Bio Flow Cytometry Clustering Phenotyping Agent Skill

Version Compatibility

Reference examples tested with: FlowSOM 2.10+, scanpy 1.10+

Before using code patterns, verify installed versions match. If versions differ:

• R: packageVersion('<pkg>') then ?function_name to verify parameters

If code throws ImportError, AttributeError, or TypeError, introspect the installed

package and adapt the example to match the actual API rather than retrying.

Clustering and Phenotyping

"Cluster my cytometry data to find cell types" → Discover cell populations in high-dimensional flow/mass cytometry data using unsupervised clustering without predefined gates.

• R: FlowSOM::FlowSOM() for self-organizing map clustering
• R: CATALYST::cluster() with Phenograph or FlowSOM

FlowSOM Clustering

Goal: Cluster cytometry events into cell populations using self-organizing maps.

Approach: Build a FlowSOM grid on marker channels, then extract metacluster assignments per cell.

library(FlowSOM)

# Prepare data
expr <- exprs(fcs)
marker_cols <- grep('CD|HLA', colnames(fcs), value = TRUE)

# Build SOM
fsom <- FlowSOM(fcs,
                colsToUse = marker_cols,
                xdim = 10, ydim = 10,
                nClus = 20,
                seed = 42)

# Get cluster assignments
clusters <- GetMetaclusters(fsom)

# Add to flowFrame
exprs(fcs) <- cbind(exprs(fcs), cluster = clusters)

CATALYST Workflow (Full Pipeline)

Goal: Run the complete CATALYST clustering pipeline from flowSet to annotated cell populations.

Approach: Convert flowSet to SingleCellExperiment with prepData, then cluster on type markers with FlowSOM via CATALYST.

library(CATALYST)
library(SingleCellExperiment)

# Create SCE from flowSet
sce <- prepData(fs, panel, md, transform = TRUE, cofactor = 5)

# Clustering
sce <- cluster(sce,
               features = 'type',  # Use 'type' markers from panel
               xdim = 10, ydim = 10,
               maxK = 20,
               seed = 42)

# View cluster assignments
table(cluster_ids(sce, 'meta20'))

Phenograph Clustering

Goal: Identify cell populations using graph-based community detection on marker expression.

Approach: Build a k-nearest-neighbor graph on type markers, then partition with Louvain community detection via Rphenograph.

library(Rphenograph)

# Extract expression matrix
expr <- assay(sce, 'exprs')

# Run Phenograph
pheno_result <- Rphenograph(t(expr[rowData(sce)$marker_class == 'type', ]), k = 30)

# Get clusters
sce$phenograph <- factor(membership(pheno_result[[2]]))

Dimensionality Reduction

Goal: Project high-dimensional cytometry data into 2D for visualization of cell populations.

Approach: Run UMAP or tSNE on type marker channels using CATALYST's runDR wrapper, then plot colored by cluster.

# UMAP
sce <- runDR(sce, dr = 'UMAP', features = 'type')

# tSNE
sce <- runDR(sce, dr = 'TSNE', features = 'type')

# Plot
plotDR(sce, 'UMAP', color_by = 'meta20')

Cluster Annotation

Goal: Assign cell type labels to clusters based on marker expression profiles.

Approach: Visualize median marker expression per cluster with a heatmap, then map cluster IDs to cell type names.

# Heatmap of marker expression by cluster
plotExprHeatmap(sce, features = 'type',
                by = 'cluster_id', k = 'meta20',
                scale = 'first', row_anno = FALSE)

# Manual annotation
cluster_annotation <- c(
    '1' = 'CD4 T cells',
    '2' = 'CD8 T cells',
    '3' = 'B cells',
    '4' = 'NK cells',
    '5' = 'Monocytes'
)

sce$cell_type <- cluster_annotation[as.character(cluster_ids(sce, 'meta20'))]

Cluster Merging

Goal: Reduce overclustering by merging similar clusters into biologically meaningful groups.

Approach: Define a mapping table from original to merged cluster IDs, then apply with CATALYST's mergeClusters.

# Merge similar clusters
merging_table <- data.frame(
    original = 1:20,
    merged = c(1, 1, 2, 2, 3, 3, 4, 4, 5, 5,
               6, 6, 7, 7, 8, 8, 9, 9, 10, 10)
)

sce <- mergeClusters(sce, k = 'meta20', table = merging_table, id = 'merged')

Abundance Analysis (per sample)

Goal: Quantify the relative frequency of each cell population across samples and conditions.

Approach: Cross-tabulate cluster assignments by sample ID, convert to proportions, and plot grouped by condition.

# Cluster frequencies per sample
abundances <- table(cluster_ids(sce, 'meta20'), sce$sample_id)
freq <- prop.table(abundances, margin = 2)

# Plot
plotAbundances(sce, k = 'meta20', by = 'cluster_id', group_by = 'condition')

Marker Expression Summary

Goal: Summarize and compare marker expression levels across clusters and conditions.

Approach: Plot per-cluster median expression with CATALYST's plotClusterExprs and pseudo-bulk expression faceted by cluster.

# Median expression per cluster
plotClusterExprs(sce, k = 'meta20', features = 'type')

# Expression by cluster and condition
plotPbExprs(sce, k = 'meta20', features = 'type', facet_by = 'cluster_id')

Export Results

Goal: Save clustering results and annotated SCE object for downstream analysis or sharing.

Approach: Extract cluster assignments into colData, export as CSV, and serialize the full SCE as RDS.

# Add cluster info to metadata
colData(sce)$cluster <- cluster_ids(sce, 'meta20')

# Export to CSV
results <- as.data.frame(colData(sce))
write.csv(results, 'clustering_results.csv', row.names = FALSE)

# Save SCE
saveRDS(sce, 'sce_clustered.rds')

Choosing Number of Clusters

Goal: Determine the optimal number of metaclusters for the dataset.

Approach: Compare normalized reduction stability (NRS) plots and heatmaps at different K values to find where clusters remain distinct.

# Delta area plot
plotNRS(sce, features = 'type')

# Or visual inspection of heatmap at different K
plotExprHeatmap(sce, features = 'type', by = 'cluster_id', k = 'meta10')
plotExprHeatmap(sce, features = 'type', by = 'cluster_id', k = 'meta20')

Batch Integration

Goal: Remove batch effects from cytometry data before or after clustering.

Approach: Detect batch effects by coloring UMAP by batch variable, then apply MNN correction with batchelor if needed.

# If batch effects present
library(batchelor)

sce <- runDR(sce, dr = 'UMAP', features = 'type')

# Check for batch effects
plotDR(sce, 'UMAP', color_by = 'batch')

# MNN correction if needed
sce_corrected <- fastMNN(sce, batch = sce$batch)

Related Skills

• gating-analysis - Manual alternative
• differential-analysis - Compare clusters between conditions
• single-cell/clustering - Similar concepts for scRNA-seq