If your sample contains a single cell type at a single state, bulk RNA-seq works fine. If your sample is heterogeneous — and most biology is — single-cell RNA-seq (scRNA-seq) resolves what bulk averages away.
The core difference
Bulk RNA-seq pools RNA from millions of cells and reports the average expression per gene. If you have a tumor with 60% cancer cells, 30% immune infiltrate, and 10% stroma, your “differentially expressed gene” might come from any of those compartments — or a shifted ratio between them.
scRNA-seq labels each cell’s transcripts with a cell-specific barcode (typically using droplet microfluidics like 10x Genomics) so each gene’s counts are tracked back to the cell of origin. The output is a gene-by-cell matrix: every cell has its own transcriptome.
Comparison summary
| Property | Bulk RNA-seq | Single-cell RNA-seq |
|---|---|---|
| Resolution | Sample average | Per-cell |
| Detects rare cell types | No | Yes |
| Genes detected per sample | Most expressed | ~1–5,000 per cell (high dropout) |
| Statistical power per gene | High | Lower (sparse data) |
| Cost | Lower | Higher |
| Bioinformatics complexity | Established | Higher; rapidly evolving |
| Sample input | ~100 ng RNA | Live cell suspension |
When bulk RNA-seq is the right choice
- Homogeneous cell populations like sorted cell lines or FACS-purified cells
- Comparing experimental conditions where you care about average response
- Detecting low-abundance transcripts — bulk has much higher per-gene read depth
- Differential expression with high statistical power on a budget
- Splicing analysis — generally easier and more sensitive in bulk
When scRNA-seq is the right choice
- Cell type discovery in tissues
- Mapping developmental trajectories
- Tumor heterogeneity studies
- Identifying rare cell populations (stem cells, tumor-initiating cells)
- Studying tissue architecture when paired with spatial transcriptomics
Limitations of single-cell
- Dropout. Most genes are not detected in any given cell because of low capture efficiency.
- Dissociation bias. Some cell types survive enzymatic dissociation poorly.
- Cost scales quickly with cells per sample × samples.
- Bioinformatics overhead. Clustering, integration, batch correction, annotation are non-trivial.
The hybrid approach
A common strategy: bulk RNA-seq for high-powered differential expression at the condition level, scRNA-seq for resolving which cell types are responsible. Sample-multiplexed scRNA-seq (CITE-seq, hashing) pools many conditions into a single 10x run.
What about snRNA-seq?
Single-nucleus RNA-seq sequences nuclei instead of whole cells. It works on flash-frozen tissue where dissociation is impractical (brain, heart, archived samples) and avoids dissociation-induced stress responses, at the cost of capturing only nuclear RNA.
Bulk and single-cell answer different questions. Bulk gives depth and statistical power; single-cell gives resolution. If you can afford it, doing both lets you have your high-confidence DE genes and know which cells they came from.
