meta_cells: Detection of metacells from single-cell gene expression...

In inferCSN: Inferring Cell-Specific Gene Regulatory Network

Detection of metacells from single-cell gene expression matrix

Description

This function detects metacells from a single-cell gene expression matrix using dimensionality reduction and clustering techniques.

Usage

meta_cells(
  matrix,
  genes_use = NULL,
  genes_exclude = NULL,
  var_genes_num = min(1000, nrow(matrix)),
  gamma = 10,
  knn_k = 5,
  do_scale = TRUE,
  pc_num = 25,
  fast_pca = FALSE,
  do_approx = FALSE,
  approx_num = 20000,
  directed = FALSE,
  use_nn2 = TRUE,
  seed = 1,
  cluster_method = "walktrap",
  block_size = 10000,
  weights = NULL,
  do_median_norm = FALSE,
  ...
)

Arguments

matrix	A gene expression matrix where rows represent genes and columns represent cells.
genes_use	Default is NULL. A character vector specifying genes to be used for PCA dimensionality reduction.
genes_exclude	Default is NULL. A character vector specifying genes to be excluded from PCA computation.
var_genes_num	Default is min(1000, nrow(matrix)). Number of most variable genes to select when genes_use is not provided.
gamma	Default is 10. Coarse-graining parameter defining the target ratio of input cells to output metacells (e.g., gamma=10 yields approximately n/10 metacells for n input cells).
knn_k	Default is 5. Number of nearest neighbors for constructing the cell-cell similarity network.
do_scale	Default is TRUE. Whether to standardize (center and scale) gene expression values before PCA.
pc_num	Default is 25. Number of principal components to retain for downstream analysis.
fast_pca	Default is TRUE. Whether to use the faster irlba algorithm instead of standard PCA. Recommended for large datasets.
do_approx	Default is FALSE. Whether to use approximate nearest neighbor search for datasets with >50000 cells to improve computational efficiency.
approx_num	Default is 20000. Number of cells to randomly sample for approximate nearest neighbor computation when do_approx = TRUE.
directed	Default is FALSE. Whether to construct a directed or undirected nearest neighbor graph.
use_nn2	Default is TRUE. Whether to use the faster RANN::nn2 algorithm for nearest neighbor search (only applicable with Euclidean distance).
seed	Default is 1. Random seed for reproducibility when subsampling cells in approximate mode.
cluster_method	Default is walktrap. Algorithm for community detection in the cell similarity network. Options: walktrap (recommended) or louvain (gamma parameter ignored).
block_size	Default is 10000. Number of cells to process in each batch when mapping cells to metacells in approximate mode. Adjust based on available memory.
weights	Default is NULL. Numeric vector of cell-specific weights for weighted averaging when computing metacell expression profiles. Length must match number of cells.
do_median_norm	Default is FALSE. Whether to perform median-based normalization of the final metacell expression matrix.
...	Additional arguments passed to internal functions.

Value

A matrix where rows represent metacells and columns represent genes.

References

https://github.com/GfellerLab/SuperCell https://github.com/kuijjerlab/SCORPION

Examples

data("example_matrix")
meta_cells_matrix <- meta_cells(
  example_matrix
)
dim(meta_cells_matrix)
meta_cells_matrix[1:6, 1:6]

inferCSN documentation built on April 13, 2025, 5:11 p.m.