Single Cell RNA Sequencing Workflow: A Researcher's Primer

Principles, the seven bench stages, platform trade-offs, applications, and the QC failures that quietly wreck experiments.

Amanda Hu

| August 21, 2026 · 14 Single cell RNA sequencing scRNA-seq workflow Single nucleus RNA-seq NGS library prep Quality control

Single Cell RNA Sequencing Workflow: A Researcher's Primer

The single cell RNA sequencing workflow turns a piece of tissue into a table with one column per cell and one row per gene — and almost every decision that determines whether that table is trustworthy is made at the bench, before any software runs. This primer walks that workflow stage by stage: the five principles it rests on, the seven bench stages from dissociation to count matrix, how droplet, plate, combinatorial and single-nucleus platforms differ, and the quality-control signatures that separate a real biological result from a dissociation artifact.

Single cell vs bulk RNA sequencing: what changes when you stop averaging

A bulk RNA-seq library made from a tissue sample is a weighted average. Every transcript count is summed across however many thousands or millions of cells went into the lysate, and the composition of that mixture is invisible in the output. Suppose a treatment doubles the expression of a gene. There are two very different ways that could happen. In the first, every cell in the sample doubled its output. In the second, most cells did not change at all, and a small new subpopulation appeared that expresses the gene very highly. Bulk data cannot tell these apart, because both produce the same average — yet they imply completely different biology, and they are routinely confused.

Single cell vs bulk RNA sequencing — a 2-fold bulk average is equally consistent with every cell doubling or with a new high-expressing subpopulation appearing — Figure 1. Why the single cell RNA sequencing workflow exists. A single bulk number (left) is consistent with two incompatible explanations (right): a uniform shift across one population, or a new, small, highly expressing subpopulation. Only per-cell resolution distinguishes them. Illustrative schematic — not experimental data. Click to enlarge.

Single cell RNA sequencing (scRNA-seq) resolves this by keeping the transcripts from each cell separately labelled, so the final data matrix has cells as columns and genes as rows rather than one column per sample. The first demonstration came in 2009, when a single mouse blastomere was profiled and found to express thousands more genes than microarray methods had detected.¹ The field then spent roughly six years scaling: from one cell at a time, to plates of a few hundred, to droplet methods in 2015 that handled thousands of cells in a single run.^2,3 By 2017, droplet implementations were reported profiling tens of thousands of cells per run at low reagent cost per cell, which is roughly where routine practice sits today.⁴

The trade is unavoidable and worth stating plainly at the outset. You gain resolution across cells and lose depth within each one. A bulk library might see 30–50 million reads assigned to one sample; a droplet-based single-cell library divides a similar budget across thousands of cells, giving each one a few thousand detected genes at best. Most genes in most cells register zero counts. That sparsity is not a defect to be corrected away — it is the expected result of sampling a few thousand molecules from a cell containing hundreds of thousands, and it shapes every analytical decision downstream.¹⁸

What actually happens at the bench, in plain terms

Before the principles, here is the physical sequence, stripped of jargon.

Start with a sample. That might be a piece of tissue from surgery or a core needle biopsy, a tube of blood, a bone-marrow aspirate, a dissected organ from a model animal, or simply a flask of cultured cells.

Take the sample apart into loose cells. Solid tissue is held together by extracellular matrix, so it is incubated with enzymes that digest that scaffolding, with gentle agitation to help it along. What was a piece of tissue becomes a cloudy liquid of free-floating individual cells. Blood and cultured cells largely skip this, because their cells are already separate.

Check the suspension. It is passed through a fine filter to remove clumps and debris, then counted and scored to confirm most cells are still intact and alive.

Give every cell its own compartment and its own label. This is the step that makes the method single-cell. The suspension is loaded onto an instrument that wraps each cell in a separate watery droplet suspended in oil, each droplet roughly the width of a human hair. Every droplet also receives a tiny bead coated with millions of copies of one short DNA sequence — that droplet's barcode, different from every other droplet's. The cell is broken open inside its droplet, and its RNA is copied into DNA with the barcode attached to every single copy. Plate-based protocols do the same thing with wells instead of droplets: one cell sorted or pipetted per well, one barcode per well.

