Morphology over time
Shape variability correlates with speed far more strongly (ρ≈0.6) than any fixed Cell Painting feature (ρ≈0.15).
NF03 · Crowding & collective motion
Crowded cells move measurably slower (ρ=-0.091); weak local coordination in heading between neighbors.
NF04 · Multiple-testing correction
Real BH-FDR code, all 87 surviving CP↔motility associations browsable in one table + overview plot.
NF05 · Collisions ("bumper cars")
96,408 genuine polygon-contact collisions; weak contact-inhibition-of-locomotion signal.