Skip to content

Spillover spreading matrix

The SSM node measures how much each single-stain control spreads the signal in every other channel. Spillover you can compensate away; spillover spread you cannot - it is the extra variance a bright dye adds to a neighbouring detector, and it is what limits how dim a population you can resolve there.

Create one with SSM on the project panel toolbar.

How the node is organised

The plot is a ribbon per well, like a well ribbon, with one important difference: a well ribbon plots the same parameter for every well, while an SSM node gives each well its own parameter - the channel that well is stained in.

A well whose name - as the node's well list shows it - names one of the node's parameters is given that channel: CD38 BV711 (Beads) takes CD38_BV711. Spaces and punctuation are ignored, but only whole words count, so CD3 is not found in CD38. For any other well PlatAnJr detects the channel: the one with the highest positive median among the channels whose positive gate holds at least 1% of the well's events, so a few bright outliers in another channel do not win. Right-click a well's bar to correct the choice. A well you have not set by hand follows its name, also once an unmixing adds the parameter it names.

Each bar carries two gates - the negative and the positive population of that single-stain control - placed automatically and draggable. A gate reaches SSM Gate Width half-maximum widths from its peak - on the MoreOptions page, 2 by default, about 98% of a Gaussian population - but never past the valley between the two, nor into the end bins where off-scale events pile up. Changing the setting refits every SSM node's gates, as Reset does.

Button Effect
Detect Re-run the automatic detection of channels and gates
Reset Reset the gates only; the channel assignments stay
Export Write the matrix out as CSV
Show % Switch the bars between counts and percentages

The tabs

SSM
The matrix itself. Each row is a single-stain control, each column a detector, and each cell is the spread that stain causes in that detector. The cells are coloured as the plate heat map, from the smallest value to the largest; an exact 0 - no spread measured, as in a control's own channel - stays white.
Stain Index
Per control: the negative and positive medians, their robust SDs and the resulting stain index.
Histogram
The distribution of the selected control with its two bands drawn on it - the place to check whether a band has been placed sensibly.

Reading the matrix

Look down a column. A detector with several large entries is receiving spread from several dyes: any dim marker you put in it will be hard to resolve. That is a panel design problem, not a compensation problem, and no amount of adjusting the spillover matrix will fix it.

Look along a row and you see which detectors a given dye damages. A very bright dye with a wide row is a candidate for moving to a different detector or dropping to a lower concentration.

The definition used here

PlatAnJr's SSM is unitless:

SSM = sqrt(varS − varR) / (Fs − Fr)

where varS and Fs are the variance and median of the stained control in the detector being examined, and varR and Fr those of its reference (negative). Variances are computed robustly, from the 84th and 50th percentiles, so a few outlying events do not dominate.

This is deliberately not the standard formulation

The usual Nguyen SSM divides by the square root of the fluorescence difference and carries units. PlatAnJr's definition is unitless, which makes values directly comparable between detectors and between instruments. Do not compare a number from here against a published SSM computed the other way.

The matrix is computed on linear values even though the gates are drawn on the display scale - a spread statistic is only meaningful on linear data.

Running one

  1. Acquire a single-stain control per dye, plus an unstained control, in their own wells.
  2. Open the plate and select those wells on the heat map.
  3. Create an SSM node.
  4. Check the detected channel for each control; right-click any bar that was assigned the wrong one.
  5. Check the band placements on the Histogram tab; drag any that are wrong.
  6. Read the matrix and Export it.

A node in a project opened before its unmixing finds its parameters once Add Unmixed Parameters puts them in; its saved gates are kept until then.