Pool everything and sequence it together. Because each DNA copy now carries a label recording which cell it came from, the droplets can be broken and the entire contents tipped into one tube without losing that information. The pooled material is amplified, cleaned up, and sequenced as a single library — which is what makes the method affordable.

Sort it back out computationally. Software reads the barcode on each sequence, sorts the sequences into per-cell groups, and counts how many distinct RNA molecules of each gene each cell contained. The output is a large table: one column per cell, one row per gene.

Five principles the single cell RNA sequencing workflow rests on

Every scRNA-seq platform — a 96-well plate, a droplet microfluidic chip, a combinatorial-indexing protocol — is a different engineering answer to the same five problems. Understanding the ideas makes the platform comparison later in this primer straightforward rather than a list of brand names.

1. Compartmentalization — keeping one cell's RNA away from another's

Before anything can be labelled per cell, each cell has to be physically separated: into a well, a droplet, a nanowell, or (in combinatorial approaches) a pool that is split and re-split so that no two cells share the same sequence of pool assignments. Everything about a platform's throughput, cost per cell, and doublet rate follows from how it does this.

2. Cell barcodes — a short sequence that says "this read came from that cell"

Each compartment supplies primers carrying a distinct barcode, typically 12–16 nucleotides. The barcode is incorporated during reverse transcription, so it is physically attached to every cDNA molecule from that cell. Once barcoding is complete the compartments no longer matter: everything can be pooled into a single tube, amplified together, and sequenced together, because the barcode preserves cell identity in sequence space. This pooling step is the reason droplet methods scale. Barcode collisions and ambient barcode swapping are the two mechanisms behind doublets and background contamination; both are corrected computationally rather than experimentally, so both need to be checked in every dataset.

3. Unique molecular identifiers (UMIs) — counting molecules, not reads

Single-cell input is a few picograms of RNA, so heavy PCR amplification is unavoidable — and PCR is not uniform. A random 8–12 nucleotide UMI added alongside the cell barcode tags each original mRNA molecule individually, so all reads sharing one UMI can be collapsed to a single count.⁸ Without UMIs, a gene that happened to amplify well would look highly expressed. With them, the count approximates the number of transcripts actually captured.

4. 3′ counting vs. full-length coverage — the single biggest design choice

High-throughput methods sequence only a short window at the 3′ end of each transcript. This is enough to identify the gene and count it, and it means read length can stay short and cheap. Full-length protocols such as Smart-seq2 instead cover the whole transcript, which costs far more per cell but is the only way to see splice isoforms, allelic expression, or sequence variants.^5,6 Later full-length chemistries added UMIs to combine molecule counting with isoform resolution.⁷

Decide early. If your question is which cell types are here and how do their abundances shift, 3′ counting is correct and cheaper. If it is which isoform, which allele, which variant, you need full-length coverage and will profile far fewer cells.

5. Sparse, count-based data — zeros are information, not missing values

A typical count matrix is 90–95% zeros. Some of those zeros are biological (the gene is off) and some are sampling (the gene is on but no molecule was captured), and no method distinguishes them with certainty for a given gene in a given cell. Analysis therefore works with distributions across groups of cells rather than point estimates per cell, and the field's normalization, dimensionality-reduction and clustering conventions all exist to handle this.^16,17

The single cell RNA sequencing workflow: seven stages from tissue to count matrix

The following is the shape of a droplet-based 3′ counting experiment, the most common configuration. Plate-based and combinatorial protocols differ mainly at stages 3 and 5.

Seven-stage droplet-based single cell RNA sequencing workflow diagram — dissociation, suspension QC, partitioning and barcoding, reverse transcription, pooling and library prep, sequencing, and analysis — Figure 2. The seven-stage droplet-based single cell RNA sequencing workflow, with the dominant failure mode at each stage. Orange-outlined stages (1, 3, 7) are where most runs are decided — dissociation stress, Poisson doublets, and treating cells rather than samples as replicates. Illustrative schematic. Click to enlarge.

