Chapter 16: Single-Cell Analysis Workflow: Seurat v5, SCTransform, Clustering & UMAP

1. Modern Single-Cell Workflows in Seurat v5

Seurat v5 introduces a modular Layers architecture, decoupling data storage from experimental batches to enable seamless integration across multi-modal assays (CITE-seq, spatial transcriptomics, scATAC-seq). Instead of basic log-normalization, modern single-cell analysis utilizes SCTransform: a regularized negative binomial regression that eliminates sequencing depth variation while preserving biological heterogeneity.

2. Graph-Based Community Detection: Shared Nearest Neighbors & Louvain

Unlike bulk clustering, single-cell experiments contain tens of thousands of individual points. Seurat constructs a Shared Nearest Neighbor (SNN) graph in reduced PCA space, connecting cells that share nearest neighbors. Cellular communities are then discovered by optimizing modularity using the Louvain or Leiden algorithms across user-defined resolutions (\(0.4 - 1.2\)).

3. End-to-End Seurat v5 Processing Script

library(Seurat)
library(ggplot2)

# 1. SCTransform: Modern Regularized Negative Binomial Normalization
# Replaces NormalizeData, FindVariableFeatures, and ScaleData in one step
pbmc <- SCTransform(pbmc, vars.to.regress = "percent.mt", verbose = FALSE)

# 2. Linear Dimensionality Reduction (PCA)
pbmc <- RunPCA(pbmc, verbose = FALSE)

# Determine Dimensionality via ElbowPlot
ElbowPlot(pbmc, ndims = 30)

# 3. Graph-Based Clustering (Shared Nearest Neighbor + Louvain Algorithm)
# Utilizing the top 20 principal components
pbmc <- FindNeighbors(pbmc, dims = 1:20)
pbmc <- FindClusters(pbmc, resolution = 0.6)

# 4. Non-Linear Dimensionality Reduction (UMAP)
pbmc <- RunUMAP(pbmc, dims = 1:20)

# 5. Visualizing Cellular Clusters on 2D UMAP Space
DimPlot(pbmc, reduction = "umap", label = TRUE, pt.size = 0.6) +
    labs(
        title = "Seurat v5 Single-Cell UMAP Cluster Map",
        subtitle = "Identified cellular communities across resolved PBMC subpopulations"
    ) +
    theme_minimal(base_size = 14)