bio-flow-cytometry-doublet-detection

name	bio-flow-cytometry-doublet-detection
description	Detect and remove doublets from flow and mass cytometry data. Covers FSC/SSC gating and computational doublet detection methods. Use when filtering out cell aggregates before clustering or quantitative analysis.
tool_type	r
primary_tool	flowCore

Doublet Detection

FSC-A vs FSC-H Gating (Standard Method)

library(flowCore)
library(ggcyto)

# Load data
fs <- read.flowSet(list.files('data/', pattern = '\\.fcs$', full.names = TRUE))

# FSC-A vs FSC-H for doublet discrimination
# Singlets fall on diagonal, doublets have higher FSC-A for given FSC-H

# Manual rectangular gate
singlet_gate <- rectangleGate(
    filterId = 'singlets',
    'FSC-A' = c(50000, 250000),
    'FSC-H' = c(50000, 250000)
)

# Or use polygon gate for diagonal
singlet_polygon <- polygonGate(
    filterId = 'singlets',
    .gate = data.frame(
        'FSC-A' = c(50000, 250000, 250000, 50000),
        'FSC-H' = c(40000, 200000, 260000, 60000)
    )
)

# Apply gate
singlets <- Subset(fs, singlet_gate)

# Visualize
autoplot(fs[[1]], 'FSC-A', 'FSC-H') + geom_gate(singlet_gate)

Automated Singlet Gating with flowDensity

library(flowDensity)

# Automatic singlet gate
singlet_result <- flowDensity(
    fs[[1]],
    channels = c('FSC-A', 'FSC-H'),
    position = c(TRUE, TRUE),
    gates = c(NA, NA)
)

# Get gated population
singlets <- getflowFrame(singlet_result)

# Percentage singlets
pct_singlets <- nrow(singlets) / nrow(fs[[1]]) * 100
cat('Singlets:', round(pct_singlets,  

flowAI Quality Control

library(flowAI)

# flowAI performs comprehensive QC including:
# - Flow rate anomaly detection
# - Signal acquisition anomaly detection
# - Dynamic range anomaly detection

# Run flowAI
fs_qc <- flow_auto_qc(
    fs,
    folder_results = 'flowAI_results',
    fcs_QC = TRUE,
    fcs_highQ = TRUE
)

# Results include singlet detection based on flow rate stability

FSC-A/FSC-W Method (Width Parameter)

# Some instruments provide FSC-W (width) instead of FSC-H
# FSC-A = FSC-H × FSC-W
# Doublets have higher width

if ('FSC-W' %in% colnames(fs[[1]])) {
    singlet_gate_w <- rectangleGate(
        filterId = 'singlets',
        'FSC-A' = c(50000, 250000),
        'FSC-W' = c(50000, 100000)  # Lower width = singlets
    )

    singlets <- Subset(fs, singlet_gate_w)
}

Ratio-Based Doublet Detection

# Calculate FSC-A/FSC-H ratio
# Singlets have ratio close to constant (based on pulse geometry)
# Doublets have elevated ratio

calculate_fsc_ratio <- function(ff) {
    fsc_a <- exprs(ff)[, 'FSC-A']
    fsc_h <- exprs(ff)[, 'FSC-H']

    ratio <- fsc_a / (fsc_h + 1)  # Add small value to avoid division by zero
    return(ratio)
}

# Add ratio as derived parameter
for (i in 1:length(fs)) {
    ratio <- calculate_fsc_ratio(fs[[i]])
    fs[[i]] <- cbind2(fsi ratio
    colnamesfsincolfsi  



ratio_cutoff  quantileexprsfs  
singlet_gate_ratio  rectangleGatefilterId      ratio_cutoff

SSC-Based Doublet Detection

# For cell types where FSC doesn't discriminate well,
# use SSC-A vs SSC-H additionally

ssc_singlet_gate <- rectangleGate(
    filterId = 'ssc_singlets',
    'SSC-A' = c(10000, 200000),
    'SSC-H' = c(10000, 200000)
)

# Combine FSC and SSC gates
combined_gate <- singlet_gate & ssc_singlet_gate
singlets <- Subset(fs, combined_gate)

CyTOF Doublet Detection

library(CATALYST)

# For CyTOF data, use DNA channels or event length

# DNA-based doublet detection (if DNA channels present)
# Doublets have ~2x DNA content
sce <- prepData(fs, panel, md)

# If Event_length channel exists
if ('Event_length' %in% rownames(sce)) {
    event_length <- assay(sce)['Event_length', ]
    singlet_idx <- event_length < quantile(event_length, 0.95)

    sce_singlets <- sce[, singlet_idx]
    cat('Removed', sum(!singlet_idx), 'doublets based on event length\n')
}

# DNA intercalator method
if (all(c   rownamessce 
    dna_total  assaysce   assaysce 
    dna_cutoff  quantiledna_total 

    singlet_idx  dna_total  dna_cutoff
    sce_singlets  sce singlet_idx

CATALYST Workflow with Doublet Removal

library(CATALYST)

# Load and prepare data
sce <- prepData(fs, panel, md, transform = TRUE, cofactor = 5)

# Remove doublets using marker-based method
sce <- filterSCE(sce, !is_doublet(sce))

# Custom doublet detection based on FSC
fsc_a <- colData(sce)$FSC_A
fsc_h <- colData(sce)$FSC_H

# Model expected singlet relationship
fit <- lm(fsc_a ~ fsc_h)
residuals <- abs(fsc_a - predict(fit))
threshold <- quantile(residuals, 0.95)

# Mark doublets
colData(sce)$doublet <- residuals > threshold
sce_singlets  sce colDatascedoublet

cat meancolDatascedoublet    

Batch Processing

# Process all samples
detect_doublets <- function(ff, method = 'fsc') {
    if (method == 'fsc') {
        fsc_a <- exprs(ff)[, 'FSC-A']
        fsc_h <- exprs(ff)[, 'FSC-H']

        fit <- lm(fsc_a ~ fsc_h)
        residuals <- abs(fsc_a - predict(fit))
        threshold <- quantile(residuals, 0.95)

        singlet_idx <- residuals <= threshold
    } else if (method == 'ratio') {
        ratio <- exprsff   exprsff   
        singlet_idx  ratio  quantileratio 
    

    ffsinglet_idx 



fs_singlets  fsApplyfs detect_doublets method  


doublet_rates  sapplyfs i 
      nrowfs_singletsi  nrowfsi

cat meandoublet_rates    

Visualization

library(ggplot2)

# Extract data for plotting
plot_data <- data.frame(
    FSC_A = exprs(fs[[1]])[, 'FSC-A'],
    FSC_H = exprs(fs[[1]])[, 'FSC-H']
)

# Calculate doublet status
fit <- lm(FSC_A ~ FSC_H, data = plot_data)
plot_data$residual <- abs(plot_data$FSC_A - predict(fit))
plot_data$doublet <- plot_data$residual > quantile(plot_data$residual, 0.95)

# Plot
ggplot(plot_data aesx  FSC_H y  FSC_A color  doublet 
    geom_pointalpha   size   
    scale_color_manualvalues    
    theme_bw 
    labstitle   x   y  
ggsave width   height  

Related Skills

Workflow order: cytometry-qc → doublet-detection → bead-normalization → clustering

• cytometry-qc - Run first: identify flow rate and signal issues
• bead-normalization - Run after: correct remaining instrument drift
• fcs-handling - Load FCS files
• gating-analysis - Manual gating workflows
• clustering-phenotyping - Downstream analysis after doublet removal