Stage 1 — Sample preparation and dissociation

Tissue is enzymatically and mechanically broken into a suspension of intact, individual cells. Cultured cells need only gentle detachment; solid tissue needs a protease cocktail matched to its matrix. This is the stage where most experiments are decided. Warm proteolytic dissociation at 37 °C activates a transcriptional stress response — immediate-early genes such as Fos and Jun plus heat-shock genes — in a subpopulation of cells, and the induced signature is strong enough that a "quiescent" population can appear partly activated purely as an artifact.¹² Cold-active proteases from psychrophilic organisms, used at 4–6 °C, substantially reduce this signature in kidney and in solid tumors.^13,14 If the tissue resists dissociation, single-nucleus RNA-seq is the alternative rather than harsher digestion.

Stage 2 — Suspension QC: viability, concentration, debris

The suspension is counted and scored for viability, then filtered and, if needed, depleted of dead cells and red blood cells. Most platforms want high viability — commonly above 80–90% — and a narrow concentration range. Dead and dying cells leak mRNA into the buffer, and that free RNA is captured by every droplet in the run. It appears as a low-level "ambient" or soup profile added to every cell, blurring cluster boundaries and generating apparent co-expression of markers that no real cell expresses together. High mitochondrial read fraction per cell is the standard downstream signature of compromised cells. Debris and clumps also drive doublet rates, because a clump loaded into one droplet is barcoded as one cell.

Stage 3 — Partitioning and barcoding

Cells, barcoded primer-bearing beads and reagent are co-flowed so that each droplet ideally receives one cell and one bead. The cell lyses inside the droplet, releasing its mRNA into a compartment already containing its unique barcode. Loading follows Poisson statistics, so single occupancy cannot be guaranteed — it is traded off. Loading fewer cells lowers the doublet rate but wastes reagent; loading more recovers more cells but raises multiplets. The barcode-plus-UMI oligo also carries a poly(dT) stretch, which selects polyadenylated mRNA out of the total RNA pool and simultaneously positions the barcode at the 3′ end — the structural reason 3′ methods are 3′ methods.

Stage 4 — Reverse transcription

Within each droplet or well, a reverse transcriptase copies captured mRNA into first-strand cDNA, incorporating the cell barcode and UMI. Template switching appends a known sequence to the 3′ end of the new strand, giving both ends a defined handle for amplification. Nearly all single-cell chemistries use engineered Moloney murine leukemia virus (MMLV) reverse transcriptases with the RNase H domain inactivated, because the parent enzyme's RNase H activity degrades the RNA template mid-synthesis and truncates products. Engineered variants also add thermostability, which helps read through GC-rich secondary structure, and it is the enzyme's intrinsic terminal transferase and template-switching behaviour — not a separate step — that installs the second handle. RT efficiency sets the capture rate, and capture rate is the hard ceiling on how many of a cell's transcripts you will ever see.

Stage 5 — Pooling, cDNA amplification and library construction

Droplets are broken and the barcoded cDNA from every cell is combined into one tube. PCR amplifies the pooled material; the product is cleaned and size-selected, then fragmented, adapter-ligated and indexed into a sequencer-ready library. Because the barcode is already covalently attached, pooling loses no information — this is what makes the method economical. Amplification bias is real but is absorbed by UMI collapsing. Bead-based size selection using solid-phase reversible immobilization (SPRI) chemistry is used repeatedly here: adjusting the polyethylene glycol and salt ratio changes which fragment lengths precipitate onto the beads, letting you clear primers, adapter dimers and off-target short products between steps.

