Pseudobulk Aggregation Workflows for Single-Cell RNA-seq Data

A quick practical overview on aggregating scRNA-seq gene expression matrices into pseudobulk profiles for robust differential expression analysis using Seurat and edgeR.

When analyzing single-cell RNA sequencing (scRNA-seq) datasets, treating individual cells as independent statistical replicates can lead to inflated false-positive rates due to cell-to-cell dependencies within the same biological sample.

Aggregating single-cell counts into pseudobulk profiles by sample and cell type allows us to leverage well-established bulk differential expression frameworks (like edgeR or DESeq2).

Key Steps in Pseudobulk Differential Expression

1. Cell Type Annotation & QC: Ensure low-quality cells and doublets are removed prior to aggregation.
2. Aggregation by Sample & Cell Type: Sum unnormalized raw counts for each gene within each combination of cell type and donor sample.
3. Statistical Testing: Model counts using negative binomial generalized linear models (GLMs).

# Load necessary libraries
suppressPackageStartupMessages({
  library(Seurat)
  library(edgeR)
  library(dplyr)
})

# Example: Aggregating raw counts in Seurat v5
# pbmc is a Seurat object with 'cell_type' and 'donor_id' in meta.data
pseudobulk_counts <- AggregateExpression(
  object = pbmc,
  group.by = c("cell_type", "donor_id"),
  slot = "counts"
)

# Inspect aggregated matrix for a specific cell type
head(pseudobulk_counts$RNA)