Chapter 18: Automated Cell Type Annotation (SingleR, Azimuth, CellTypist) & Functional Analysis

1. Automated Reference-Based Cell Label Transfer

Manual marker inspection is subjective, time-consuming, and prone to investigator bias. Automated reference annotation tools map single-cell query profiles against expertly curated, high-depth reference cell atlases:

2. Complete SingleR Automated Annotation Script

if (!requireNamespace("SingleR", quietly = TRUE)) BiocManager::install("SingleR")
if (!requireNamespace("celldex", quietly = TRUE)) BiocManager::install("celldex")

library(SingleR)
library(celldex)
library(Seurat)

# 1. Download Curated Reference Transcriptome (Human Primary Cell Atlas)
hpca_ref <- celldex::HumanPrimaryCellAtlasData()

# 2. Extract Single-Cell Test Matrix (SCT or RNA counts)
query_counts <- GetAssayData(pbmc, assay = "RNA", slot = "counts")

# 3. Execute SingleR Automated Annotation
# Computes correlation per cell and iteratively prunes ambiguous cell types
predictions <- SingleR(
    test = query_counts,
    ref = hpca_ref,
    labels = hpca_ref$label.main
)

# 4. Integrate Predicted Labels into Seurat Object
pbmc$SingleR_Label <- predictions$labels

# Diagnostic: Score heatmap checking classification confidence
plotScoreHeatmap(predictions)

# 5. Visualizing Final Annotated Single-Cell UMAP
DimPlot(pbmc, group.by = "SingleR_Label", reduction = "umap", label = TRUE, repel = TRUE) +
    labs(
        title = "Final Annotated Single-Cell Immune Atlas",
        subtitle = "Automated reference label transfer via SingleR & Human Primary Cell Atlas"
    ) +
    theme_minimal(base_size = 14)

3. Downstream Single-Cell Functional Frontiers