Seurat-Local Skill

Comprehensive assistance with Seurat v5 for single-cell RNA-seq analysis, generated from official documentation.

When to Use This Skill

This skill should be triggered when:

Core Seurat Analysis:

• Working with single-cell RNA-seq data analysis in R
• Setting up Seurat objects and performing quality control
• Normalizing data and finding variable features
• Running dimensional reduction (PCA, UMAP, t-SNE)
• Performing clustering and cell type identification

Data Integration:

• Integrating multiple single-cell datasets
• Using FindIntegrationAnchors and IntegrateData workflows
• Performing batch correction across experiments
• Working with both CCA and RPCA integration methods
• Transferring cell type annotations between datasets

Advanced Analysis:

• Differential expression testing and marker gene identification
• Spatial transcriptomics analysis with 10x Visium data
• Trajectory inference and lineage analysis
• Multi-modal data integration (e.g., scRNA-seq + scATAC-seq)
• Pseudobulk analysis and aggregated expression calculations

Seurat v5 Specific Features:

• Working with the new layered assay structure
• Using SCTransform v2 for normalization
• Performing integration in low-dimensional space
• Using the streamlined IntegrateLayers workflow
• Working with on-disk matrices for large datasets

Common Use Cases:

• "How do I integrate two scRNA-seq datasets?"
• "Help me find marker genes for my clusters"
• "How do I normalize my single-cell data?"
• "What's the difference between CCA and RPCA integration?"
• "How do I analyze spatial transcriptomics data?"

Quick Reference

Essential Seurat Workflow

Basic Setup and Normalization

library(Seurat)
library(SeuratData)

# Load example dataset
InstallData("pbmc3k")
pbmc <- LoadData("pbmc3k")

# Basic preprocessing
pbmc <- NormalizeData(pbmc, verbose = FALSE)
pbmc <- FindVariableFeatures(pbmc, selection.method = "vst", nfeatures = 2000)

Dimensional Reduction and Clustering

# Run PCA and determine dimensions
pbmc <- RunPCA(pbmc, verbose = FALSE)
pbmc <- FindNeighbors(pbmc, dims = 1:10)
pbmc <- FindClusters(pbmc, resolution = 0.5)

# Run UMAP for visualization
pbmc <- RunUMAP(pbmc, dims = 1:10)
DimPlot(pbmc, reduction = "umap")

Differential Expression

# Find markers for cluster 1
cluster1.markers <- FindMarkers(pbmc, ident.1 = 1, min.pct = 0.25)
head(cluster1.markers)

# Find all markers
pbmc.markers <- FindAllMarkers(pbmc, only.pos = TRUE, min.pct = 0.25)

Data Integration Workflows

Standard Integration (CCA-based)

# Split dataset by condition
ifnb.list <- SplitObject(ifnb, split.by = "stim")

# Normalize and find variable features
ifnb.list <- lapply(ifnb.list, function(x) {
    x <- NormalizeData(x)
    x <- FindVariableFeatures(x, selection.method = "vst", nfeatures = 2000)
})

# Find integration anchors
immune.anchors <- FindIntegrationAnchors(object.list = ifnb.list, dims = 1:20)

# Integrate data
immune.combined <- IntegrateData(anchorset = immune.anchors, dims = 1:20)

Fast Integration (RPCA-based)

# Preprocess with PCA individually
features <- SelectIntegrationFeatures(object.list = ifnb.list)
ifnb.list <- lapply(ifnb.list, function(x) {
    x <- ScaleData(x, features = features, verbose = FALSE)
    x <- RunPCA(x, features = features, verbose = FALSE)
})

# Find anchors using RPCA (faster, more conservative)
immune.anchors <- FindIntegrationAnchors(
    object.list = ifnb.list,
    anchor.features = features,
    reduction = "rpca"
)

SCTransform Integration

# Normalize with SCTransform
ifnb.list <- lapply(ifnb.list, SCTransform, method = "glmGamPoi")
features <- SelectIntegrationFeatures(object.list = ifnb.list, nfeatures = 3000)

# Prepare for integration
ifnb.list <- PrepSCTIntegration(object.list = ifnb.list, anchor.features = features)
ifnb.list <- lapply(ifnb.list, RunPCA, features = features)

# Integrate with SCTransform
immune.anchors <- FindIntegrationAnchors(
    object.list = ifnb.list,
    normalization.method = "SCT",
    anchor.features = features,
    reduction = "rpca"
)
immune.combined.sct <- IntegrateData(
    anchorset = immune.anchors,
    normalization.method = "SCT"
)

Spatial Transcriptomics

Load and Process 10x Visium Data

library(Seurat)
library(SeuratData)
library(ggplot2)

# Load spatial dataset
InstallData("stxBrain")
brain <- LoadData("stxBrain", type = "anterior1")

# Normalize with SCTransform (recommended for spatial data)
brain <- SCTransform(brain, assay = "Spatial", verbose = FALSE)

# Run dimensional reduction and clustering
brain <- RunPCA(brain, assay = "SCT", verbose = FALSE)
brain <- FindNeighbors(brain, reduction = "pca", dims = 1:30)
brain <- FindClusters(brain, verbose = FALSE)
brain <- RunUMAP(brain, reduction = "pca", dims = 1:30)