Stage 6 — Sequencing

The library is sequenced with an asymmetric read configuration: a short read covering the barcode and UMI, and a longer read covering the transcript fragment. Depth is budgeted per cell rather than per sample. Because 3′ counting only needs enough transcript sequence to assign a gene, the cDNA read can be short. The practical planning question is the split between cells and reads at fixed cost: more cells at shallow depth resolves composition and rare cell types; fewer cells at greater depth resolves expression within cell types. Method benchmarking studies show that sensitivity per cell and cost per cell vary several-fold across protocols at matched depth, so this is worth deciding against published comparisons rather than by default.^9,10

Stage 7 — Processing and analysis

Reads are demultiplexed by cell barcode, aligned, and collapsed by UMI into a counts matrix. Then: cell filtering, normalization, feature selection, dimensionality reduction, clustering, annotation and differential testing. This stage carries as much methodological risk as the bench work, and the choices interact. Published tutorials lay out the pre-processing sequence and the reasoning for each step, and are worth following rather than reinventing.¹⁶ Two cautions recur: clustering will always return clusters, including from noise, so annotations need marker evidence and ideally orthogonal validation; and trajectory inference methods disagree substantially with one another on the same data, so a single inferred lineage is a hypothesis rather than a result.²⁰

Choosing a platform: droplet, plate, combinatorial and single nucleus RNA sequencing

The four families below cover most published work. The right choice follows almost entirely from two questions: how many cells you need, and whether you need full-length transcript coverage.

Approach	Cells per run	Coverage	Best for	Main limitation
Droplet, 3′ counting	10³–10⁵	3′ end, UMI	Cell-type census, composition shifts, rare population discovery, atlas building	No isoform or allele resolution; ambient RNA background
Plate-based, full-length	10²–10³	Whole transcript	Splice isoforms, allelic expression, variant detection, index-sorted rare cells	Cost and labour per cell; low throughput
Combinatorial indexing	10⁴–10⁶	3′ end, UMI	Very large cell numbers without instrument-limited partitioning	More hands-on steps; lower sensitivity per cell
Single-nucleus (snRNA-seq)	10³–10⁵	3′ end, UMI	Frozen or archived tissue, brain, adipose, muscle and other hard-to-dissociate tissue	Misses cytoplasmic transcripts; higher intronic fraction

Single-nucleus profiling deserves particular attention because it solves the dissociation problem rather than mitigating it: nuclei can be released from frozen tissue by mechanical homogenization without any warm proteolysis, so the stress-response artifact does not arise.¹¹ The cost is that the nuclear transcriptome is not the cellular transcriptome — mature cytoplasmic mRNAs are underrepresented and unspliced nuclear transcripts are enriched. Systematic comparisons of single-cell and single-nucleus methods on matched samples quantify these differences and are the right basis for choosing between them.¹⁰

Plan multiplexing before the run. Pooling several samples into one run and separating them afterwards — by natural genetic variation, or by antibody- or lipid-tagged sample barcodes — is now routine, and it removes batch effect as a confounder for the comparison you actually care about. If you are comparing conditions, design this in rather than treating it as an add-on.

Single cell RNA sequencing applications: where the resolution pays off

The unifying feature of productive scRNA-seq applications is that the biological question is about heterogeneity — composition, state distribution, or rare populations. Questions about the average behaviour of a homogeneous population are usually answered better and far more cheaply by bulk RNA-seq.

Cell-type discovery and reference atlases

Unbiased profiling finds populations that marker-based approaches cannot, because it does not require knowing what to stain for. Published single-cell atlases — large annotated datasets built from many donors and validated against protein and imaging data — now serve as reference maps for new experiments. Cells in your dataset are matched to atlas cells and inherit their labels, rather than being named by manual marker inspection. A human lung atlas, for example, resolved dozens of cell types, including populations that had not previously been characterized.²¹ The limit: reference mapping only recognizes cell types the atlas already contains. A cluster that maps poorly is a signal that something may be novel — not an identification of what it is.

