Compensation, without the hand-waving

01 / The problem

Detectors cannot tell dyes apart

A fluorochrome does not emit one wavelength. It emits across a broad band, rising steeply on the blue side of its peak and trailing a long way off to the red. A detector is a bandpass filter in front of a sensor, so it collects every photon that falls inside its window and has no idea which dye sent it.

Put FITC and PE in the same tube and the detector assigned to PE sees real PE, plus a share of the FITC whose tail reaches into the same window. That share is called spillover.

02 / The fix

What compensation actually does

Because spillover is a fixed fraction, it can be measured and removed. Stain a sample with one dye only, look at what lands in every other detector, and you have the fraction that dye contributes to each. Do that for every dye in the panel and the fractions form a matrix. Inverting it recovers an estimate of how much of each dye was really on the cell.

Compensation is a calculation with a correct answer, not a setting to be adjusted until the plot looks reasonable. Two people with the same controls should arrive at the same numbers. If they do not, one of them is adjusting rather than measuring.

The arithmetic runs on linear values, before anything reaches a log axis. That ordering matters for the reason in section 04.

03 / Controls

The rules that make the number right

Compensation is only as good as the single-stained controls behind it. These rules are not style preferences. Break one and the matrix is wrong in a way the software cannot detect.

1. One control per fluorochrome, and it must be that exact fluorochrome

   Tandem dyes are the trap. Two vials of the same catalog tandem from different lots can have different emission, because the acceptor degrades. Compensate with the vial you stained with.

2. The control has to be at least as bright as the sample

   The fraction is estimated from the positive population. A dim control estimates it from a short lever, and the error scales up across the whole range of your data.

3. Positive and negative in the same control must share autofluorescence

   The negative defines the baseline the positive is measured against. A bead positive against a cell negative compares two different baselines, so the subtraction inherits the difference. Keep each control internally consistent.

4. Collect enough events in both populations

   The matrix is built from medians. A few hundred events in a sparse control is a noisy median, and every downstream sample carries that noise.

5. Recalculate when the instrument changes

   New detector settings, new optical configuration, or a service visit means the old matrix describes an instrument that no longer exists.

04 / The limit

What compensation cannot fix

Correct compensation puts the median of a negative population back where it belongs. It does not put back the precision that was lost getting there, and this is the part that surprises people.

Detecting light is a counting process, so every measurement carries error that grows with the size of the signal. When compensation subtracts a large spillover from a large signal, the error of both survives into the answer. The population lands in the right place and arrives wider than it started.

Compensation corrects position, not resolution. A detector receiving heavy spillover ends up with a negative population that is correctly centered and badly spread, and a dim marker measured in that detector can be impossible to separate from it. Nothing was done wrong. The information was gone before the arithmetic started.

This is spillover spreading, and it can be measured for every detector pair on an instrument and written down as a matrix. That matrix is the honest input to panel design. It tells you which channels can carry a dim marker and which cannot, before you spend money on antibodies and find out the hard way. Putting the brightest dyes on the dimmest markers is only half the job. The other half is keeping your dim markers out of channels that other dyes spread into.

Modeling that spread and choosing panels to minimize it is the subject of US 11,935,629, which I invented while at BD.

05 / Spectral

What changes on a full spectral instrument

A conventional instrument gives each dye one detector and corrects the overlap between them. A full spectral instrument measures the whole emission signature across every detector and solves for how much of each dye is present using all of it at once. The vocabulary changes from compensation to unmixing.

The underlying algebra is the same, with more equations. What it buys you is the ability to separate dyes that look nearly identical in any single channel, and the option to treat autofluorescence as its own signature and pull it out rather than living with it.

What does not change is section 04. Unmixing is still subtraction, subtraction still carries error, and spread still decides whether you can see a dim population. More detectors do not exempt a panel from being designed.

Further reading

Credit where it is due

The clearest free explanation of compensation on the internet belongs to Mario Roederer, and has for about as long as there has been an internet to put it on. Anyone serious about this should read his pages rather than settle for mine, which are a summary of what the field learned largely from him.

• Pages Mario Roederer on compensation. Preserved at the Internet Archive back to 1999 if the original is ever unreachable.
• 2001 Roederer M. Spectral compensation for flow cytometry: visualization artifacts, limitations, and caveats. Cytometry 45(3):194–205. Where the negative values and the case for FMO controls are laid out properly.
• 2013 Nguyen R, Perfetto S, Mahnke YD, Chattopadhyay P, Roederer M. Quantifying spillover spreading for comparing instrument performance and aiding in multicolor panel design. Cytometry Part A 83A(3):306–315. The paper that turned spreading into a number you can design against.