Scientific Reports · 2026

when a signal takes sides

i wrote the image analysis: nucleus segmentation, the perinuclear region, the R metric, and the figure. this visual cycles through six deposited nuclei, following each from its outline to a measure of one-sided signal, then placing it among every cell in the analysis.

  1. start with the nucleus. the outline comes from the deposited Cellpose segmentation of DAPI, not a schematic circle. the six examples use the same selection rule, one per knockdown × infection group: among interior nuclei with at least five times the minimum accepted perinuclear pixel count, choose the one nearest the full group’s median R. ties use table order.
  2. find the perinuclear signal. Otsu thresholding selects above-background mCherry pixels outside all nuclei; each pixel belongs to the nearest nucleus, within the configured 1 μm radius cap. the glow sweeps around the nucleus by angle. each example displays 400 deterministically selected real pixels, not simulated signal.
  3. turn position into direction. each pixel has an angle θ around the nucleus center and an intensity weight w. each point flicks into a short weighted stroke at an enlarged display scale; the strokes then chain head-to-tail behind a moving head. the displayed pixels carry aggregate, positively adjusted weights that preserve both components of the full-data normalized resultant, rather than changing the measured score. all six full-data scores reproduce the deposited table.
  4. read the closing arrow. the asymmetry score is the intensity-weighted circular resultant length: R = |sum_i w_i (cos theta_i, sin theta_i)| / sum_i w_i. the arrow runs from the path’s start to its end; the dotted ring is unit length. R = 0 means the weighted directions cancel; R = 1 means they all point the same way. a low R need not imply a uniformly bright ring.
  5. put one number in context. each point in the strip plot is one of all 545 deposited cells, grouped by knockdown and infection using the table’s condition names. horizontal position is R; vertical jitter only separates overlapping points. white ticks mark group medians; six outlined points mark the examples. the current group stays highlighted, and its example lights up and settles into place when the calculation finishes, before the next nucleus fades in. reduced motion shows one finished calculation with all six markers indicated. edge-flagged cells remain in the distribution. these are descriptive cell-level distributions, not an estimate of independent experimental replication.

[paper][code][back]

data: signal-aggregation-comparison @ 252ce3a, authored by me · results/cells.csv, the six selected examples’ deposited images and cached nucleus labels, and the analysis configuration · generated by scripts/work/polarization-index.py; input checksums and the exact example-selection rule are in the visual’s data.