Tumour heterogeneity and the microenvironment

A tumour biopsy is a mixture of malignant cells, fibroblasts, endothelium and immune cells in proportions that vary between patients. Bulk expression from such a sample confounds tumour-intrinsic programs with infiltration level. Single-cell data separates them, and additionally exposes malignant subclones with distinct transcriptional states — the substrate on which treatment resistance is selected. Solid tumour dissociation is exactly the scenario where collagenase-associated stress signatures are strongest, and they can be mistaken for a biologically interesting cell state; cold-active protease protocols were developed for this case.¹⁴

Immune profiling and activation states

Immune cells occupy continuous activation, exhaustion and differentiation gradients that flow cytometry compresses into discrete gates on a handful of markers. Transcriptome-wide profiling recovers the gradient. Combined protein-and-RNA measurement on the same cells — antibody-derived tags read out alongside the transcriptome — bridges the two data types directly, letting established surface-marker definitions be mapped onto transcriptional states.¹⁵

Development, differentiation and lineage

Because a developing or differentiating tissue contains cells at many points along the same trajectory simultaneously, one snapshot can be computationally ordered into a pseudotemporal sequence. This is genuinely powerful and easy to over-interpret: a systematic comparison of trajectory inference methods found substantial disagreement between tools on identical data.²⁰ Treat an inferred trajectory as a hypothesis for lineage tracing or time-course validation.

Perturbation, drug response, and multi-omic extensions

When a compound or genetic perturbation affects only a subset of cells, bulk readouts dilute the effect toward nothing. Single-cell readouts identify the responsive population and its state change directly, and pooled formats allow many perturbations to be assayed in one experiment with the perturbation identity captured alongside the transcriptome. The cell-barcode principle also generalizes beyond mRNA: the same partitioning and barcoding scheme has been extended to chromatin accessibility, surface protein, immune receptor sequence and spatial position, measured jointly with the transcriptome. Each addition constrains interpretation of the others — accessibility suggests regulatory mechanism, protein anchors to established phenotype definitions, position restores tissue context lost at dissociation.

Where the workflow fails, and how to see it in the QC data

Almost every disappointing scRNA-seq dataset fails for one of a short list of reasons, and most are visible in QC metrics before any biological conclusion is drawn. Checking for each of these explicitly is faster than debugging a confusing clustering result.

Three single cell RNA sequencing quality control failure modes — dissociation stress signature, ambient RNA soup and doublets — with their data signatures and mitigations — Figure 3. The three artifacts that are visible in scRNA-seq QC before any biological conclusion is drawn, with the mitigation for each. All three are diagnosed from the count matrix itself, not from the biology you are hoping to find. Illustrative schematic. Click to enlarge.

Failure mode	Signature in the data	Mitigation
Dissociation stress response	A cluster defined by immediate-early and heat-shock genes, spanning multiple cell types	Cold-active protease at 4–6 °C, shorter digestion, or switch to single-nucleus; score and regress the signature during analysis^12,13
Ambient (free) RNA	Highly expressed markers of one cell type appearing at low level in all clusters	Improve viability and washing before loading; apply ambient-RNA correction and verify markers are cluster-restricted
Doublets and multiplets	Cells with unusually high UMI and gene counts, expressing two mutually exclusive marker sets	Load at lower concentration; remove clumps by filtration; apply doublet detection; use genetic or tag-based demultiplexing where available
Dead and dying cells	Elevated mitochondrial read fraction, low gene count per cell	Dead-cell depletion before loading; filter on a threshold set from the actual distribution rather than a fixed default
Low RNA capture	Uniformly low median genes per cell across all populations	Check RNA integrity in the input, reverse transcription conditions, and whether the cell type is simply RNA-poor
Batch effect confounded with condition	Clusters separating by sample rather than by biology	Multiplex conditions into shared runs; if impossible, replicate across batches so batch and condition are not collinear
Under-powered comparison	Differential expression driven by a handful of cells from one donor	Treat biological replicates, not cells, as the unit of replication; aggregate to pseudobulk per sample for condition-level testing

The last row is the most common statistical error in the literature. Thousands of cells from two mice is n = 2, not n = thousands. Cells within one animal are not independent observations of a treatment effect, and testing them as if they were produces confidently wrong p-values.

