Automated Gating of a Published Panel

Reference only

This page runs on the OMIP-039 deposit, FR-FCM-ZYY6, which is far too large to ship in a container. The code and every number below come from a real run in the everything-flow-cytometry repository, in scripts/02_omip39_compensation_and_gating.R and scripts/04_omip39_reproduce_paper.R. The automated gating page runs the same method on packaged data and does execute here.

The automated gating page shows a template fitting each sample’s own cut point, and argues that this beats copying one boundary across a plate. It cannot show whether the template lands where a human would, because there is no human answer to compare against.

OMIP-039 is a deposit where there is. The authors published their panel, their files, and their own gates, so an automated template can be run against the same data and the two answers put side by side.

The comparison

Each row is one population. The manual percentage is the authors’ gate, the automated one is the template’s, and the note records how the two strategies differ in construction.

"manual","automated","manual_percent","automated_percent","difference","verdict"
"Viable CD14- CD19-","Viable",72.1005412783453,74.185526675711,2.08498539736573,"agree"
"Single Cells","SingleCells",96.4719847325319,86.9317822066125,-9.54020252591943,"differ"
"Lymphocytes","Lymphocytes",77.3182331426279,71.6551541682242,-5.66307897440365,"differ"
"CD56+ CD3-","CD56posCD3neg",8.24798411438465,5.79516255125271,-2.45282156313194,"agree"
"CD56+CD3+","CD56posCD3pos",3.64169511640632,2.22006639870013,-1.42162871770619,"agree"
"CD56- CD3+","CD56negCD3pos",85.1141714382596,91.0861015651976,5.97193012693806,"differ"
"CD56- CD3-","CD56negCD3neg",0.913345817824051,0.898669484849527,-0.0146763329745241,"agree"

The viability gate agrees to two points. The double negative population agrees to a hundredth of a point. The singlet gate is nine and a half points apart.

Where the disagreements come from

The largest gaps are not in the immunophenotyping. They are in the cleanup gates at the top of the hierarchy, and each one traces to a difference in how the boundary is drawn rather than to a difference of opinion about the biology.

The singlet gate is forward scatter area against height. The analyst drew a polygon by hand and the template fits singletGate, which is stricter, so the template throws away nine percent more events. The lymphocyte gate is the same story with flowClust at K equal to two against a hand drawn polygon.

The T cell gate differs by six points for a different reason. The analyst used one quadrant gate on CD56 against CD3, and the template uses two one dimensional gates combined with a boolean. A quadrant forces every event into exactly one of four boxes. Two independent cuts do not, so the populations do not have to sum to a hundred and the T cell fraction lands higher.

None of that is a bug in either method. It is the cost of the hierarchy being a set of choices, and it is invisible until someone runs both.

Whether the paper’s claims survive

The percentages are a means to an end. What matters is whether the conclusions of the paper hold under an analysis its authors did not run.

"route","verdict","claims"
"manual, the authors' gates","reproduced",6
"automated, openCyto template","reproduced",5
"automated, openCyto template","too small to call",1

Six of the paper’s claims reproduce from the authors’ own gates. Five of the six reproduce from the template, and the sixth is not contradicted, it lands on a population the template gates too small to test. So the automated route agrees with every claim it is able to evaluate, and the one it cannot evaluate is a consequence of the stricter cleanup gates above it.

That is the honest summary of what template gating buys and costs. It removes the analyst from the loop and it reproduces the science. It also gates one rare population out of existence, and the only reason anyone knows that is the comparison.

What to take from this

Run both when the deposit lets you. A template that agrees with a published manual analysis on the biology, while disagreeing on the cleanup gates, is telling you something true about where analyst variation actually lives, and the harmonisation page measures the same effect from the other direction.