Single-Cell QC

practice

single-cell

Published

May 7, 2026

Purpose

This page covers basic cell-level QC metrics after loading a filtered scRNA-seq count matrix into a Seurat object.

The goal is to inspect cell quality before normalization, dimensionality reduction, clustering, and annotation.

This page does not cover empty-droplet calling or formal doublet detection. Those are related but separate steps.

Common QC Metrics

Metric	Meaning	Interpretation
`nCount_RNA`	total UMI counts per cell	too low may indicate poor quality or low sequencing depth; very high may suggest doublets
`nFeature_RNA`	number of detected genes per cell	too low may indicate dead or low-quality cells; very high may suggest doublets or contamination
`percent.mt`	mitochondrial gene percentage	high values often indicate stressed, damaged, or dying cells
`percent.hb`	hemoglobin gene percentage	high values may indicate red blood cell contamination
`percent.ribo`	ribosomal gene percentage	reflects ribosomal gene signal; interpret with tissue and experiment context
`log10_ratio_features_to_umi`	`log10(nFeature_RNA) / log10(nCount_RNA)`	low values may indicate low expression complexity or RNA degradation

Default Seurat Metrics

CreateSeuratObject() automatically calculates:

nCount_RNA: column sums of the counts matrix
nFeature_RNA: number of non-zero features per cell

Check them in metadata:

metadata <- seu[[]]

head(metadata[, c("nCount_RNA", "nFeature_RNA")])
summary(metadata[, c("nCount_RNA", "nFeature_RNA")])

Calculate Default Metrics

If nCount_RNA or nFeature_RNA is missing, calculate them from the raw count matrix.

counts <- Seurat::GetAssayData(
  object = seu,
  assay = "RNA",
  slot = "counts"
)

seu$nCount_RNA <- Matrix::colSums(counts)
seu$nFeature_RNA <- Matrix::colSums(counts > 0)

Check again:

summary(seu$nCount_RNA)
summary(seu$nFeature_RNA)

What they mean:

nCount_RNA: total UMI counts per cell.
nFeature_RNA: number of detected genes per cell.

Add QC Metrics

Mitochondrial percentage for human data:

seu <- Seurat::PercentageFeatureSet(
  object = seu,             # Seurat object
  pattern = "^MT-",         # human mitochondrial genes
  col.name = "percent.mt"   # metadata column to store the result
)

For mouse data, mitochondrial genes usually start with mt-:

seu <- Seurat::PercentageFeatureSet(
  object = seu,             # Seurat object
  pattern = "^mt-",         # mouse mitochondrial genes
  col.name = "percent.mt"   # metadata column to store the result
)

Hemoglobin percentage:

hb_genes <- c("HBA1", "HBA2", "HBB", "HBD", "HBE1", "HBG1", "HBG2", "HBM", "HBQ1", "HBZ")
hb_genes <- Seurat::CaseMatch(search = hb_genes, match = rownames(seu)) # case-insensitive gene matching

seu <- Seurat::PercentageFeatureSet(
  object = seu,             # Seurat object
  features = hb_genes,      # hemoglobin gene set found in the object
  col.name = "percent.hb"   # metadata column to store the result
)

Ribosomal percentage:

ribo_genes <- grep("^RP[SL]", rownames(seu), value = TRUE, ignore.case = TRUE) # case-insensitive RPL/RPS matching

seu <- Seurat::PercentageFeatureSet(
  object = seu,              # Seurat object
  features = ribo_genes,     # ribosomal protein genes found in the object
  col.name = "percent.ribo"  # metadata column to store the result
)

Expression complexity:

seu$log10_ratio_features_to_umi <- log10(seu$nFeature_RNA) / log10(seu$nCount_RNA)

Check added metrics:

metadata <- seu[[]]

summary(metadata[, c(
  "nCount_RNA",
  "nFeature_RNA",
  "percent.mt",
  "percent.hb",
  "percent.ribo",
  "log10_ratio_features_to_umi"
)])

Visualize QC Metrics

Violin plots:

Violin plots show the distribution of each QC metric across cells. They are useful for seeing outliers and deciding whether thresholds should be sample-specific.