Where catalog reagents touch the single cell RNA sequencing workflow

Scope, stated plainly. BioHippo does not currently stock single-cell partitioning instruments, barcoded bead kits, or complete scRNA-seq library chemistries. Those come from the platform vendor whose instrument you are running, and nothing below substitutes for them. What the catalog does cover are the surrounding steps a single-cell experiment depends on: culture-level quality control before you commit a precious sample, the enzyme class used at the reverse transcription step, bead chemistry for cleanup and size selection, and matched bulk RNA-seq for the validation arm.

Upstream: is the sample worth committing?

A single-cell run is expensive enough that it is worth confirming a culture or treatment condition is healthy before loading. Plate-based viability assays answer that question at the population level: the ATP Luminescent Cell Viability Assay Kit reads intracellular ATP as a proxy for viable biomass, Cell Counting Kit (CCK-8) uses tetrazolium reduction by cellular dehydrogenases, and CellQuanti-Blue™ Cell Viability Assay Kits use resazurin reduction. To be clear about what these are and are not: they report on a well of cells, not on individual cells in suspension, and they do not replace the per-cell viability count on the dissociated suspension immediately before loading. They are a screening tool for the condition, not a substitute for the pre-load check. Browse the full range in Assay Kits.

Reverse transcription: the enzyme class

Stage 4 of the workflow depends on an engineered MMLV-derived reverse transcriptase. The same enzyme class is stocked as standalone catalog enzymes for bulk cDNA synthesis: Hifair™ Super Reverse Transcriptase, Glycerol-free (600 U/µL) is an engineered M-MLV (H−) derivative that operates up to 70 °C and synthesizes cDNA up to 10 kb, with a glycerol-free formulation suited to lyophilization; Hifair™ V Reverse Transcriptase (600 U/µL, Glycerol free) is formulated for templates with strong secondary structure or low-copy targets; and OneScript® Hot Reverse Transcriptase is a thermostable option. These are not validated single-cell chemistries and will not substitute for a platform kit — but they are the practical reagents for the RT-qPCR validation of markers your single-cell analysis nominates. See the full Molecular Biology Enzymes range.

Cleanup and size selection

Hieff NGS™ DNA Selection Beads are SPRI-principle magnetic beads for DNA purification and size selection during NGS library preparation, stated by the supplier to be compatible with major DNA and RNA library prep kits. Two specifics worth knowing before you order: the standard formulation recovers fragments of roughly 150 bp and longer, and the bead buffer is designed for DNA rather than RNA, so it is not the right choice for RNA purification steps.

The bulk RNA-seq validation arm

Most well-designed single-cell studies include matched bulk RNA-seq, both as an orthogonal check on pseudobulk aggregates and because bulk is far cheaper for questions that do not need cell resolution. Hieff™ Cell/Tissue Total RNA Kit covers extraction. For library construction, Hieff NGS™ Ultima Dual-mode RNA Library Prep Kit builds strand-specific or non-strand-specific libraries from total RNA for Illumina and MGI platforms, using dUTP incorporation in the second strand plus a uracil-intolerant high-fidelity polymerase to achieve strand specificity; note that it requires a separately purchased platform-specific primer mix and an upstream mRNA-purification or rRNA-depletion step. Hieff NGS™ EvoMax RNA Library Prep Kit (Strand-specific) is an alternative strand-specific route, and 2×Super Canace™ II High-Fidelity Mix for Library Amplification covers the final amplification step. These kits, beads and mixes are part of the Yeasen Biotechnology line; related nucleic-acid reagents sit in DNA & RNA.

Single cell RNA sequencing FAQ

How does single cell RNA sequencing work?

