What Is Single-Cell RNA Sequencing and How Does It Work?

Single-cell RNA sequencing, usually abbreviated scRNA-seq, measures which genes are active in individual cells rather than averaging the signal across millions of them. That shift from bulk to single-cell resolution has reshaped how researchers study tissues, tumors, immune responses, and embryonic development, because two cells sitting side by side can be running entirely different genetic programs. The technology has matured rapidly since its first demonstrations in the early 2010s, branching into multiple platform designs, spawning a rich ecosystem of computational tools, and generating datasets that now routinely contain hundreds of thousands of cells from a single experiment.

How Cells Get Their Barcodes

The core challenge in scRNA-seq is deceptively simple: you need to figure out which RNA molecules came from which cell. In a bulk experiment, you grind up a tissue, extract all the RNA, and sequence it together. In single-cell work, every cell’s RNA has to be tagged with a unique molecular label before everything gets pooled. The dominant strategy today uses tiny droplets as reaction vessels. A suspension of cells flows through a microfluidic chip alongside a stream of hydrogel beads, each carrying millions of copies of a unique DNA barcode. When a single cell and a single bead land in the same nanoliter-sized droplet, the cell is lysed, its messenger RNA is released, and the bead’s barcoded primers grab hold of those transcripts during a reverse-transcription reaction.

1Nature Protocols. Single-cell barcoding and sequencing using droplet microfluidics

After barcoding, all droplets are broken open and the complementary DNA from every cell is combined into one sequencing library. When the sequencer reads a fragment, the barcode tells you which cell it belonged to, and the transcript sequence tells you which gene was active. The inDrop platform, one of the pioneering droplet systems, used hydrogel microspheres each carrying around a billion copies of one of more than 147,000 distinct barcodes, making the chance of two cells sharing the same tag negligibly small.

2Cell. Droplet Barcoding for Single-Cell Transcriptomics Applied to Embryonic Stem Cells

Platform Trade-Offs and Costs

Three droplet-based platforms have dominated the field: 10X Genomics Chromium, Drop-seq, and inDrop. They share the same basic logic but differ in sensitivity, noise levels, and price. A direct comparison found that 10X Chromium generally captures more transcripts per cell and produces less technical noise, but the instrument alone costs upward of $50,000, and each cell runs about $0.50 in reagent costs before sequencing. Drop-seq, an open-source system, costs roughly $0.10 per cell with modest trade-offs in sensitivity. inDrop falls somewhere in between, with per-cell costs about half those of 10X.

3Molecular Cell. Comparative Analysis and Refinement of Drop-seq, inDrop, and 10X Genomics Chromium Platforms for Single-Cell RNA Sequencing

Those per-cell costs add up quickly when you want to profile tens or hundreds of thousands of cells from multiple samples. Beyond the droplet platforms, full-length transcript methods like Smart-seq3 capture more of each gene’s sequence, which matters if you care about splice variants or want to distinguish closely related gene family members. A benchmarking study found that Smart-seq3 delivered the highest gene detection per cell at the lowest price among the protocols compared, while commercial kits from companies like Takara matched that gene detection with better reproducibility but at a much steeper cost.

4PubMed Central. Benchmarking full-length transcript single cell mRNA sequencing protocols

The practical upshot: researchers choose their platform based on what question they are asking. If the goal is profiling a huge number of cells to find rare populations, a high-throughput droplet method makes sense despite its shallower read depth per cell. If the goal is deep characterization of a smaller number of cells, a full-length protocol provides richer data per cell.

When Whole Cells Are Not an Option

Standard scRNA-seq requires dissociating tissue into a suspension of intact single cells, and that is not always possible. Brain tissue, for instance, is full of neurons with long, fragile projections that shatter during enzymatic digestion. Large cells, tightly interconnected cells, and cells embedded in dense extracellular matrix all resist gentle isolation. Single-nucleus RNA sequencing, or snRNA-seq, sidesteps this problem by extracting nuclei from cells instead of keeping cells whole.

5Journal of Pathology and Translational Medicine. Perspectives on single-nucleus RNA sequencing in different cell types and tissues

The nucleus contains a representative sample of the cell’s active transcripts, and importantly, nuclei can be isolated from frozen archived tissue. That is a major advantage for human studies, where most biobanked samples are flash-frozen at the time of surgery or autopsy. Work on the human brain has shown that snRNA-seq on frozen postmortem tissue performs nearly on par with whole-cell sequencing from fresh samples in terms of transcript coverage and read length.

