PeacoQC and Rare Events: Keep the Question When Cleaning the File

A clean cluster can still be the wrong biological story

When a small population appears in an automated analysis, the first question should include time: did these events occur throughout acquisition, or mainly during an unstable segment? A coherent cluster is evidence of similar recorded features. It is not, by itself, evidence that those features represent a distinct biological population. Reviewing acquisition quality before interpreting clusters makes the inference easier to audit.

The original methods evidence

Emmaneel and colleagues introduced PeacoQC, which follows density peaks across acquisition bins and removes anomalous regions. Their quantitative comparison used 55 manually annotated FlowCAP IV files. PeacoQC and flowCut did not differ significantly on the reported quantitative measures. A low-event file was overfiltered in the broader evaluation. This is a 2022 methods paper, first published online in 2021: Cytometry Part A, DOI 10.1002/cyto.a.24501.

CFCM's practical analysis below considers how to evaluate a cleaning step for a rare-event question; it does not claim that the paper validated a clinical MRD assay.

Keep an auditable path from raw events to the answer

For a clinical transfer study, retain the source FCS file, preprocessing choices, software version, parameters, channel list and the event indices excluded by cleaning. Save the temporal plots alongside the final population plot. A second reviewer should be able to identify which operation changed the denominator and to reconstruct the result without guessing from a screenshot.

Do not optimise cleaning parameters against the answer you hoped to obtain. Set them on representative development material, define exceptions, and then assess a separate challenge set. An informative challenge set should include both true acquisition problems and clean samples with unusual biology. Otherwise an algorithm can look successful simply because the evaluation contains too few difficult counterexamples.

Evaluate the population affected by exclusion

Consider an explicitly hypothetical file containing one million evaluable leukocytes and twenty candidate abnormal events. If a cleaning step removes one hundred thousand background events but retains all twenty candidates, the reported percentage rises from 0.002% to approximately 0.00222%. If it instead excludes half of those candidates, the interpretation changes for a different reason. These numbers are arithmetic illustrations, not patient observations or performance estimates for PeacoQC.

The appropriate response is to inspect both the removed region and the retained events. Record whether candidate events are concentrated in time, whether other channels show a simultaneous disturbance, and whether the sample needs reacquisition or a qualified no-call. A rule that always retains suspicious events is no more defensible than a rule that automatically removes every unusual cluster.

Agreement with annotation has a defined scope

Human annotation is a reference process with its own uncertainty. For local evaluation, specify what annotators saw, how disagreements were resolved and what counted as a true error. Keep event-level cleaning metrics separate from specimen-level sensitivity, quantitative recovery and the clinical consequences of a missed population. A high average score can coexist with an unacceptable result in the small subset that matters most.

Also assess what happens when there are too few events to support a stable estimate. The absence of a software error is not an adequacy decision. A clinical workflow needs an explicit route for low-count, truncated or otherwise non-evaluable files, with an explanation that survives export into the final report.

Daily Learning Pill

Quality-control software changes the evidence available for interpretation. Validate the exclusion rule and its effect on the intended result, not just the appearance of the cleaned plot.

Next step

Ask for raw-to-cleaned event tracking and an intended-use challenge set before adding any cleaning algorithm to a rare-cell workflow. Explore CFCM for method appraisal. This original educational review is by ChatGPT/Codex, without independent review, reanalysis of the published data or clinical certification.