Each cell is placed in its own compartment — a droplet, a well, or a combinatorial pool — together with primers carrying a barcode unique to that compartment. Reverse transcription copies the cell's mRNA into cDNA with that barcode and a UMI attached to every molecule. Because cell identity is now written into the sequence itself, all compartments can be pooled, amplified and sequenced as one library, then sorted back into per-cell columns computationally. That barcode-then-pool trick is the whole reason the single cell RNA sequencing workflow is affordable at scale.

How many cells do I need?

It depends entirely on the rarest population you need to detect and how confidently. Detecting a population present at 1% requires enough cells that you sample it many times over — a few thousand cells will contain it but may not give enough of them to characterize. Composition comparisons between conditions additionally need replicates, since the unit of replication is the sample, not the cell. Work backwards from the rarest population and the effect size, not forwards from a round number.

Should I sequence more cells or more deeply?

More cells for questions about composition and rare populations; more depth for questions about expression differences within a defined cell type. At fixed budget these trade directly against each other. Published method comparisons report sensitivity and cost per cell across protocols at matched depth and are the right reference for this decision.^9,10

Can I use frozen tissue?

Not reliably for single-cell — freezing damages membranes and viability after thaw is usually too low. Frozen and archived tissue is the standard use case for single-nucleus RNA-seq, which releases intact nuclei by mechanical homogenization and skips warm proteolysis entirely.¹¹ You can also cryopreserve a viable cell suspension in advance, which is different from freezing bulk tissue and generally works.

Why do my clusters separate by sample instead of by biology?

That is a batch effect, and it usually means each sample was processed as its own run. The durable fix is upstream: multiplex conditions into shared runs so that batch and condition are not the same variable. Integration methods can align datasets after the fact, but they cannot recover a comparison that was never separable from batch in the first place.

Do I need to impute the zeros?

Usually not, and imputation carries real risk of manufacturing correlations that were not measured. Droplet count data is reasonably described as a sampling process without excess zero inflation beyond what the sampling itself predicts.¹⁸ Standard practice is to work with the sparse counts under an appropriate model rather than filling them in.

How do I know a cluster is a real cell type and not an artifact?

Three checks. Does it have specific positive markers rather than only being defined by what it lacks? Do those markers hold up in an independent modality — protein, imaging, or a reference atlas? And does it survive the artifact screens: is it a stress-response cluster, a doublet cluster, or a low-quality cluster? Clustering algorithms return clusters from any input, so the burden of evidence is on the annotation.

Is single cell replacing bulk RNA-seq?

No. Bulk remains the better tool when the question is about the average behaviour of a population you already know is homogeneous, when sensitivity for low-abundance transcripts matters, or when the budget is better spent on more biological replicates. The two are complementary, and pairing them is a common and sound design.¹⁹

The single cell RNA sequencing workflow at a glance

Stage	What happens	The thing that goes wrong
1. Dissociation	Tissue to single-cell suspension	Warm proteolysis induces a stress-gene signature
2. Suspension QC	Count, viability, debris removal	Dead cells leak RNA into every droplet
3. Partitioning	One cell + one barcoded bead per compartment	Poisson loading produces doublets
4. Reverse transcription	mRNA to barcoded, UMI-tagged cDNA	Capture efficiency caps what you can ever detect
5. Library construction	Pool, amplify, size-select, index	Incomplete cleanup carries adapter and primer artifacts
6. Sequencing	Barcode/UMI read plus transcript read	Depth budgeted wrong for the question asked
7. Analysis	Counts matrix to annotated cell states	Cells treated as replicates instead of samples