Spatial Visualization

# Visualize clusters on tissue
SpatialDimPlot(brain, label = TRUE, label.size = 3)

# Visualize gene expression
SpatialFeaturePlot(brain, features = c("Hpca", "Ttr"))

# Adjust visualization parameters
SpatialFeaturePlot(brain, features = "Ttr", alpha = c(0.1, 1))

Cell Type Label Transfer

# Load single-cell reference
allen_reference <- readRDS("allen_cortex.rds")
allen_reference <- SCTransform(allen_reference, ncells = 3000, verbose = FALSE)

# Find transfer anchors
anchors <- FindTransferAnchors(
    reference = allen_reference,
    query = cortex,
    normalization.method = "SCT"
)

# Transfer cell type labels
predictions.assay <- TransferData(
    anchorset = anchors,
    refdata = allen_reference$subclass,
    prediction.assay = TRUE
)
cortex[["predictions"]] <- predictions.assay

Advanced Features

Module Scoring

# Define gene sets
cd_features <- list(c('CD79B', 'CD79A', 'CD19', 'CD180', 'CD200',
                     'CD3D', 'CD2', 'CD3E', 'CD7', 'CD8A'))

# Calculate module scores
pbmc <- AddModuleScore(
    object = pbmc,
    features = cd_features,
    name = 'CD_Features'
)

Project UMAP Coordinates

# Project new data onto existing UMAP
query_umap <- ProjectUMAP(
    query = new_data,
    reference = reference_data,
    query.reduction = "pca",
    reference.reduction = "pca",
    reduction.model = reference_data[["umap"]]
)

Integrate Layers (Seurat v5)

# Integrate multiple layers in a single object
integrated_object <- IntegrateLayers(
    object = seurat_object,
    method = RPCAIntegration,
    orig.reduction = "pca",
    assay = "RNA",
    features = variable_features
)

Key Concepts

Core Seurat Objects

• Seurat Object: Main data structure containing assays, metadata, and reductions
• Assay: Contains different data layers (counts, data, scale.data)
• Layers: In Seurat v5, data is stored in layers instead of slots
• DimReduc: Dimensional reduction objects (PCA, UMAP, etc.)

Data Integration Methods

• CCA (Canonical Correlation Analysis): Traditional method, good for strong batch effects
• RPCA (Reciprocal PCA): Faster, more conservative, less over-correction
• SCTransform: Regularized negative binomial normalization, recommended for most analyses

Spatial Analysis

• Spots: Spatial measurement locations (50μm for 10x Visium)
• Images: Tissue histology images stored in object
• Coordinate Systems: Mapping between spots and image coordinates

Reference Files

This skill includes comprehensive documentation in references/:

announcements.md

• Seurat v5 release notes and changes
• Backwards compatibility information
• New feature descriptions and migration guide

api.md

• Complete function reference with parameters
• Detailed examples for each function
• Performance notes and best practices
• Functions covered:
  • ProjectUMAP() - Query dataset projection
  • IntegrateLayers() - Multi-layer integration
  • FastRowScale() - Efficient matrix operations
  • AddModuleScore() - Gene set scoring
  • CellSelector() - Interactive cell selection
  • TransferData() - Cross-dataset data transfer

other.md

• Code of conduct and community guidelines
• Advanced tutorials and case studies
• Integration workflows for complex scenarios
• Comprehensive PBMC stimulation analysis example

tutorials.md

• Step-by-step guides for common workflows
• RPCA vs CCA integration comparison
• Spatial transcriptomics analysis tutorials
• SCTransform normalization workflows
• Performance optimization tips

Working with This Skill

For Beginners

1. Start with the basic workflow: Load data → Normalize → Find variable features → PCA → Cluster → UMAP
2. Use the getting_started reference for fundamental concepts
3. Follow the basic examples in the Quick Reference section
4. Practice with the pbmc3k dataset which is included in SeuratData

For Intermediate Users

1. Explore integration methods when working with multiple datasets
2. Use spatial analysis features for spatial transcriptomics data
3. Leverage the API reference for advanced function parameters
4. Study the PBMC stimulation tutorial for comparative analysis

For Advanced Users

1. Use the tutorials reference for complex workflows
2. Optimize performance with RPCA integration and SCTransform
3. Implement custom analysis pipelines using the function reference
4. Contribute to the community following the code of conduct

Navigation Tips

• Search for specific functions in the api.md reference
• Find complete workflows in tutorials.md
• Check announcements.md for the latest Seurat v5 features
• Use other.md for specialized use cases and community guidelines

Resources

references/

Organized documentation extracted from official sources containing:

• Detailed function explanations with all parameters
• Real code examples with proper syntax highlighting
• Performance notes and best practices
• Links to original documentation for further reading

scripts/

Add helper scripts here for common automation tasks such as:

• Batch processing of multiple datasets
• Custom visualization functions
• Quality control automation
• Report generation

assets/

Store templates, boilerplate, and example projects:

• Example Seurat objects for testing
• Custom gene sets for module scoring
• Reference datasets for integration testing
• Visualization themes and templates

Notes

• Seurat v5 is now the default on CRAN with backwards compatibility
• New assay structure uses layers instead of slots for better data organization
• SCTransform v2 includes regularization improvements and glmGamPoi support
• Integration workflows are now streamlined and more memory-efficient
• Spatial analysis is fully integrated with enhanced visualization options

Updating

To refresh this skill with updated documentation:

1. Re-run the documentation scraper with the same configuration
2. The skill will be rebuilt with the latest Seurat documentation
3. All examples and references will be updated automatically
4. Quick reference patterns will be refreshed with new best practices