Compiled: 2021-02-08

Seurat Standard Worflow

The standard Seurat workflow takes raw single-cell expression data and aims to find clusters within the data. For full details, please read our tutorial. This process consists of data normalization and variable feature selection, data scaling, a PCA on variable features, construction of a shared-nearest-neighbors graph, and clustering using a modularity optimizer. Finally, we use a t-SNE to visualize our clusters in a two-dimensional space.

pbmc.counts <- Read10X(data.dir = "~/Downloads/pbmc3k/filtered_gene_bc_matrices/hg19/")
pbmc <- CreateSeuratObject(counts = pbmc.counts)
pbmc <- NormalizeData(object = pbmc)
pbmc <- FindVariableFeatures(object = pbmc)
pbmc <- ScaleData(object = pbmc)
pbmc <- RunPCA(object = pbmc)
pbmc <- FindNeighbors(object = pbmc)
pbmc <- FindClusters(object = pbmc)
pbmc <- RunTSNE(object = pbmc)
DimPlot(object = pbmc, reduction = "tsne")

Seurat Object Interaction

Since Seurat v3.0, we’ve made improvements to the Seurat object, and added new methods for user interaction. We also introduce simple functions for common tasks, like subsetting and merging, that mirror standard R functions.

# Get cell and feature names, and total numbers
colnames(x = pbmc)
Cells(object = pbmc)
rownames(x = pbmc)
ncol(x = pbmc)
nrow(x = pbmc)

# Get cell identity classes
Idents(object = pbmc)
levels(x = pbmc)

# Stash cell identity classes
pbmc[["old.ident"]] <- Idents(object = pbmc)
pbmc <- StashIdent(object = pbmc, save.name = "old.ident")

# Set identity classes
Idents(object = pbmc) <- "CD4 T cells"
Idents(object = pbmc, cells = 1:10) <- "CD4 T cells"

# Set identity classes to an existing column in meta data
Idents(object = pbmc, cells = 1:10) <- "orig.ident"
Idents(object = pbmc) <- "orig.ident"

# Rename identity classes
pbmc <- RenameIdents(object = pbmc, `CD4 T cells` = "T Helper cells")

# Subset Seurat object based on identity class, also see ?SubsetData
subset(x = pbmc, idents = "B cells")
subset(x = pbmc, idents = c("CD4 T cells", "CD8 T cells"), invert = TRUE)

# Subset on the expression level of a gene/feature
subset(x = pbmc, subset = MS4A1 > 3)

# Subset on a combination of criteria
subset(x = pbmc, subset = MS4A1 > 3 & PC1 > 5)
subset(x = pbmc, subset = MS4A1 > 3, idents = "B cells")

# Subset on a value in the object meta data
subset(x = pbmc, subset = orig.ident == "Replicate1")

# Downsample the number of cells per identity class
subset(x = pbmc, downsample = 100)

# Merge two Seurat objects
merge(x = pbmc1, y = pbmc2)
# Merge more than two Seurat objects
merge(x = pbmc1, y = list(pbmc2, pbmc3))

Data Access

Accessing data in Seurat is simple, using clearly defined accessors and setters to quickly find the data needed.

# View metadata data frame, stored in object@meta.data
pbmc[[]]

# Retrieve specific values from the metadata
pbmc$nCount_RNA
pbmc[[c("percent.mito", "nFeature_RNA")]]

# Add metadata, see ?AddMetaData
random_group_labels <- sample(x = c("g1", "g2"), size = ncol(x = pbmc), replace = TRUE)
pbmc$groups <- random_group_labels

# Retrieve or set data in an expression matrix ('counts', 'data', and 'scale.data')
GetAssayData(object = pbmc, slot = "counts")
pbmc <- SetAssayData(object = pbmc, slot = "scale.data", new.data = new.data)

# Get cell embeddings and feature loadings
Embeddings(object = pbmc, reduction = "pca")
Loadings(object = pbmc, reduction = "pca")
Loadings(object = pbmc, reduction = "pca", projected = TRUE)

# FetchData can pull anything from expression matrices, cell embeddings, or metadata
FetchData(object = pbmc, vars = c("PC_1", "percent.mito", "MS4A1"))

Visualization in Seurat

Seurat has a vast, ggplot2-based plotting library. All plotting functions will return a ggplot2 plot by default, allowing easy customization with ggplot2.