References

Tang F, Barbacioru C, Wang Y, et al. mRNA-Seq whole-transcriptome analysis of a single cell. Nat Methods. 2009;6(5):377–382. PMID 19349980
Macosko EZ, Basu A, Satija R, et al. Highly parallel genome-wide expression profiling of individual cells using nanoliter droplets. Cell. 2015;161(5):1202–1214. PMID 26000488
Klein AM, Mazutis L, Akartuna I, et al. Droplet barcoding for single-cell transcriptomics applied to embryonic stem cells. Cell. 2015;161(5):1187–1201. PMID 26000487
Zheng GXY, Terry JM, Belgrader P, et al. Massively parallel digital transcriptional profiling of single cells. Nat Commun. 2017;8:14049. PMID 28091601
Picelli S, Björklund ÅK, Faridani OR, et al. Smart-seq2 for sensitive full-length transcriptome profiling in single cells. Nat Methods. 2013;10(11):1096–1098. PMID 24056875
Picelli S, Faridani OR, Björklund ÅK, et al. Full-length RNA-seq from single cells using Smart-seq2. Nat Protoc. 2014;9(1):171–181. PMID 24385147
Hagemann-Jensen M, Ziegenhain C, Chen P, et al. Single-cell RNA counting at allele and isoform resolution using Smart-seq3. Nat Biotechnol. 2020;38(6):708–714. PMID 32518404
Kivioja T, Vähärautio A, Karlsson K, et al. Counting absolute numbers of molecules using unique molecular identifiers. Nat Methods. 2012;9(1):72–74. PMID 22101854
Ziegenhain C, Vieth B, Parekh S, et al. Comparative analysis of single-cell RNA sequencing methods. Mol Cell. 2017;65(4):631–643.e4. PMID 28212749
Ding J, Adiconis X, Simmons SK, et al. Systematic comparison of single-cell and single-nucleus RNA-sequencing methods. Nat Biotechnol. 2020;38(6):737–746. PMID 32341560
Habib N, Avraham-Davidi I, Basu A, et al. Massively parallel single-nucleus RNA-seq with DroNc-seq. Nat Methods. 2017;14(10):955–958. PMID 28846088
van den Brink SC, Sage F, Vértesy Á, et al. Single-cell sequencing reveals dissociation-induced gene expression in tissue subpopulations. Nat Methods. 2017;14(10):935–936. PMID 28960196
Adam M, Potter AS, Potter SS. Psychrophilic proteases dramatically reduce single-cell RNA-seq artifacts: a molecular atlas of kidney development. Development. 2017;144(19):3625–3632. PMID 28851704
O'Flanagan CH, Campbell KR, Zhang AW, et al. Dissociation of solid tumor tissues with cold active protease for single-cell RNA-seq minimizes conserved collagenase-associated stress responses. Genome Biol. 2019;20(1):210. PMID 31623682
Stoeckius M, Hafemeister C, Stephenson W, et al. Simultaneous epitope and transcriptome measurement in single cells. Nat Methods. 2017;14(9):865–868. PMID 28759029
Luecken MD, Theis FJ. Current best practices in single-cell RNA-seq analysis: a tutorial. Mol Syst Biol. 2019;15(6):e8746. PMID 31217225
Kharchenko PV. The triumphs and limitations of computational methods for scRNA-seq. Nat Methods. 2021;18(7):723–732. PMID 34155396
Svensson V. Droplet scRNA-seq is not zero-inflated. Nat Biotechnol. 2020;38(2):147–150. PMID 31937974
Haque A, Engel J, Teichmann SA, Lönnberg T. A practical guide to single-cell RNA-sequencing for biomedical research and clinical applications. Genome Med. 2017;9(1):75. PMID 28821273
Saelens W, Cannoodt R, Todorov H, Saeys Y. A comparison of single-cell trajectory inference methods. Nat Biotechnol. 2019;37(5):547–554. PMID 30936559
Travaglini KJ, Nabhan AN, Penland L, et al. A molecular cell atlas of the human lung from single-cell RNA sequencing. Nature. 2020;587(7835):619–625. PMID 33208946

This primer summarizes published single-cell RNA sequencing methodology and is provided for research reference. Method performance figures cited are from the referenced comparison studies, not from BioHippo testing. BioHippo does not supply single-cell partitioning instruments or complete scRNA-seq library chemistries; the catalog products referenced are general molecular-biology and NGS reagents and are not validated for single-cell library preparation. Confirm specifications, compatibility and intended-use statements on each product page before purchase. All products are for research use only.