Bio Flow Cytometry Cytometry Qc Agent Skill

library(flowAI)
library(flowCore)

# Load FCS file
ff <- read.FCS('sample.fcs')

# Run automated QC
# Checks: flow rate, signal stability, dynamic range
qc_result <- flow_auto_qc(
    ff,
    folder_results = 'qc_output/',
    fcs_QC = TRUE,       # Export QC'd FCS
    html_report = TRUE,  # Generate HTML report
    mini_report = TRUE   # Also make summary
)

# Get cleaned data
ff_clean <- qc_result$fcs

# QC metrics
cat('Original events:', nrow(ff), '\n')
cat('After QC:', nrow(ff_clean), '\n')
cat('Removed:', nrow(ff) - nrow(ff_clean), '(',
    round((1 - nrow(ff_clean)/nrow(ff)) * 100, 1), '%)\n')

# Check for acquisition issues via flow rate
check_flow_rate <- function(ff, time_channel = 'Time') {
    expr <- exprs(ff)
    time <- expr[, time_channel]

    # Bin by time
    n_bins <- 50
    bins <- cut(time, breaks = n_bins, labels = FALSE)

    # Events per bin (flow rate proxy)
    events_per_bin <- table(bins)
    flow_rate <- as.numeric(events_per_bin)

    # Calculate metrics
    cv <- sd(flow_rate) / mean(flow_rate) * 100

    # Detect anomalies (>2 SD from mean)
    z_scores <- abs(scale(flow_rate))
    anomalies <- which(z_scores > 2)

    list(
        mean_rate = mean(flow_rate),
        cv_percent = cv,
        anomaly_bins = anomalies,
        stable = cv < 20 && length(anomalies) < 3
    )
}

flow_qc <- check_flow_rate(ff)
cat('Flow rate CV:', round(flow_qc$cv_percent, 1), '%\n')
cat('Stable:', flow_qc$stable, '\n')

# Detect signal drift over acquisition time
detect_signal_drift <- function(ff, channels, time_channel = 'Time') {
    expr <- exprs(ff)
    time <- expr[, time_channel]

    n_bins <- 20
    bins <- cut(time, breaks = n_bins, labels = FALSE)

    drift_results <- lapply(channels, function(ch) {
        bin_medians <- tapply(expr[, ch], bins, median, na.rm = TRUE)

        # Linear trend
        trend <- lm(bin_medians ~ seq_along(bin_medians))
        slope <- coef(trend)[2]
        r_squared <- summary(trend)$r.squared

        # Percent change over acquisition
        pct_change <- (tail(bin_medians, 1) - head(bin_medians, 1)) / head(bin_medians, 1) * 100

        list(
            channel = ch,
            slope = slope,
            r_squared = r_squared,
            percent_change = pct_change,
            drift_detected = abs(pct_change) > 10 && r_squared > 0.5
        )
    })

    names(drift_results) <- channels
    drift_results
}

marker_channels <- c('CD45', 'CD3', 'CD4', 'CD8')
drift <- detect_signal_drift(ff, marker_channels)
for (ch in names(drift)) {
    if (drift[[ch]]$drift_detected) {
        cat('DRIFT DETECTED:', ch, '(', round(drift[[ch]]$percent_change, 1), '%)\n')
    }
}

# Remove events at detector saturation limits
remove_margin_events <- function(ff, channels = NULL) {
    expr <- exprs(ff)

    if (is.null(channels)) {
        channels <- colnames(expr)
    }

    # Get channel ranges from FCS parameters
    params <- parameters(ff)

    margin_mask <- rep(FALSE, nrow(expr))

    for (ch in channels) {
        if (ch %in% colnames(expr)) {
            # Get max range from parameters
            idx <- match(ch, params@data$name)
            if (!is.na(idx)) {
                max_val <- params@data$range[idx]
                # Events at max or min are margin events
                margin_mask <- margin_mask | (expr[, ch] >= max_val * 0.99) | (expr[, ch] <= 0)
            }
        }
    }

    cat('Margin events:', sum(margin_mask), '(', round(mean(margin_mask) * 100, 2), '%)\n')
    ff[!margin_mask, ]
}

ff_no_margin <- remove_margin_events(ff, c('FSC-A', 'SSC-A'))

# Exclude dead cells using viability marker
exclude_dead_cells <- function(ff, viability_channel, threshold = NULL) {
    expr <- exprs(ff)
    viability <- expr[, viability_channel]

    if (is.null(threshold)) {
        # Auto-threshold using bimodal distribution
        # Dead cells have higher viability dye uptake
        threshold <- quantile(viability, 0.9)
    }

    live_mask <- viability < threshold

    cat('Total events:', length(live_mask), '\n')
    cat('Live cells:', sum(live_mask), '(', round(mean(live_mask) * 100, 1), '%)\n')
    cat('Dead cells:', sum(!live_mask), '(', round(mean(!live_mask) * 100, 1), '%)\n')

    ff[live_mask, ]
}

# Example with zombie dye
ff_live <- exclude_dead_cells(ff, 'Zombie-Aqua')

# CyTOF-specific quality metrics
cytof_qc <- function(ff) {
    expr <- exprs(ff)

    # Event length check (cell size proxy)
    if ('Event_length' %in% colnames(expr)) {
        event_length <- expr[, 'Event_length']

