Skip to contents

Create meta-cells from a single cell object using the KNN algorithm. This function is adapted from the R package hdWGCNA (Morabito et al., 2023)

Source: R/cell_deconvolution.R
create_metacells.Rd

Create meta-cells from a single cell object using the KNN algorithm. This function is adapted from the R package hdWGCNA (Morabito et al., 2023)

Usage

create_metacells(
  sc_object,
  labels_column,
  samples_column,
  exclude_cells = NULL,
  min_cells = 50,
  k = 15,
  max_shared = 10,
  n_workers = 4,
  min_meta = 10
)

Arguments

sc_object

A Seurat object with raw counts and a PCA already computed (RunPCA()), used to find each cell's nearest neighbours.

labels_column

Name of the metadata column with the cell type labels.

samples_column

Name of the metadata column with the sample labels.

exclude_cells

Cell types to discard from metacell algorithm.

min_cells

The minimum number of cells in a particular grouping to construct metacells.

k

Number of nearest neighbors to aggregate for KNN algorithm.

max_shared

The maximum number of cells to be shared across two metacells (keep it below k, otherwise metacells can overlap completely).

n_workers

Number of cores to use for paralellization.

min_meta

Minimum number of metacells allowed. Below this number, metacells of this cell type will be discarded.

Value

A list with two elements:

  • The metacell count matrix (genes as rownames and metacells as columns): each metacell is the sum of the counts of its k cells

  • The metadata matrix corresponding to the metacell object

References

Langfelder, P., Horvath, S. WGCNA: an R package for weighted correlation network analysis. BMC Bioinformatics 9, 559 (2008). https://doi.org/10.1186/1471-2105-9-559

Morabito, S., Reese, F., Rahimzadeh, N., Miyoshi, E., & Swarup, V. (2023). hdWGCNA identifies co-expression networks in high-dimensional transcriptomics data. Cell Reports Methods, 3(6), 100498. https://doi.org/10.1016/j.crmeth.2023.100498