Seurat::VlnPlot(
  object = seu, # Seurat object
  features = c(
    "nFeature_RNA",
    "nCount_RNA",
    "percent.mt",
    "percent.hb",
    "log10_ratio_features_to_umi"
  ),            # QC metrics to visualize
  pt.size = 0.1, # point size
  ncol = 5      # number of panels per row
)

Add a threshold line for one metric:

feature_threshold <- 200

p <- Seurat::VlnPlot(
  object = seu,              # Seurat object
  features = "nFeature_RNA", # QC metric
  group.by = "sample_id",    # compare samples if available
  pt.size = 0.1              # point size
) +
  ggplot2::labs(title = "Genes per Cell") +
  ggplot2::geom_hline(
    yintercept = feature_threshold, # threshold value
    linetype = "dashed",
    color = "red"
  )

p

Scatter plots:

Scatter plots show relationships between QC metrics. They help identify cells with unusual combinations, such as high counts but low complexity, or high mitochondrial percentage.

Seurat::FeatureScatter(
  object = seu,             # Seurat object
  feature1 = "nCount_RNA",  # total counts per cell
  feature2 = "nFeature_RNA" # detected genes per cell
)

Seurat::FeatureScatter(
  object = seu,            # Seurat object
  feature1 = "nCount_RNA", # total counts per cell
  feature2 = "percent.mt"  # mitochondrial percentage
)

Seurat::FeatureScatter(
  object = seu,                            # Seurat object
  feature1 = "nCount_RNA",                 # total counts per cell
  feature2 = "log10_ratio_features_to_umi" # expression complexity
)

Comprehensive QC Scatter

A comprehensive scatter plot can show library size, detected genes, mitochondrial percentage, sample structure, and threshold lines in one figure.

This is useful for multi-sample QC.

metadata <- seu[[]]

umi_threshold <- 500
feature_threshold <- 200

p_comprehensive <- metadata |>
  ggplot2::ggplot(
    ggplot2::aes(
      x = nCount_RNA,
      y = nFeature_RNA,
      color = percent.mt
    )
  ) +
  ggplot2::geom_point(
    size = 1,
    alpha = 0.8
  ) +
  ggplot2::scale_colour_gradient(
    low = "gray90",
    high = "#8856a7"
  ) +
  ggplot2::scale_x_log10() +
  ggplot2::scale_y_log10() +
  ggplot2::geom_vline(
    xintercept = umi_threshold,
    linetype = "dashed",
    color = "red"
  ) +
  ggplot2::geom_hline(
    yintercept = feature_threshold,
    linetype = "dashed",
    color = "red"
  ) +
  ggplot2::facet_wrap(
    facets = ~sample_id,
    labeller = ggplot2::label_both
  ) +
  ggplot2::theme_classic() +
  ggplot2::labs(
    title = "UMI Counts vs Genes by Sample",
    x = "UMI Counts",
    y = "Genes Detected",
    color = "Mitochondrial %"
  )

p_comprehensive

Optional trend line:

p_comprehensive +
  ggplot2::stat_smooth(
    method = "lm",
    color = "darkblue",
    linewidth = 1,
    linetype = "solid",
    se = FALSE
  )

If sample_id exists, inspect QC by sample:

Seurat::VlnPlot(
  object = seu, # Seurat object
  features = c(
    "nFeature_RNA",
    "nCount_RNA",
    "percent.mt"
  ),            # QC metrics
  pt.size = 0.1, # point size
  group.by = "sample_id", # split distributions by sample
  ncol = 3      # number of panels per row
)

Filtering

Filtering removes cells that look technically low quality based on QC metrics.

Thresholds are dataset-specific. The values below are common starting points, not fixed rules.

Metric	Common Minimum	Common Maximum	Purpose
`nCount_RNA`	500	dataset-specific	remove cells with very low UMI counts
`nFeature_RNA`	200	6000	remove low-complexity cells and unusually feature-rich cells
`percent.mt`	0	10-20	remove cells with high mitochondrial signal
`percent.hb`	0	5	remove cells with high hemoglobin signal, if relevant
`percent.ribo`	0	50	check high ribosomal signal, if relevant
`log10_ratio_features_to_umi`	0.8	`Inf`	optional metric for low expression complexity

Filter cells:

seu <- subset(
  x = seu,
  subset = nCount_RNA > 500 &
    nFeature_RNA > 200 &
    nFeature_RNA < 6000 &
    percent.mt < 10 &
    percent.hb < 5 &
    percent.ribo < 50 &
    log10_ratio_features_to_umi > 0.8
)

Check after filtering:

dim(seu)

if ("sample_id" %in% colnames(seu[[]])) {
  table(seu$sample_id)
}

Note

Basic QC mainly asks whether each cell has reasonable library size, detected gene number, mitochondrial signal, blood-related signal, and expression complexity.

Doublet detection and empty-droplet calling can use related signals, but they should be treated as separate analysis steps.