Variable Features

Purpose

Variable feature selection identifies genes with high biological variation across cells.

This step is usually done after normalization and before scaling and PCA.

Typical order:

NormalizeData() -> FindVariableFeatures() -> ScaleData() -> RunPCA()

Why Variable Features

Single-cell RNA-seq data contain thousands of genes, but many genes are uninformative for cell-state differences.

Variable features are used to focus downstream dimensionality reduction on genes that capture meaningful cell-to-cell variation.

What to remember:

• variable features are not all genes
• they are usually used for PCA
• they are selected after normalized expression is available
• the selected genes depend on dataset, normalization, and method

Find Variable Features

Standard Seurat workflow:

seu <- Seurat::FindVariableFeatures(
  object = seu,              # Seurat object after NormalizeData()
  selection.method = "vst",  # default and common method
  nfeatures = 2000,          # number of variable features to keep
  verbose = TRUE             # show progress messages
)

Common parameters:

Parameter	Meaning
object	Seurat object
selection.method	method for selecting variable features
nfeatures	number of variable features to keep

For basic scRNA-seq analysis, selection.method = "vst" and nfeatures = 2000 are common starting choices.

Check Variable Features

Extract selected features:

variable_features <- Seurat::VariableFeatures(seu)

length(variable_features)
head(variable_features, 10)
sample(variable_features, 10)

Check whether expected marker genes appear:

"IL7R" %in% variable_features
"MS4A1" %in% variable_features

This is only a sanity check. A gene does not need to be variable to be biologically meaningful.

Visualize Variable Features

Plot variable features:

variable_feature_plot <- Seurat::VariableFeaturePlot(
  object = seu,
  log = NULL,               # decide automatically based on data
  col = c("black", "red"),  # black = ordinary genes, red = variable genes
  pt.size = 1               # point size
)

variable_feature_plot

Label top features:

top_variable_genes <- head(variable_features, 10)

variable_feature_plot <- Seurat::LabelPoints(
  plot = variable_feature_plot,
  points = top_variable_genes,
  repel = TRUE, # avoid label overlap
  xnudge = 0.3,
  ynudge = 0.05
)

variable_feature_plot

Note

Variable feature selection is a preprocessing step for dimensionality reduction.

It should not be interpreted as a final list of disease genes, marker genes, or differentially expressed genes.