Analysis
Validation & QC
flowAI and Dynamic Range: Review the Exclusion Before Trusting the Clean Plot
The original flowAI paper checks flow rate, signal stability and dynamic range. Its range checks offer a specific reason to preserve the evidence behind event removal.
A clean plot needs an exclusion history
When a preprocessing step removes events, the immediate output may look easier to interpret. A methods review should also ask which observations were removed and why. This is especially useful when an unusual cluster disappears after cleaning: its disappearance alone does not explain the acquisition problem.
What the original flowAI paper describes
Monaco and colleagues' 2016 Bioinformatics article presents automatic and interactive quality-control tools. The workflow checks flow rate, signal stability over time and dynamic range. Its lower-range procedure identifies negative outliers using a robust threshold; it is not a rule to discard every negative value. The upper-range check addresses margin events. A report records the affected events and parameters.
The evaluation covers 4469 files from 11 datasets. Examples show how acquisition anomalies can affect downstream clustering and dimensionality reduction. In a B-cell example, an apparent population involved CD19 values at the upper dynamic-range margin. These demonstrations motivate careful preprocessing review; they do not establish diagnostic sensitivity or a universal rejection threshold for another assay.
Review the reason, not just the retained percentage
For a local audit, choose one file and preserve its original form, the actual preprocessing configuration and the event-selection result. Ask the reviewer to explain a removed category using the diagnostic output. A single retained percentage is a useful summary, but it cannot replace the explanation of what happened in a particular parameter.
Write down whether the team accepted the proposed exclusion, changed a setting or retained the file for further investigation. If the software and reviewer disagree, keep that disagreement visible. This is a proposed review practice, not a claim that manual intervention automatically produces the correct answer.
Keep historical and local evidence separate
This article reads the original publication. It does not assert that a current release has the same defaults, or that its published examples validate a local clinical panel. Verify the implementation actually used and assess its suitability for the intended application. Where an exclusion changes a clinically important interpretation, the laboratory needs evidence appropriate to that decision.
Daily Learning Pill
A negative outlier rule is not the same statement as 'negative means invalid'. Locate the actual criterion before explaining an exclusion.
Discuss one preprocessing record with your team using the CFCM resources. Original ChatGPT/Codex educational analysis; not independent scientific review or clinical validation.