Can I combine snRNA-seq and scRNA-seq data from the same tissue in one analysis?

You can, but treat the assay itself as a batch effect, not just a technical replicate. Depth heterogeneity and cell-type composition bias differ enough between the two that simple batch correction can over- or under-correct depending on the depth regime, so check for depth-driven clusters before you trust a merged UMAP or shared clustering.

Why is my snRNA-seq gene detection lower than scRNA-seq from the same tissue?

Exon-only counting on nuclei misses the unspliced pre-mRNA that dominates nuclear RNA, leaving detection around 7,000 genes against roughly 11,000 for matched whole cells. Including intronic reads, the current Cell Ranger default, raises median genes per cell by about 59% and closes most of that gap.

Why is my mitochondrial percentage near zero in snRNA-seq? Is my library broken?

That is expected, not a problem. Nuclei lack mitochondria, so a clean snRNA-seq library should show almost no mitochondrial UMIs; any elevated mito signal actually indicates ambient contamination from lysed cytoplasm rather than cell stress or death. A flat >20% filter, or an outlier test without a minimum-difference threshold, can misread this and over-filter a low-coverage dataset.

Should I use snRNA-seq or scRNA-seq for frozen tumor samples?

Use snRNA-seq. Frozen tissue rules out the enzymatic dissociation scRNA-seq needs, and nuclei isolation avoids the immune-infiltration inflation that dissociation causes, giving tumor-purity estimates that track histopathology and bulk RNA-seq more closely.

Do I need special Cell Ranger settings for single-nucleus data?

You need an intron-aware reference and counting mode, but current Cell Ranger versions already default to counting exonic and intronic reads together, which is the setting to keep rather than override. Beyond that, expect and design your QC around a near-zero mitochondrial baseline and a larger ambient RNA fraction than you would see in scRNA-seq.

Comparison · assay

Single-Nucleus RNA-seq vs Single-Cell RNA-seq: Which One Should You Use?

Frozen tissue and unbiased cell-type capture on one side, fuller transcriptomes and dissociation bias on the other, and the QC thresholds do not transfer between them.

By Ming "Tommy" Tang, Director of Bioinformatics in Big Pharma · Reviewed September 2026 · 5 min read

The verdict

Default to scRNA-seq when you have any choice in the matter. Fresh tissue that dissociates cleanly gives you higher raw gene detection without relying on intronic reads to close a gap, a bigger ecosystem of reference atlases built on cytoplasmic transcriptomes, and RNA velocity or splicing analyses that do not need caveats about nuclear export rates. Most published cell-type annotation references, most integration tutorials, and most downstream tools assume whole-cell input, so picking scRNA-seq by default means less friction fitting your data into that ecosystem.

Switch to snRNA-seq when the tissue makes the decision for you: frozen or biobanked samples where fresh dissociation is not possible, or tissue that enzymatic digestion damages or kills selectively, adipose, cardiac, bone, or brain tissue, and anything where you specifically need adherent structural cell types (hepatocytes, cholangiocytes, epithelial cells) that scRNA-seq dissociation tends to lose. It is also the better call when dissociation-induced stress genes would contaminate your readout, or when you need a tumor-purity estimate that matches histopathology instead of one inflated by dissociation-favored immune infiltration. Once you are on snRNA-seq, count introns by default, treat any mitochondrial signal as ambient contamination rather than cell health, and budget time for an ambient RNA correction step you would not need with fresh cells.

Single-nucleus RNA-seq (snRNA-seq) starts with frozen tissue. You lyse cells to release nuclei, skip enzymatic digestion entirely, and load the nuclei into the same droplet chemistry (10x Chromium, typically) used for whole cells. Single-cell RNA-seq (scRNA-seq) starts with fresh tissue that gets enzymatically or mechanically dissociated into a single-cell suspension before it ever reaches the droplet instrument. That one difference in sample prep, frozen and undissociated versus fresh and digested, is the root of almost everything else that differs downstream.

Because nuclei are full of unspliced pre-mRNA, 34 to 58 percent of UMIs in a nucleus sample map to introns, against a small fraction in whole cells where transcripts are mostly mature and spliced. Count exons only and a nucleus looks like it has about 7,000 detected genes against roughly 11,000 for a matched whole cell. Include the intronic reads, which is the current Cell Ranger default, and detection climbs back toward parity, with about 97 percent agreement in clustering compared to exon-only analysis. This is not noise: less than 0.1 percent of nucleus UMIs come from genomic DNA, so the intronic signal is real RNA, mostly from poly(dT) priming inside introns and some from strand invasion during reverse transcription.

The other consequence is which cells survive the workflow. Enzymatic dissociation for scRNA-seq favors immune cells (T/NK, myeloid, B cells) and damages or loses adherent, structurally anchored types like hepatocytes, cholangiocytes, epithelial and stellate cells, which snRNA-seq captures intact because there is no digestion step to survive. Dissociation also switches on oxidative-stress and apoptotic-signaling genes that have nothing to do with the biology you set out to study, while snRNA-seq preserves adhesion and junction-assembly gene expression closer to the in-vivo state. The flip side: because nuclei lack mitochondria, a clean snRNA-seq library should show almost no mitochondrial UMIs, so your mito QC threshold means something different in each assay, and ambient RNA carries a bigger share of each nucleus droplet since isolation lyses cytoplasm straight into the buffer.

