Introduction to scRNA-seq & Seurat

What is single-cell RNA sequencing?

Every cell in your body carries the same genome — the same DNA sequence. But cells are wildly different: a root cell in Sorghum bicolor behaves nothing like a leaf mesophyll cell, even though they share identical DNA. The reason is gene expression: different cells switch different genes on and off.

Traditional bulk RNA sequencing (RNA-seq) measures the average gene expression across millions of cells in a tissue sample. That average destroys the identity of individual cells. Rare cell types — a stress-activated immune cell, a newly differentiating stem cell — are completely masked by the majority.

Single-cell RNA sequencing (scRNA-seq) solves this by measuring gene expression in each cell individually. You end up with a matrix where each row is a cell and each column is a gene — giving you a transcriptomic fingerprint for thousands of cells at once.

What scRNA-seq actually measures

In a typical scRNA-seq experiment:

• Cells are dissociated from tissue into a single-cell suspension
• Each cell is captured in a tiny droplet (e.g. 10x Genomics Chromium) containing a bead with a unique cell barcode
• RNA from the cell is reverse-transcribed and labelled with the barcode plus a Unique Molecular Identifier (UMI) per transcript
• Libraries are sequenced; reads are demultiplexed by barcode → each barcode = one cell
• UMI counts per gene per cell form the raw count matrix

Bulk RNA-seq vs Single-cell RNA-seq

Understanding the differences shapes every decision you make in a Seurat workflow.

Why Seurat?

Seurat is not just a collection of functions — it introduces a dedicated data structure called the Seurat Object that stores everything in one place:

• The raw count matrix (stored as a sparse matrix to save memory)
• Normalised and scaled data
• Dimensionality reduction results (PCA, UMAP)
• Cluster identities (called "idents")
• Cell-level metadata (QC metrics, sample origin, annotations)
• Gene-level metadata

By keeping everything in one object, you never lose track of which cells belong to which analysis state. This is fundamentally different from the DESeq2 workflow, where you manage separate data frames for counts, metadata, and results.

Seurat versions

Understanding 10x Genomics data formats

The most common scRNA-seq platform is the 10x Genomics Chromium system. After sequencing and alignment (using Cell Ranger), 10x produces a standard output folder containing three files:

The three files together are called the filtered feature-barcode matrix and are typically located at:

sample_output/
└── outs/
    └── filtered_feature_bc_matrix/
        ├── matrix.mtx.gz
        ├── barcodes.tsv.gz
        └── features.tsv.gz

Seurat's Read10X() function reads all three files at once and combines them into a single sparse matrix — you just point it at the folder and it does the rest.

Installing Seurat

Step 1 — Check your R version

# Check current R version — must be ≥ 4.1.0
R.version.string

# Check if Seurat is already installed
packageVersion("Seurat")

Step 2 — Install system libraries (Ubuntu)

# Required C libraries for R packages that Seurat depends on
sudo apt-get install -y \
  libhdf5-dev \
  libcurl4-openssl-dev \
  libssl-dev \
  libxml2-dev \
  libgeos-dev \
  libgdal-dev \
  libproj-dev

# libhdf5-dev → needed for reading HDF5 (.h5) 10x files
# libcurl4-openssl-dev → needed for internet downloads inside R

Step 3 — Install Seurat from CRAN

# Install Seurat v5 from CRAN
install.packages("Seurat")

# Install SeuratObject (data structure package — sometimes installed automatically)
install.packages("SeuratObject")

# Install supporting packages
install.packages(c(
  "dplyr",       # data manipulation
  "ggplot2",     # plotting (Seurat uses ggplot2 internally)
  "patchwork",   # arrange multiple plots
  "Matrix",      # sparse matrix support
  "scales"       # colour scales for plots
))

Step 4 — Install Bioconductor dependencies

# Install BiocManager if not already present
if (!requireNamespace("BiocManager", quietly = TRUE))
  install.packages("BiocManager")

# Install Bioconductor packages used by Seurat
BiocManager::install(c(
  "limma",       # differential expression testing backend
  "glmGamPoi",   # fast GLM fitting for SCTransform
  "DESeq2"       # used for pseudobulk DE in later lessons
))

First look in R — verifying your installation

Once installed, let's verify everything works and take a first look at the PBMC dataset we will use throughout this module. PBMC stands for Peripheral Blood Mononuclear Cells — a mixture of immune cells (T cells, B cells, NK cells, monocytes) that is the standard benchmark dataset for Seurat tutorials.

Load Seurat and check version

# Load Seurat
library(Seurat)
library(SeuratObject)
library(dplyr)
library(ggplot2)

# Confirm version — should be ≥ 5.0.0
packageVersion("Seurat")
# [1] '5.1.0'

# See all exported functions (over 300!)
ls("package:Seurat") |> head(20)

Download the PBMC 3k dataset

# Create a working directory for this module
mkdir -p ~/scrna-seurat-pbmc/data/raw
cd ~/scrna-seurat-pbmc/data/raw

# Download the PBMC 3k filtered feature-barcode matrix (22 MB)
wget -q \
  https://cf.10xgenomics.com/samples/cell/pbmc3k/pbmc3k_filtered_gene_bc_matrices.tar.gz

# Extract — creates pbmc3k_filtered_gene_bc_matrices/hg19/
tar -xzf pbmc3k_filtered_gene_bc_matrices.tar.gz

# Inspect the three files Seurat will read
ls -lh pbmc3k_filtered_gene_bc_matrices/filtered_gene_bc_matrices/hg19/
# -rw-r--r--  barcodes.tsv   70K
# -rw-r--r--  genes.tsv      502K
# -rw-r--r--  matrix.mtx     17M

Peek at the raw files in R

# Point to the data directory
data_dir <- "data/raw/pbmc3k_filtered_gene_bc_matrices/filtered_gene_bc_matrices/hg19"

# Read the sparse count matrix — returns a dgCMatrix
counts <- Read10X(data.dir = data_dir)

# What class is this?
class(counts)
# [1] "dgCMatrix"
attr(class(counts), "package")
# [1] "Matrix"

# Dimensions: genes × cells
dim(counts)
# [1] 32738  2700
# 32,738 genes measured in 2,700 cells

# First 5 genes, first 4 cells — most values are zero
counts[1:5, 1:4]
# 4 x 4 sparse Matrix of class "dgCMatrix"
#           AAACATACAACCAC AAACATTGAGCTAC AAACATTGATCAGC AAACCGTGCTTCCG
# AL627309.1             .              .              .              .
# AP006222.2             .              .              .              .
# RP11-206L10.2          .              .              .              .
# RP11-206L10.9          .              .              .              .
# LINC00115              .              .              .              .

# Dots (.) represent zero — the matrix is very sparse
# Calculate sparsity: % of entries that are zero
sparsity <- 1 - nnzero(counts) / prod(dim(counts))
cat(sprintf("Sparsity: %.1f%%\n", sparsity * 100))
# Sparsity: 94.3%

94.3 % of all entries in this matrix are zero. That single number explains why every decision in the Seurat workflow — normalisation method, feature selection, dimensionality reduction — is specifically designed for sparse count data.

Exercises