Cell Composition

Purpose

After cell annotation, a natural next step is to examine cell type composition.

Typical questions:

• how many cells are assigned to each cell type?
• what fraction of each sample is each cell type?
• does cell type composition differ between conditions?

Cell composition is descriptive unless a proper statistical model is used.

Required Metadata

The Seurat object should contain cell type labels:

table(seu$cell_type)

It should also contain sample or condition metadata:

table(seu$sample)
table(seu$condition)

Use sample-level metadata when comparing groups. Do not rely only on pooled cell counts across all samples.

Cell Type Counts

Count cells by cell type:

cell_type_counts <- as.data.frame(
  table(seu$cell_type)
)

colnames(cell_type_counts) <- c("cell_type", "n_cells")

cell_type_counts

Count cells by sample and cell type:

sample_cell_type_counts <- as.data.frame(
  table(seu$sample, seu$cell_type)
)

colnames(sample_cell_type_counts) <- c(
  "sample",
  "cell_type",
  "n_cells"
)

sample_cell_type_counts

Cell Type Proportions

Calculate cell type proportions within each sample:

sample_cell_type_props <- sample_cell_type_counts |>
  dplyr::group_by(sample) |>
  dplyr::mutate(
    prop = n_cells / sum(n_cells)
  ) |>
  dplyr::ungroup()

sample_cell_type_props

Add condition metadata if each sample has one condition:

sample_metadata <- seu@meta.data |>
  dplyr::select(sample, condition) |>
  dplyr::distinct()

sample_cell_type_props <- sample_cell_type_props |>
  dplyr::left_join(
    sample_metadata,
    by = "sample"
  )

Plot Composition

Stacked bar plot by sample:

ggplot2::ggplot(
  data = sample_cell_type_props,
  mapping = ggplot2::aes(
    x = sample,
    y = prop,
    fill = cell_type
  )
) +
  ggplot2::geom_col(width = 0.8) +
  ggplot2::scale_y_continuous(labels = scales::percent) +
  ggplot2::labs(
    x = "Sample",
    y = "Cell type proportion",
    fill = "Cell type"
  ) +
  ggplot2::theme_classic()

Stacked bar plot grouped by condition:

ggplot2::ggplot(
  data = sample_cell_type_props,
  mapping = ggplot2::aes(
    x = condition,
    y = prop,
    fill = cell_type
  )
) +
  ggplot2::geom_col(
    position = "fill",
    width = 0.8
  ) +
  ggplot2::scale_y_continuous(labels = scales::percent) +
  ggplot2::labs(
    x = "Condition",
    y = "Cell type proportion",
    fill = "Cell type"
  ) +
  ggplot2::theme_classic()

Compare One Cell Type

Plot sample-level proportions for one cell type:

target_cell_type <- "T cell"

one_cell_type <- sample_cell_type_props |>
  dplyr::filter(cell_type == target_cell_type)

ggplot2::ggplot(
  data = one_cell_type,
  mapping = ggplot2::aes(
    x = condition,
    y = prop,
    color = condition
  )
) +
  ggplot2::geom_boxplot(outlier.shape = NA) +
  ggplot2::geom_jitter(width = 0.15, height = 0) +
  ggplot2::scale_y_continuous(labels = scales::percent) +
  ggplot2::labs(
    x = "Condition",
    y = paste(target_cell_type, "proportion")
  ) +
  ggplot2::theme_classic()

Each point should represent a sample, not an individual cell.

Note

Cell composition plots are useful for exploration.

For formal differential abundance analysis, use methods designed for sample-level or replicate-aware testing. Do not treat cells from the same sample as independent biological replicates.