6Nature Biotechnology. Single-nuclei isoform RNA sequencing unlocks barcoded exon connectivity in frozen brain tissue

The trade-off is that nuclear RNA captures only a subset of the transcriptome. Cytoplasmic transcripts, particularly those involved in fast signaling responses, can be underrepresented. For many tissue types, though, the ability to work with frozen samples and difficult-to-dissociate cells makes snRNA-seq the more practical choice.

7PubMed Central. Comparative analysis of nuclei isolation methods for brain single-nucleus RNA sequencing

The Dissociation Problem

One of the less glamorous but genuinely important issues in single-cell work is that the act of preparing cells for sequencing changes the cells themselves. Chopping tissue into a single-cell suspension typically involves enzymes and warm incubation, and cells respond to that stress by switching on immediate-early genes and heat shock proteins. A systematic study of kidney tissue found that warm dissociation induced higher expression of 71 genes compared to cold dissociation, with the top hits being classic stress-response genes like Fos, Jun, and heat shock proteins Hspa1a and Hspa1b. Gene ontology analysis flagged “regulation of cell death” as the most enriched biological process among the upregulated genes.

8PubMed Central. Systematic assessment of tissue dissociation and storage biases in single-cell and single-nucleus RNA-seq workflows

This is not a minor footnote. If you are studying inflammation or apoptosis in a tumor, and your sample preparation protocol itself activates inflammatory and apoptotic gene programs, you risk confusing artifact with biology. Researchers have developed metabolic labeling strategies that tag only newly synthesized RNA during the dissociation window, letting them computationally separate genuine biological signal from prep-induced noise.

9PubMed Central. A single-cell RNA labeling strategy for measuring stress response upon tissue dissociation

Cleaning Up the Data

Raw scRNA-seq data is messy. Some barcodes correspond to empty droplets that captured ambient RNA floating in solution rather than a cell. Others correspond to doublets, where two cells landed in the same droplet. Dying cells leak their RNA, contaminating the counts of healthy neighbors. And many genuine transcripts simply fail to be captured, producing an abundance of zero counts in the data matrix that may or may not reflect true biological absence.

Quality control starts with filtering out low-quality cells. A common heuristic is to flag cells with an unusually high proportion of reads mapping to mitochondrial genes, because damaged cells leak cytoplasmic RNA while retaining mitochondrial transcripts. Adaptive, data-driven approaches jointly model mitochondrial read fraction and the number of detected genes to set thresholds that fit each dataset rather than applying arbitrary cutoffs.

10PLoS Computational Biology. miQC: An adaptive probabilistic framework for quality control of single-cell RNA-sequencing data

Doublet detection has its own suite of tools. Computational pipelines now run multiple doublet-calling algorithms in parallel and flag cells identified as doublets by more than one method, since no single algorithm catches everything. Ambient RNA contamination can similarly be estimated and subtracted from each cell’s expression profile.

11Nature Communications. Comprehensive generation, visualization, and reporting of quality control metrics for single-cell RNA sequencing data

Normalization, the step that makes expression levels comparable across cells with different sequencing depths, presents its own challenges. Because single-cell data contains far more zeros than bulk RNA-seq, standard normalization methods can break down. One influential approach pools cells together, computes size factors from the pooled counts, and then deconvolves those back into cell-level factors, sidestepping the sparsity problem.

12PubMed Central. Pooling across cells to normalize single-cell RNA sequencing data with many zero counts

The zero-inflation issue extends to differential expression testing. Many genes appear to be “off” in a given cell simply because the transcript was not captured, not because the gene is biologically silent. A weighting strategy that models these excess zeros and assigns gene- and cell-specific weights can plug directly into standard statistical tools originally built for bulk data, often outperforming bespoke single-cell methods.

13PubMed Central. Observation weights unlock bulk RNA-seq tools for zero inflation and single-cell applications

Finding Cell Types in the Noise

Once the data is cleaned and normalized, the next step is clustering cells by their expression profiles and figuring out what each cluster represents biologically. Dimensionality reduction comes first: the expression matrix, which may have measurements across 20,000-plus genes, gets compressed into a handful of principal components that capture the most meaningful variation. From there, algorithms group cells into clusters, and those clusters get projected onto two-dimensional plots for visualization. A recent benchmarking effort found that GPU-accelerated computation can speed up the dimensionality reduction step by about 15 times compared to the best CPU methods, an important practical consideration when datasets contain millions of cells. Among full analysis pipelines tested, some achieved clustering accuracy scores above 0.97 in datasets with known cell identities.