Head to head

Criterion	Single-Nucleus RNA-seq	Single-Cell RNA-seq	Edge
Tissue requirement	Works from frozen or banked tissue; nuclei are released mechanically, with no enzymatic digestion step to survive.	Needs fresh tissue that gets enzymatically or mechanically dissociated into a single-cell suspension before loading.	Single-Nucleus RNA-seq
Raw gene detection per cell/nucleus	Exon-only counting typically detects around 7,000 genes per nucleus in matched samples.	Whole-cell capture of mature mRNA detects roughly 11,000 genes in the same matched comparison (PMC6306246).	Single-Cell RNA-seq
Fix for the detection gap	Including intronic reads (the current Cell Ranger default) raises median genes per cell by about 59% and total UMIs by about 48%, closing most of the gap.	No equivalent step needed; mature mRNA is already mostly captured without relying on unspliced reads.	Tie
Mitochondrial QC meaning	Nuclei lack mitochondria, so near-zero mito UMIs is the healthy baseline; any signal above that indicates ambient contamination, and isOutlier() with a minimum difference of +0.5% avoids over-filtering low-coverage libraries.	Mitochondrial percentage reflects real cytoplasmic content and cell stress; a flat >20% mito UMI filter is common practice for removing dying or dead cells.	Tie
Ambient RNA burden	Ambient RNA is a larger fraction of every droplet because nuclei isolation lyses cytoplasm into the buffer, making ambient-correction tools (CellBender, SoupX, DecontX, controlAmbience) close to mandatory.	Ambient contamination exists but is typically a smaller fraction per droplet, and contaminating transcripts more often originate from genuinely co-captured cell states.	Single-Cell RNA-seq
Cell-type composition fidelity	Enriches adherent and structural cell types (epithelial, hepatocyte, stellate, cholangiocyte) and under-represents immune cells relative to tissue truth.	Enzymatic dissociation enriches immune and other easily-liberated cell types while damaging or losing adherent types; in colon tissue epithelial fraction was 69.9% in snRNA-seq versus lower in matched scRNA-seq (PMC11166281).	Tie
Dissociation-induced artifact genes	Preserves cell-substrate adhesion and junction-assembly gene expression close to the in-vivo state.	Digestion triggers oxidative-stress and apoptotic-signaling gene programs, and can spike immediate-early genes (Fos, Egr1, Arc) up to 10-fold, that get mistaken for biology.	Single-Nucleus RNA-seq
Tumor purity estimation	Matches histopathology and bulk RNA-seq purity estimates more closely because it avoids dissociation's immune-infiltration bias.	Tends to overestimate immune infiltration relative to histopathology, skewing purity calls.	Single-Nucleus RNA-seq
RNA velocity and splicing dynamics	Unspliced/spliced ratios are confounded by variable nuclear export rates across cell types (about 50% nuclear localization in layer 4 cortical neurons versus about 20% in layer 5), complicating velocity interpretation and reuse of cytoplasm-trained annotation references.	Velocity and splicing analyses run on more standard assumptions since transcripts have already been exported to the cytoplasm.	Single-Cell RNA-seq
Cross-assay integration	Nuclei preparations carry different depth and gene-detection profiles that must be modeled explicitly before merging with cell-based data.	Serves as the more familiar baseline for reference atlases, but still needs depth-aware integration to avoid batch-by-depth artifacts when merged with other data.	Tie

Use Single-Nucleus RNA-seq when

Only frozen or archival/biobanked tissue is available and fresh dissociation cannot happen in time.
The tissue is hard to dissociate without damage: adipose, cardiac muscle, bone, or brain tissue, where enzymatic digestion destroys or selectively kills fragile cell types.
You need a tumor-purity estimate that agrees with histopathology, and dissociation-driven immune-cell inflation would bias the result.
You want expression profiles free of dissociation-induced stress and immediate-early gene artifacts, for example when Fos, Egr1, or Arc induction would confound your readout.
Your question depends on adherent, structural cell types (hepatocytes, cholangiocytes, epithelial cells) that scRNA-seq dissociation tends to lose or damage.

Use Single-Cell RNA-seq when

Fresh tissue is available and dissociates cleanly without major stress artifacts.
You need the fuller cytoplasmic transcriptome, including mature spliced mRNA and cytoplasm-enriched transcripts.
Immune cell subtypes (T/NK, myeloid, B cells) are central to your question, since dissociation preserves these better than adherent types.
You plan to run RNA velocity or lean on published cytoplasm-based reference atlases for annotation without correcting for nuclear export effects.
You want the highest raw gene detection per cell without depending on intronic reads to close a detection gap.