        # Typical single cells: 15-45
        good_length <- event_length >= 15 & event_length <= 45
        cat('Event length filter:', sum(good_length), '/', length(good_length),
            '(', round(mean(good_length) * 100, 1), '%)\n')
    }

    # DNA intercalator check (nucleated cells)
    dna_channels <- grep('(Ir191|Ir193|DNA)', colnames(expr), value = TRUE)
    if (length(dna_channels) > 0) {
        dna_signal <- rowMeans(expr[, dna_channels, drop = FALSE])

        # Cells should have DNA signal
        has_dna <- dna_signal > quantile(dna_signal, 0.1)
        cat('DNA+ events:', sum(has_dna), '(', round(mean(has_dna) * 100, 1), '%)\n')
    }

    # Gaussian parameters (if present)
    gauss_channels <- grep('(Center|Offset|Width|Residual)', colnames(expr), value = TRUE)
    if (length(gauss_channels) > 0) {
        cat('Gaussian parameters available for additional QC\n')
    }
}

cytof_qc(ff)

library(dplyr)

# Generate QC summary for batch of files
batch_qc_summary <- function(fcs_files) {
    results <- lapply(fcs_files, function(f) {
        ff <- read.FCS(f)

        # Basic metrics
        n_events <- nrow(ff)

        # Flow rate CV
        flow_qc <- check_flow_rate(ff)

        # Signal range check
        expr <- exprs(ff)
        signal_channels <- grep('(FSC|SSC)', colnames(expr), value = TRUE, invert = TRUE)
        median_signals <- apply(expr[, signal_channels, drop = FALSE], 2, median)

        data.frame(
            file = basename(f),
            events = n_events,
            flow_rate_cv = flow_qc$cv_percent,
            flow_stable = flow_qc$stable,
            median_signal = mean(median_signals, na.rm = TRUE)
        )
    })

    summary_df <- do.call(rbind, results)

    # Flag outliers
    summary_df$outlier <- with(summary_df,
        events < median(events) * 0.5 |
        events > median(events) * 2 |
        flow_rate_cv > 30
    )

    summary_df
}

# Run batch QC
fcs_files <- list.files('data/', pattern = '\\.fcs$', full.names = TRUE)
qc_summary <- batch_qc_summary(fcs_files)
print(qc_summary)

# Flag problematic samples
cat('\nSamples with QC issues:\n')
print(qc_summary[qc_summary$outlier, ])

library(ggplot2)

# Flow rate plot
plot_flow_rate <- function(ff, time_channel = 'Time') {
    expr <- exprs(ff)
    time <- expr[, time_channel]

    n_bins <- 100
    bins <- cut(time, breaks = n_bins, labels = FALSE)
    events_per_bin <- table(bins)

    plot_data <- data.frame(
        bin = as.numeric(names(events_per_bin)),
        events = as.numeric(events_per_bin)
    )

    ggplot(plot_data, aes(x = bin, y = events)) +
        geom_line() +
        geom_smooth(method = 'loess', color = 'red', se = FALSE) +
        theme_bw() +
        labs(title = 'Flow Rate Over Acquisition', x = 'Time Bin', y = 'Events per Bin')
}

# Signal stability plot
plot_signal_stability <- function(ff, channel, time_channel = 'Time') {
    expr <- exprs(ff)

    n_bins <- 50
    bins <- cut(expr[, time_channel], breaks = n_bins, labels = FALSE)
    bin_stats <- tapply(expr[, channel], bins, function(x) {
        c(median = median(x), q25 = quantile(x, 0.25), q75 = quantile(x, 0.75))
    })

    plot_data <- data.frame(
        bin = seq_along(bin_stats),
        median = sapply(bin_stats, '[', 'median'),
        q25 = sapply(bin_stats, '[', 'q25'),
        q75 = sapply(bin_stats, '[', 'q75')
    )

    ggplot(plot_data, aes(x = bin)) +
        geom_ribbon(aes(ymin = q25, ymax = q75), alpha = 0.3) +
        geom_line(aes(y = median), color = 'blue') +
        theme_bw() +
        labs(title = paste('Signal Stability:', channel), x = 'Time Bin', y = 'Intensity')
}

# Generate QC plots
p1 <- plot_flow_rate(ff)
ggsave('qc_flow_rate.png', p1, width = 10, height = 4)

p2 <- plot_signal_stability(ff, 'CD45')
ggsave('qc_signal_stability.png', p2, width = 10, height = 4)

# Generate comprehensive QC report
generate_qc_report <- function(ff, output_file = 'qc_report.txt') {
    sink(output_file)

    cat('=== FLOW CYTOMETRY QC REPORT ===\n\n')
    cat('File:', description(ff)$`$FIL`, '\n')
    cat('Date:', description(ff)$`$DATE`, '\n')
    cat('Total events:', nrow(ff), '\n\n')

    cat('--- Flow Rate ---\n')
    flow_qc <- check_flow_rate(ff)
    cat('CV:', round(flow_qc$cv_percent, 1), '%\n')
    cat('Status:', ifelse(flow_qc$stable, 'PASS', 'FAIL'), '\n\n')

    cat('--- Signal Channels ---\n')
    expr <- exprs(ff)
    for (ch in colnames(expr)[1:min(10, ncol(expr))]) {
        cat(ch, ': median =', round(median(expr[, ch]), 1), '\n')
    }

    sink()
    cat('Report saved to', output_file, '\n')
}

generate_qc_report(ff)