14PubMed Central. Benchmarking large-scale single-cell RNA-seq analysis

Labeling those clusters with actual cell type names used to require manual inspection of marker genes by a domain expert. Automated annotation tools have changed that. Some match clusters against curated databases of known marker genes. Others transfer labels from previously annotated reference datasets using supervised classification. ScType, for example, cross-references both positive and negative marker genes to assign cell types in a fully automated fashion.

15Nature Communications. Fully-automated and ultra-fast cell-type identification using specific marker combinations from single-cell transcriptomic data16PubMed Central. Automated methods for cell type annotation on scRNA-seq data

Combining Datasets Without Losing the Biology

Real-world studies rarely consist of a single sequencing run. Samples processed on different days, in different labs, or on different platforms carry batch effects that can overwhelm biological differences. If left uncorrected, cells from two batches will cluster by their processing origin rather than by cell type. Batch correction methods attempt to align shared cell populations across batches while preserving genuine biological variation.

A benchmark of batch-correction methods found that Harmony, LIGER, and Seurat 3 all performed well, with Harmony recommended as the first method to try because of its significantly shorter runtime.

17PubMed Central. A benchmark of batch-effect correction methods for single-cell RNA sequencing data For more complex integration tasks, such as building cross-study atlases or combining data across technologies, methods like scANVI, Scanorama, and scVI have shown strong performance.

18Nature Methods. Benchmarking atlas-level data integration in single-cell genomics

An early foundational approach used mutual nearest neighbors in high-dimensional expression space: if a cell from batch A and a cell from batch B are each other’s closest match, they are likely the same cell type, and the difference between them can be attributed to batch effects and removed. This strategy does not require knowing the cell type composition of each batch ahead of time, which makes it flexible for exploratory work.

19PubMed Central. Batch effects in single-cell RNA-sequencing data are corrected by matching mutual nearest neighbors

Tracing How Cells Change Over Time

A snapshot of gene expression across thousands of cells often captures cells at different stages of a continuous process: stem cells differentiating, immune cells activating, cancer cells evolving. Pseudotime analysis computationally orders those cells along a trajectory that reconstructs the progression, even though every cell was sequenced at a single moment. Methods like Slingshot infer branching lineage structures, correctly identifying developmental paths in datasets with multiple branch points.

20PubMed Central. Slingshot: cell lineage and pseudotime inference for single-cell transcriptomics

RNA velocity takes this further by exploiting a subtle feature of the sequencing data itself. Standard protocols capture both spliced (mature) and unspliced (freshly transcribed) RNA. The ratio between the two for a given gene indicates whether that gene’s expression is ramping up or winding down. By computing this ratio across all genes, researchers can estimate a “velocity” vector for each cell, predicting its future transcriptional state on a timescale of hours.

21PubMed Central. RNA velocity of single cells Dynamical modeling extensions have refined this approach by fitting full kinetic models to the spliced-unspliced dynamics, extending RNA velocity to transient cell states that the original steady-state assumptions missed.

22PubMed Central. Generalizing RNA velocity to transient cell states through dynamical modeling

Differential pseudotime analysis adds another layer, allowing researchers to compare trajectories across conditions. If cells from a treated sample progress through differentiation faster or take a different path than cells from a control sample, statistical frameworks can now quantify that difference.

23Nature Communications. A statistical framework for differential pseudotime analysis with multiple single-cell RNA-seq samples

Adding Protein and Spatial Information

Gene expression alone does not tell the whole story. Protein levels do not always track mRNA levels, and knowing where a cell sits within a tissue matters for understanding its behavior. Two extensions of scRNA-seq address these gaps.

CITE-seq measures both gene expression and surface protein abundance in the same cell. Antibodies conjugated to short DNA tags bind to cell surface proteins, and those tags get sequenced alongside the cell’s mRNA. This is especially valuable for immunology, where immune cell subtypes are traditionally defined by surface markers. CITE-seq has been widely applied to immune-related disorders and infectious diseases.