Switching between them

Switching from scRNA-seq to snRNA-seq changes more than the wet-lab step. Rebuild your counting around an intron-aware reference: current Cell Ranger versions already default to counting exonic and intronic reads together for a single gene, so don't override that to exon-only or you will silently lose about a third of detected genes. Your mitochondrial QC threshold flips meaning: a percentage that flags dying cells in scRNA-seq flags ambient contamination in snRNA-seq, and a flat >20% cutoff applied to a near-zero baseline over-filters a low-coverage nucleus dataset; use isOutlier() with a minimum difference threshold instead. Add an ambient RNA correction step (CellBender, SoupX, DecontX, or controlAmbience) that most scRNA-seq pipelines skip, since lysed cytoplasm makes ambient contamination proportionally larger in nuclei. Do not treat gene-detection counts, mitochondrial percentages, or cell-type proportions as directly comparable across the two assays: adherent and immune cell fractions differ by assay design, not by biology, and any merged analysis needs depth-aware integration before it can separate technical depth effects from real cell state.

Pitfalls with either

Excluding intronic reads from an snRNA-seq count matrix cuts median genes per cell by roughly 59% and total UMIs by roughly 48%; keep an intron-aware reference and Cell Ranger's default of counting exonic and intronic reads together.
Applying a fixed >20% mitochondrial cutoff to nuclei without checking the baseline over-filters a dataset where most libraries have near-zero mito UMIs; use isOutlier() with a minimum difference threshold instead.
Skipping ambient RNA correction on snRNA-seq because it looks like standard scRNA-seq QC misses that lysed cytoplasm makes ambient contamination a bigger fraction of each nucleus droplet; run CellBender, SoupX, DecontX, or controlAmbience before differential expression.
Comparing cell-type proportions across snRNA-seq and scRNA-seq datasets from the same tissue as if they measure the same ground truth ignores that adherent and immune cell types are systematically over- or under-represented by assay, not by biology.
Merging snRNA-seq and scRNA-seq, or any depth-heterogeneous batches, without checking for depth-driven clustering first lets library size dominate PCA/UMAP structure; check for depth-driven clusters and use depth-aware integration before trusting the merge.
Trusting RNA velocity direction from snRNA-seq without caveat ignores that variable nuclear export rates across cell types distort spliced/unspliced ratios; interpret velocity cautiously and re-check cytoplasm-based annotation assumptions.

Questions people ask

Can I combine snRNA-seq and scRNA-seq data from the same tissue in one analysis?: You can, but treat the assay itself as a batch effect, not just a technical replicate. Depth heterogeneity and cell-type composition bias differ enough between the two that simple batch correction can over- or under-correct depending on the depth regime, so check for depth-driven clusters before you trust a merged UMAP or shared clustering.
Why is my snRNA-seq gene detection lower than scRNA-seq from the same tissue?: Exon-only counting on nuclei misses the unspliced pre-mRNA that dominates nuclear RNA, leaving detection around 7,000 genes against roughly 11,000 for matched whole cells. Including intronic reads, the current Cell Ranger default, raises median genes per cell by about 59% and closes most of that gap.
Why is my mitochondrial percentage near zero in snRNA-seq? Is my library broken?: That is expected, not a problem. Nuclei lack mitochondria, so a clean snRNA-seq library should show almost no mitochondrial UMIs; any elevated mito signal actually indicates ambient contamination from lysed cytoplasm rather than cell stress or death. A flat >20% filter, or an outlier test without a minimum-difference threshold, can misread this and over-filter a low-coverage dataset.
Should I use snRNA-seq or scRNA-seq for frozen tumor samples?: Use snRNA-seq. Frozen tissue rules out the enzymatic dissociation scRNA-seq needs, and nuclei isolation avoids the immune-infiltration inflation that dissociation causes, giving tumor-purity estimates that track histopathology and bulk RNA-seq more closely.
Do I need special Cell Ranger settings for single-nucleus data?: You need an intron-aware reference and counting mode, but current Cell Ranger versions already default to counting exonic and intronic reads together, which is the setting to keep rather than override. Beyond that, expect and design your QC around a near-zero mitochondrial baseline and a larger ambient RNA fraction than you would see in scRNA-seq.

Sources

Single-nucleus and single-cell transcriptomes compared in matched cortical cell types — Matched cell/nucleus gene detection (~7,000 vs ~11,000) and expression correlation (~0.70) numbers
Comparison of cell type distribution between single-cell and single-nucleus RNA sequencing: enrichment of adherent cell types in single-nucleus RNA sequencing — Adherent vs immune cell-type enrichment, dissociation-induced stress genes, and tumor purity claims
Chapter 11 Single-nuclei RNA-seq processing — Mitochondrial QC with isOutlier() and ambient RNA estimation with controlAmbience()
QClus: a droplet filtering algorithm for enhanced snRNA-seq data quality in challenging samples — Ambient RNA correction tools and the >20% mitochondrial filter convention
Single-cell and single-nucleus RNA-sequencing from paired normal-adenocarcinoma lung samples provide both common and discordant biological insights — Cell-type proportion differences between snRNA-seq and scRNA-seq across tissues