Pseudobulk Analysis using pseudoBulkDGE()

LIBD Rstats Club: pseudobulk analysis using pseudoBulkDGE()
Presented By : Manisha Barse
Date: May 02, 2025
Recording link to our rstats club presentation: https://youtu.be/0OkWBLyLrfg

pseudoBulkDGE(): A handy wrapper for DE analysis following OSCA guidelines
#RStats #transcriptomics

Transcript

1. Pseudobulk Analysis using pseudoBulkDGE() RStats presentation Presented By: Manisha Barse May 2, 2025
2. Objectives Understand what pseudobulk analysis is and why it’s used Learn the steps of pseudoBulkDGE() following OSCA guidelines Compare registration_pseudobulk() vs aggregateAcrossCells() + pseudoBulkDGE() workflows Understand input parameters, normalization, filtering, and output interpretation
3. What is Pseudobulk Analysis? Aggregates single-cell or spatial counts into group-level profiles Treats groups as “bulk samples” → improves statistical power Suitable for differential expression analysis (DEA) Reduces false positives compared to cell-level models Maynard, K.R., Collado-Torres, L., Weber, L.M. et al. Transcriptome-scale spatial gene expression in the human dorsolateral prefrontal cortex. Nat Neurosci 24, 425–436 (2021).
4. OSCA Workflow for Pseudobulk Analysis 1. Aggregate counts → pseudobulk matrix 2. Normalize counts 3. Filter lowly expressed genes (filterByExpr) 4. Fit models (edgeR, voom) 5. Extract DE results 6. Interpret log-fold changes, p-values, FDR https://bioconductor.org/books/3.18/OSCA.multisample/multi-sample-comparisons.html#multi-sample-comparisons
5. pseudoBulkDGE() Automates pseudobulk DE pipeline. Supports edgeR, voom. Input: SummarizedExperiment, label, design, coef, condition. Handles normalization, filtering, logcounts computation, GLM or voom models.
6. Overview of pseudoBulkDGE() Workflow 1. Input: SummarizedExperiment with counts 2. Labeling: Define groups (e.g., BayesSpace domains) 3. Aggregation: Sum counts within groups → pseudobulks 4. Normalization: calcNormFactors() 5. Filtering: Remove lowly expressed genes (filterByExpr) 6. Modeling: Fit linear model, estimate dispersion 7. Testing: Compute DE statistics 8. Output: Table of DE genes, logFC, p-values
7. Normalization and Filtering Normalization: edgeR::calcNormFactors() Filtering: edgeR::filterByExpr() Logcounts: After normalization, generates logCPM/logcounts matrix.
8. Comparison to existing method • registration_pseudobulk() applies edgeR::filterByExpr() across all spots- might remove genes with overall low expression which by might be expressed in specific spatial domains. • So, for situations with limited samples size or domain-wise expression is important: pseudobulk samples using aggregateAcrossCells() and then use pseudobulkDGE().
9. psuedoBulkDGE() • Main function: Performs DEA on pseudobulked data • Important arguments: ◦ data: SummarizedExperiment ◦ label: grouping variable (e.g., BayesSpace) ◦ design: model matrix (~ diagnosis + covariates) ◦ coef: coefficient of interest (e.g., “Case”) ◦ condition: primary condition (diagnosis) ◦ row.data: gene-level metadata ◦ method: "edgeR" or "voom" (edgeR handles low counts better via its count-based model but voom supports variable sample precision when quality=TRUE)
10. R script Demo script https://github.com/manishabarse/LIBD_presentation/blob/main/pseudobulk_demo.R
11. Thank you