Key Points

Introduction to Single-Cell RNA-Seq

Single-cell RNA-seq measures gene expression in individual cells, revealing cellular heterogeneity that bulk RNA-seq averages out
The 10x Chromium platform uses droplet-based barcoding: each cell gets a unique 16 bp cell barcode and each mRNA molecule gets a 12 bp UMI
The UMI count matrix is very sparse (~95% zeros) because each cell expresses only a fraction of all genes and capture efficiency is limited
The scRNA-seq analysis pipeline progresses through alignment, QC, normalization, dimensionality reduction, clustering, annotation, integration, and differential expression
This workshop focuses on 10x Chromium 3’ GEX data analyzed with Seurat v5 in R on the Purdue Negishi HPC cluster

Workshop data should be organized under ${RCAC_SCRATCH}/scrna_workshop/ with separate directories for FASTQ files, reference, and outputs
Cell Ranger count aligns 10x FASTQ files and produces a filtered count matrix, a web summary report, and alignment metrics
STARsolo replicates Cell Ranger’s pipeline but runs approximately 9x faster and is fully open source
Both tools produce a filtered count matrix in the same format: barcodes.tsv, features.tsv, and matrix.mtx
Always inspect the Cell Ranger web_summary.html to check estimated cell count, median genes per cell, sequencing saturation, and fraction of reads in cells before proceeding

Read10X() loads a Cell Ranger count matrix and CreateSeuratObject() creates the central data structure for Seurat analysis
The three key QC metrics are nFeature_RNA (genes per cell), nCount_RNA (UMIs per cell), and percent.mt (mitochondrial percentage)
High mitochondrial percentage indicates damaged cells that are losing cytoplasmic RNA through a leaky membrane
Filtering thresholds should be chosen by inspecting the data distributions, not by applying universal fixed cutoffs
Dedicated doublet detection tools (scDblFinder, DoubletFinder) are available for high-loading experiments but are not always necessary at standard loading rates

Log normalization corrects for sequencing depth by dividing by total counts, scaling, and log-transforming each cell independently
Highly variable features are the ~2,000 genes with the most cell-to-cell variation, capturing biological differences while excluding housekeeping genes and noise
Scaling (z-score transformation) centers and standardizes gene expression so that PCA is not dominated by highly expressed genes
Regressing out confounders like percent.mt during scaling is optional and should only be done when technical variation is obscuring biology
SCTransform is a one-step alternative that uses regularized negative binomial regression and is recommended for publication-quality analyses

PCA reduces ~2,000 variable genes to ~15 principal components that capture the major axes of biological variation
The elbow plot helps choose how many PCs to retain; for PBMCs, 15 PCs is typically sufficient
UMAP provides an intuitive 2D visualization but should not be used to infer distances, cluster sizes, or quantitative relationships
Graph-based clustering builds a shared nearest neighbor graph in PCA space and detects communities using the Louvain or Leiden algorithm
The resolution parameter controls clustering granularity; clustree helps identify the resolution where clusters begin to split unstably

FindAllMarkers identifies genes upregulated in each cluster versus all others; key output columns are avg_log2FC, pct.1, pct.2, and p_val_adj
Manual annotation maps cluster marker genes to known cell-type-specific markers and requires domain expertise
SingleR automates annotation by comparing each cell’s expression profile to a labeled reference dataset
Combining manual and automated approaches increases confidence; agreement between methods strongly supports the assigned label
Annotation is iterative and should be cross-validated with multiple references and marker gene visualization

Batch effects cause cells to cluster by sample rather than by cell type, preventing cross-condition comparisons
CCA integration finds shared correlation structure between conditions and aligns matching cell types using anchor pairs
In Seurat v5, multi-sample data is handled by splitting assay layers, integrating, then rejoining layers for downstream analysis
Always verify integration by checking that conditions mix within clusters AND that known biological differences (like ISG expression) are preserved
Integration is not always appropriate; skip it when conditions have no shared cell types or when batch effects are minimal

Per-cell DE tests (Wilcoxon, MAST) treat each cell as independent, inflating p-values through pseudoreplication
Pseudobulk analysis aggregates cells by sample and cell type, providing statistically valid results with proper type I error control
Volcano plots visualize the relationship between effect size (log2FC) and significance (-log10 p-value) for all tested genes
Gene ontology enrichment connects lists of DE genes to biological pathways, providing functional interpretation of differential expression results
Always export session info, save intermediate objects, and document parameter choices for reproducibility