SingleCellForge v2

Paired pseudobulk single-cell RNA-seq analysis workflow

SingleCellForge v2 is a reproducible R/Bioconductor workflow for preprocessing, quality control, dimensionality reduction, clustering, contaminant identification, paired pseudobulk differential expression, cross-cell-type comparison and functional enrichment of single-cell RNA-sequencing data.

The workflow was demonstrated using the Kang18 stimulated-versus-control peripheral blood mononuclear cell dataset.

Project highlights

• Processed 29,065 original cells.
• Retained 24,563 QC-filtered singlet cells.
• Retained 13,306 expressed genes.
• Selected 2,000 highly variable genes.
• Performed PCA using 50 principal components.
• Used the first 30 principal components for UMAP and graph construction.
• Identified 14 Leiden clusters.
• Detected and excluded an erythroid-contaminant cluster.
• Generated 128 donor-condition-cell-type pseudobulk profiles.
• Modelled paired donors using ~ donor + condition.
• Analysed six major immune cell types.
• Identified 737 genes shared across at least two cell types.
• Identified 107 core-response genes shared across all six cell types.
• Identified 1,253 direction-specific response genes.

Main biological result

The stimulation response was dominated by interferon-associated and antiviral transcriptional programmes across all analysed immune populations.

CD14+ monocytes were the strongest overall responders, with:

• 3,185 significant genes;
• 61.56% of tested genes significant;
• 271 significant GO Biological Process and Reactome terms.

Analysed cell types

Cell type	Significant genes	Significant percentage	Enriched terms
CD14+ Monocytes	3,185	61.56%	271
CD4 T cells	1,823	28.86%	232
FCGR3A+ Monocytes	1,447	49.00%	129
B cells	898	32.49%	202
NK cells	656	28.73%	187
CD8 T cells	417	22.34%	126

Analysis workflow

1. Dataset import and metadata standardization
2. Singlet and annotation filtering
3. Haemoglobin-dominated contaminant removal
4. Gene prevalence filtering
5. Deconvolution-based normalization
6. Highly variable gene selection
7. PCA and UMAP
8. Shared-nearest-neighbour graph construction
9. Leiden clustering
10. Erythroid-contaminant identification
11. Donor-condition-cell-type pseudobulk aggregation
12. Paired edgeR quasi-likelihood differential expression
13. MDS, BCV, MD and volcano diagnostics
14. Top-gene and core-response heatmaps
15. Cross-cell-type response analysis
16. GO Biological Process and Reactome enrichment
17. Integrated response comparison
18. Automated HTML reporting

Statistical design

Differential expression was performed separately for each cell type using the paired donor model:

~ donor + condition

The stimulated-versus-control coefficient was tested using the edgeR quasi-likelihood framework.

Main report

The complete HTML report is located at:

results/v2/final_report/SingleCellForge_v2_final_report.html

Important output directories

results/v2/differential_expression/
results/v2/differential_expression/functional_enrichment/
results/v2/differential_expression/cross_celltype_comparison/
results/v2/differential_expression/final_celltype_comparison/
results/v2/final_report/
results/v2/project_audit/

Interpretation note

The integrated response score combines normalized DEG counts and enrichment counts for within-project visualization. It is not intended as a formal biological effect-size statistic.

Reproducibility

The workflow uses saved intermediate RDS objects, scripted analyses, tabular outputs and automated report generation. The final report includes the R session information used during rendering.

Author

Pratik Ramchandra Chaudhari