Bead Normalization - Usage Guide

Overview

Bead-based normalization corrects CyTOF instrument sensitivity drift using EQ four-element beads as a physical internal standard, while cross-batch normalization (CytoNorm) harmonizes staining/acquisition batches using a shared reference sample. The key expert distinction this skill encodes: these are two different layers - beads fix within/between-run drift, CytoNorm fixes batch effects - and they are not interchangeable. It also covers why the anchor/reference sample is the design element everything depends on, why normalization is per-cluster with many quantiles, and the over-correction risk that argues for modeling batch in the statistical design rather than cleaning it out of the data.

Prerequisites

# R/Bioconductor
BiocManager::install(c('CATALYST', 'flowCore'))
# CytoNorm (cross-batch)
# remotes::install_github('saeyslab/CytoNorm')

Quick Start

Tell your AI agent what you want to do: - "Normalize my CyTOF run with EQ beads and remove the bead events" - "Harmonize my multi-batch study using my reference PBMC sample" - "Plot bead signal over acquisition time to check for drift" - "Should I normalize batch out or model it in my statistics?"

Example Prompts

EQ-bead drift correction

"Run CATALYST normCytof on this CyTOF SCE using the DVS EQ bead masses, correct the drift, remove the beads, and tell me what fraction were bead events." "Plot the Ce140 bead-channel median over Time before and after normalization to confirm the drift is gone."

Cross-batch

"Train CytoNorm on my reference sample that was run in all four batches, then normalize the experimental samples - and run testCV first so I know the per-cluster splines are safe." "One of my batches doesn't have the reference aliquot - what are my options?"

Strategy

"I'm going to do differential abundance downstream - should I CytoNorm the data or put batch in the diffcyt design? Explain the over-correction tradeoff."

What the Agent Will Do

1. Apply EQ-bead normalization first (on raw counts) to correct sensitivity drift, returning the cleaned SCE from normCytof()$data.
2. For multi-batch data, confirm an anchor/reference sample exists in every batch.
3. Run testCV() to verify FlowSOM clustering is batch-stable, then train and apply CytoNorm per-cluster.
4. Visualize bead/marker signal before and after to confirm correction without over-correction.
5. Recommend modeling batch in the diffcyt design for inference, reserving normalization for display/clustering.

Tips

• Bead normalization (drift) and CytoNorm (batch) are different layers - do bead first, on raw counts; CytoNorm after, on transformed scale.
• normCytof() returns a list - the normalized SCE is in $data; beads="dvs" is the EQ four-element mass set.
• CytoNorm requires the same anchor/reference sample in every batch; without it, that batch can't be safely normalized.
• Run testCV() first - if per-cluster CV is high the FlowSOM model is batch-unstable and splines will distort (fall back to nClus=1).
• Per-cluster + ~99 quantiles handles cell-type-specific, non-linear drift; a single global shift reintroduces distortion.
• Over-normalization can erase real biology; for inference prefer modeling batch in the design, normalizing only for visualization.
• Fully confounded batch (batch == condition) cannot be rescued by any normalization.

Related Skills

• cytometry-qc - EQ-bead-median-vs-Time is the primary CyTOF drift readout
• doublet-detection - Remove doublets before normalization
• compensation-transformation - The transform scale CytoNorm operates on
• clustering-phenotyping - Cluster across normalized batches
• differential-analysis - Model batch in the design rather than over-cleaning
• experimental-design/batch-design - Anchor/reference-sample design and batch strategy