Reading Single-Cell QC Metrics
Single-cell QC metrics are measurements used to assess whether a barcode represents a cell and whether that captured cell profile is usable. Their input is a raw or filtered feature-barcode matrix, and their output supports decisions to retain, review, or exclude barcodes.
Barcode-rank curve
Section titled “Barcode-rank curve”This plot sorts all barcodes by total UMI.
| axis | meaning |
|---|---|
| x | barcode rank by UMI count |
| y | total UMI per barcode, commonly on a log scale |
The high-count beginning is enriched for cells, while the long low-count tail is enriched for background. A knee or inflection is a candidate boundary, not a perfect separation.
Low-RNA cell types can lie near the tail. A total-count cutoff can remove them selectively.
Total UMI count
Section titled “Total UMI count”The sum of UMI counts across genes for one barcode reflects both library complexity and cellular RNA content.
- Very low: empty droplet, damage, or poor capture
- Very high: doublet or a naturally large, RNA-rich cell
Interpret extremes within each sample and with cell-type markers.
Detected genes
Section titled “Detected genes”The number of genes with at least one UMI describes the diversity of detected transcripts. It is strongly related to total UMI, so view both together. High UMI with unusually few genes may indicate domination by a small set of abundant RNAs or background.
Mitochondrial fraction
Section titled “Mitochondrial fraction”This is the fraction of all UMIs assigned to mitochondrial genes. Loss of cytoplasmic RNA from a damaged cell can increase it, but normal ranges differ by tissue and type.
Do not rely blindly on an MT- symbol prefix across species and annotations. Record the mitochondrial gene list used.
Ribosomal, haemoglobin, and stress signals
Section titled “Ribosomal, haemoglobin, and stress signals”Ribosomal fraction, haemoglobin genes, heat-shock genes, and immediate-early genes can reveal sample condition and dissociation artifacts. They are contextual signals, not automatic row-deletion rules.
Doublet score
Section titled “Doublet score”A model-based score estimates how much a row resembles two cells captured together. Scrublet simulates doublets by adding observed cells and compares them with observed neighbourhoods.
Review expected loading-related doublet rate, high UMI and gene counts, incompatible lineage markers, and doublet-enriched clusters together with the score.
Ambient RNA contamination
Section titled “Ambient RNA contamination”RNA released from broken cells enters droplets. The expression profile of empty droplets can estimate this background and its contribution to occupied droplets.
SoupX and CellBender outputs are model estimates. Preserve matrices before and after correction, then examine marker specificity and count changes.
Sample-level metrics
Section titled “Sample-level metrics”| metric | question |
|---|---|
| estimated cells | Were the expected cell-associated barcodes recovered? |
| reads per cell | How deeply was each captured cell sequenced? |
| sequencing saturation | How many new UMIs are likely from more sequencing? |
| fraction reads in cells | What fraction of valid reads belongs to called cells? |
| mapping rate | How many cDNA reads aligned to the reference? |
One poor value can have several causes. Low fraction reads in cells may reflect ambient RNA, poor viability, or cell-calling settings.
What to report
Section titled “What to report”- Raw barcode and called-cell counts
- Per-sample metric distributions
- Lower and upper thresholds
- Cells removed at each stage
- Doublet and ambient-RNA tools and versions
- Paths to matrices before and after correction
- Review of candidate cell types preferentially removed