Flow Cytometry

Flow cytometry data present two common statistical pitfalls: treating multiple wells or samples from the same donor as independent observations (pseudoreplication), and applying standard parametric tests to proportion data (compositional data). Licklider detects both and guides the researcher to valid, defensible analyses.

STEP 1 — The Scenario

Researcher Profile

An immunologist studying the effect of an immunosuppressant on regulatory T cell (Treg) frequency in peripheral blood. PBMCs from 6 healthy donors and 6 patients with autoimmune disease are stained with a Treg panel (CD4+CD25+FoxP3+) and run in duplicate wells on the cytometer.

Research Question

"Is the frequency of CD4+CD25+FoxP3+ Tregs (as % of CD4+ T cells) significantly different between healthy donors and autoimmune patients?"

Dataset

• 6 healthy donors, 6 autoimmune patients
• 2 technical replicate wells per donor
• Outcome: Treg frequency (% of CD4+ T cells), compositional (bounded 0–100%)
• Data format: DonorID, Group (Healthy/Patient), Well (1/2), Treg_pct

STEP 2 — Licklider Analysis Path

1. Data Upload & Structure Detection

Licklider detects the DonorID column with 12 unique values (6 per group) and the Well column with 2 levels. It flags: "Each donor has 2 well replicates — these are technical replicates. Statistical comparison must be performed at the donor level, not the well level."

2. Technical Replicate Averaging

The Biological vs. Technical Replicate Guard (4.5.2) requires averaging the two well replicates per donor before statistical testing. After averaging: 6 biological observations per group (n=12 total). The averaging step is logged in the Preprocessing Audit Log.

3. Compositional Data Assessment

Treg_pct is recognized as a proportion (% of CD4+) — bounded [0, 100]. Licklider flags: "Proportion data should not be analyzed with standard parametric tests without transformation." Options offered:

• Arcsine-square-root transformation: arcsin(√(x/100)) — traditional approach, stabilizes variance for proportion data
• Logit transformation: log(x/(100-x)) — appropriate when values cluster near 0% or 100%
• Typical lab option (rank-based test on untransformed data): avoids parametric assumptions — in practice often Mann–Whitney; Licklider does not currently run Mann–Whitney end-to-end in the stats engine (see Non-Parametric Alternatives)

The researcher selects the approach; Licklider applies it and runs Shapiro-Wilk to verify normality post-transformation if a parametric test is chosen.

4. Statistical Test

After averaging and transformation (e.g., arcsine): two-sample Welch's t-test (n=6 per group). A rank-based path on untransformed proportions is a common lab protocol; Licklider's verified path here is parametric-first. Effect size: Cohen's d (parametric) or, if you analyze ranks outside Licklider, rank-biserial correlation as appropriate.

5. Figure Generation

Box plot (or strip plot) at the donor level — not the well level. Each point represents one donor. Individual donor trajectories visible. Transformation applied is disclosed in the axis label if data are displayed on transformed scale, or the original scale with a note about the transformation used for testing.

STEP 3 — Guards & Disclosures Activated

Pseudoreplication Detection (4.5.1)

If the researcher attempts to analyze all 24 well observations as independent (n=12 per group instead of n=6), the pseudoreplication guard fires: "Multiple wells from the same donor are not independent. Analysis at the well level inflates n and underestimates variability. Averaging per donor is required before group comparison."

Biological vs. Technical Replicate Guard (4.5.2)

Technical replicate averaging is enforced and logged. The report discloses: "Data are presented as mean of 2 technical replicates per donor; n=6 biological donors per group."

Compositional Data Warning (4.7.2)

The percentage data are flagged as compositional (bounded). The transformation method applied and its rationale are disclosed in the Methods text. If no transformation is applied (non-parametric test chosen), this is also disclosed with the rationale.

N Disclosure (1.2.1)

n=6 biological replicates per group (not n=12 wells) is the reported n in all figure legends, Methods text, and statistical reports.

Batch / Plate Confounding (4.5.3)

If all 6 healthy donors were run on Day 1 and all 6 patients on Day 2, the batch confounding guard fires: "All samples within a group appear to have been processed on the same date. Batch effects cannot be separated from biological group effects — this should be disclosed as a limitation."

STEP 4 — Export Package

• Figure: Donor-level box + strip plot — Healthy vs. Autoimmune Treg frequency — 300 dpi TIFF/PDF
• Statistical Report: Well-level to donor-level averaging summary; transformation applied; Shapiro-Wilk post-transformation; Welch/t-test result (verified path); effect size + 95% CI; n per group (biological)
• Methods Text: "Treg frequency (% of CD4+ T cells) was measured by flow cytometry in n=6 healthy donors and n=6 autoimmune patients (2 technical replicates per donor, averaged prior to analysis). Data were arcsine-square-root transformed before statistical comparison using a two-sided Welch's t-test. Effect size is reported as Cohen's d. Data are presented at the donor level."
• Figure Legend: Each point = one donor; transformation disclosure; p-value; Cohen's d; n
• Preprocessing Audit Log: Technical replicate averaging step; transformation applied; batch pattern detected (if applicable)