How many samples each population is actually measurable in

One bar per population, split by whether that population's event count in each sample was enough to quantify it, enough to detect it, or neither.

Usage

fig_detection_limits(ufreq, outfile, dpi = 200, colors = fcs_colors())

Arguments

ufreq

the frequencies element of run_gate_uncertainty()

outfile

Path to write the figure to.

dpi

Resolution in dots per inch. Default 200.

colors

Named list of colours; defaults to the package palette. Default fcs_colors().

Details

WHY A COUNT RATHER THAN A SCATTER. The limit of detection depends only on how many parent-gate events a sample contributed, so it is a property of the sample and identical across the populations within it. Plotting frequencies against it would put every population on the same reference line and say nothing extra. What varies, and what decides whether a population is worth testing, is how many samples clear the limit at all.

HOW TO READ IT. A population quantified in every sample is measurable in this cohort. One that is mostly below the limit of quantification is not, and no change to the gating strategy will fix it: the numerator is small because few cells were acquired, so the answer is a longer acquisition or a higher --max-events-per-file, not a different threshold. A population split between the two is the dangerous case, because its group difference can be driven entirely by which samples happened to clear the limit.