# Dimensional reduction plot for PCA or tSNE
DimPlot(object = pbmc, reduction = "tsne")
DimPlot(object = pbmc, reduction = "pca")

# Dimensional reduction plot, with cells colored by a quantitative feature
FeaturePlot(object = pbmc, features = "MS4A1")

# Scatter plot across single cells, replaces GenePlot
FeatureScatter(object = pbmc, feature1 = "MS4A1", feature2 = "PC_1")
FeatureScatter(object = pbmc, feature1 = "MS4A1", feature2 = "CD3D")

# Scatter plot across individual features, repleaces CellPlot
CellScatter(object = pbmc, cell1 = "AGTCTACTAGGGTG", cell2 = "CACAGATGGTTTCT")

VariableFeaturePlot(object = pbmc)

# Violin and Ridge plots
VlnPlot(object = pbmc, features = c("LYZ", "CCL5", "IL32"))
RidgePlot(object = pbmc, feature = c("LYZ", "CCL5", "IL32"))

# Heatmaps
DoHeatmap(object = pbmc, features = heatmap_markers)
DimHeatmap(object = pbmc, reduction = "pca", cells = 200)

# New things to try!  Note that plotting functions now return ggplot2 objects, so you can add themes, titles, and options
# onto them
VlnPlot(object = pbmc, features = "MS4A1", split.by = "groups")
DotPlot(object = pbmc, features = c("LYZ", "CCL5", "IL32"), split.by = "groups")
FeaturePlot(object = pbmc, features = c("MS4A1", "CD79A"), blend = TRUE)
DimPlot(object = pbmc) + DarkTheme()
DimPlot(object = pbmc) + labs(title = "2,700 PBMCs clustered using Seurat and viewed\non a two-dimensional tSNE")

Seurat provides many prebuilt themes that can be added to ggplot2 plots for quick customization

# Plotting helper functions work with ggplot2-based scatter plots, such as DimPlot, FeaturePlot, CellScatter, and
# FeatureScatter
plot <- DimPlot(object = pbmc) + NoLegend()

# HoverLocator replaces the former `do.hover` argument It can also show extra data throught the `information` argument,
# designed to work smoothly with FetchData
HoverLocator(plot = plot, information = FetchData(object = pbmc, vars = c("ident", "PC_1", "nFeature_RNA")))

# FeatureLocator replaces the former `do.identify`
select.cells <- FeatureLocator(plot = plot)

# Label points on a ggplot object
LabelPoints(plot = plot, points = TopCells(object = pbmc[["pca"]]), repel = TRUE)

Multi-Assay Features

With Seurat, you can easily switch between different assays at the single cell level (such as ADT counts from CITE-seq, or integrated/batch-corrected data). Most functions now take an assay parameter, but you can set a Default Assay to avoid repetitive statements.

cbmc <- CreateSeuratObject(counts = cbmc.rna)
# Add ADT data
cbmc[["ADT"]] <- CreateAssayObject(counts = cbmc.adt)
# Run analyses by specifying the assay to use
NormalizeData(object = cbmc, assay = "RNA")
NormalizeData(object = cbmc, assay = "ADT", method = "CLR")

# Retrieve and set the default assay
DefaultAssay(object = cbmc)
DefaultAssay(object = cbmc) <- "ADT"
DefaultAssay(object = cbmc)

# Pull feature expression from both assays by using keys
FetchData(object = cbmc, vars = c("rna_CD3E", "adt_CD3"))

# Plot data from multiple assays using keys
FeatureScatter(object = cbmc, feature1 = "rna_CD3E", feature2 = "adt_CD3")

Seurat v2.X vs v3.X

<colgroup style="box-sizing: border-box;"><col width="50%" style="box-sizing: border-box;"><col width="50%" style="box-sizing: border-box;"></colgroup>

<details style="box-sizing: border-box; display: block;"><summary style="box-sizing: border-box; display: list-item;">Session Info</summary></details>

<nav id="toc" data-toggle="toc" style="box-sizing: border-box; display: block;">

Contents

• Seurat Standard Worflow
• Seurat Object Interaction
• Data Access
• Visualization in Seurat
• Multi-Assay Features
• Seurat v2.X vs v3.X

Developed by Paul Hoffman, Satija Lab and Collaborators.