scRNA-seq QC Calculator

Calculate key scRNA-seq quality metrics: genes per cell, UMI counts, mitochondrial fraction.

>0.25: likely doublet
Please check your inputs and try again.

The Three Core QC Metrics

Single-cell quality control filters on three numbers, and each identifies a different failure mode.

MetricLow value meansHigh value means
UMI countEmpty droplet or dying cellDoublet
Genes detectedPoor capture, ambient RNADoublet
Mitochondrial %—Damaged or dying cell

The mitochondrial fraction is the most diagnostic. When a cell's membrane is compromised, cytoplasmic RNA leaks out while mitochondrial transcripts remain enclosed within mitochondria — so the mitochondrial proportion rises as the cell dies.

TissueTypical mito threshold
Most tissues5–10%
Heart, muscle20–30% — genuinely mito-rich
Nuclei preparations< 5%

Applying a 10% threshold to cardiac tissue would discard the cardiomyocytes, which are legitimately mitochondria-dense. Thresholds must be set per tissue from the observed distribution, not copied from a tutorial.

Doublets and Ambient RNA

Doublet rate scales with loading concentration, running roughly 0.8% per 1,000 cells loaded on droplet platforms. Loading 10,000 cells gives around 8% doublets — enough to create spurious intermediate cell types in clustering.

Computational detection simulates artificial doublets from the observed data and flags cells resembling them. It works well for doublets of different cell types and poorly for doublets of the same type, which remain largely undetectable.

Ambient RNA is the other pervasive artefact: free transcripts from lysed cells are captured alongside genuine ones in every droplet. This creates apparent low-level expression of markers a cell should not express, and correction tools estimate the contamination from empty droplets.

Worked Examples

Example 1: High quality cell: 5000 UMI, 2000 genes, 6% mito
All metrics pass QC filters
Result: Include in downstream analysis
Typical PBMC cell from 10x Genomics
Example 2: Low quality: 500 UMI, 100 genes, 45% mito
Low UMI + high mito = damaged cell
Result: Filter out before clustering
Empty droplet or apoptotic cell
Example 3: Why cardiac tissue needs different thresholds
Cardiomyocyte with 25% mitochondrial reads
Result: Legitimate, not damaged
Cardiac muscle is genuinely mitochondria-dense. A standard 10% filter would remove the principal cell type from the dataset.
Example 4: Doublet rate and loading
10,000 cells loaded on a droplet platform
Result: About 8% doublets expected
At roughly 0.8% per thousand loaded. These appear as intermediate clusters that can be mistaken for transitional cell states.

Common Mistakes

⚠️
Using a fixed mitochondrial threshold across tissues

Heart and muscle are legitimately mitochondria-rich. A 10% cutoff would discard exactly the cells of interest. Set thresholds from the observed distribution.

⚠️
Ignoring doublet rate when loading

Doublets rise roughly 0.8% per 1,000 cells loaded. Overloading creates artefactual intermediate populations in clustering.

⚠️
Trusting low-level marker expression

Ambient RNA from lysed cells is captured in every droplet, producing apparent expression of markers the cell does not make.

⚠️
Filtering only on low quality

Very high UMI counts and gene numbers indicate doublets. Both tails of the distribution need attention.

Frequently Asked Questions

Why filter mitochondrial reads?
Dying cells lose cytoplasmic mRNA but retain mitochondrial RNA (mitochondria have own genome). High % mito = dead/damaged cell that will confound clustering. Typical threshold: 10-25% depending on cell type. Cardiomyocytes: high mito normally (metabolically active).
Doublet detection?
Two cells captured in one droplet: appears as one cell with 2× expected UMIs and genes. Tools: Scrublet, DoubletFinder. Remove doublets before clustering — otherwise form artificial clusters with mixed identity. Higher loading density → more doublets.
Why does high mitochondrial content indicate a dying cell?
Because cytoplasmic RNA leaks out through the compromised membrane while mitochondrial transcripts stay enclosed, raising their relative proportion.
What mitochondrial threshold should I use?
It depends on tissue. Five to ten percent suits most, but heart and muscle are legitimately mitochondria-rich and need 20–30%.
How many doublets should I expect?
Roughly 0.8% per 1,000 cells loaded on droplet platforms, so 10,000 cells gives about 8%. Doublets of different cell types can be detected computationally.
What is ambient RNA?
Free transcripts released by lysed cells and captured in every droplet. It produces apparent low-level expression of markers the cell does not actually express.