24PubMed Central. Key Considerations on CITE-Seq for Single-Cell Multiomics25Nature Machine Intelligence. A multi-use deep learning method for CITE-seq and single-cell RNA-seq data integration with cell surface protein prediction and imputation

Spatial transcriptomics takes a different approach. Instead of dissociating tissue, it captures gene expression directly on a tissue section, preserving the physical layout of cells. The trade-off is resolution and transcriptome coverage: current spatial methods do not match the depth or completeness of scRNA-seq. Integrating the two, using scRNA-seq to define cell types and spatial data to map where those types sit, has become a major area of computational development. In cancer research, this combination reveals how immune cells, fibroblasts, and tumor cells organize into spatial niches that drive therapy resistance.

26PubMed Central. Single-cell and spatial transcriptomics integration: new frontiers in tumor microenvironment and cellular communication27Nature Reviews Genetics. Integrating single-cell and spatial transcriptomics to elucidate intercellular tissue dynamics

What the Technology Has Revealed in Cancer

Cancer has been one of the most productive application areas for scRNA-seq, because tumors are defined by their heterogeneity. A single tumor can contain dozens of molecularly distinct cell subpopulations, each with different growth rates, drug sensitivities, and metastatic potential. Bulk sequencing averages all of that into one blurry signal. Single-cell profiling of advanced non-small cell lung cancer, for example, identified rare cell populations like follicular dendritic cells and T helper 17 cells that were invisible in earlier studies, along with massive patient-to-patient variation in cellular composition and signaling networks.

28Nature Communications. Single-cell profiling of tumor heterogeneity and the microenvironment in advanced non-small cell lung cancer

In recurrent glioblastoma, scRNA-seq has mapped the landscape of drug-resistant cell states and the microenvironment interactions that sustain them, offering clues for why these tumors are so difficult to treat.

29PubMed Central. Single-cell RNA sequencing reveals tumor heterogeneity, microenvironment, and drug-resistance mechanisms of recurrent glioblastoma Ovarian cancer studies have used single-cell data combined with bulk transcriptomics to build prognostic gene signatures tied to specific tumor cell differentiation outcomes.

30PubMed Central. Integrated analysis of single-cell RNA-seq and bulk RNA-seq unveils heterogeneity and establishes a novel signature for prognosis and tumor immune microenvironment in ovarian cancer

The clinical translation is still in early stages. Researchers are working to leverage single-cell data for personalized therapeutic strategies, using a patient’s single-cell tumor profile to predict which drugs will target the right cell subpopulations. The scale of these datasets and the speed of computational tools are the current bottlenecks, not the biology itself.

31PubMed Central. Leveraging Single-Cell Approaches in Cancer Precision Medicine

Building Cell Atlases Across Organs and Species

Beyond disease-focused studies, scRNA-seq has enabled the construction of comprehensive reference maps of healthy tissues. A single-donor atlas of 15 major human organs cataloged over 84,000 cells spanning 252 subtypes, including novel fibroblast populations and previously uncharacterized cell-cell communication networks. T and B cells from different organs showed direct clonal sharing, meaning immune cells circulate and take up residence across the body in ways that only become visible at single-cell resolution.

32PubMed Central. Single-cell transcriptome profiling of an adult human cell atlas of 15 major organs

Cross-species comparisons have pushed this atlas-building effort into evolutionary biology. A study comparing single-cell transcriptomes across seven animal species constructed a hierarchy of cell type evolution and found that muscle and neuron cells are among the most conserved cell types across animals, with shared transcription factor programs underlying their identity.

33Cell Reports. Cross-species single-cell transcriptomic comparison reveals evolutionary conservation and divergence of cell types A more targeted comparison across amniotes, including chickens, turtles, and humans, revealed that the same cell types display highly similar global gene expression patterns despite hundreds of millions of years of evolutionary separation.

34Nature Communications. Cross-species comparison of amniote single-cell transcriptomes reveals evolutionary conservation and divergence in the chicken immune system

These atlases are becoming reference scaffolds for the field. When a researcher profiles a new sample, they can project their cells onto an existing atlas to quickly identify cell types, spot unusual populations, and compare disease tissue against a healthy baseline. The infrastructure is still growing: multi-million-cell atlases of the human body are underway through large consortium efforts, and the computational tools to query them are evolving alongside the data.