Skip to contents

Run PAGA graph abstraction

run_paga_sc.Rd

PAGA abstracts the single cell kNN graph into a cluster-level graph, where the edge weight between two clusters measures how much more connected they are than expected under a random null model. It also returns the maximum spanning forest of that graph, which is the backbone typically plotted. For details, please refer to Wolf, et al.

Only the kNN indices are used, no distances and no expression data, so the method is safe on any of the available graphs, including the WNN one. Run find_neighbours_sc() and a clustering first.

Usage

run_paga_sc(
  object,
  cluster_col,
  modality = c("rna", "adt", "wnn"),
  .verbose = TRUE
)

Arguments

object

One of SingleCells, SingleCellsSubset, MetaCells or SingleCellsMultiModal.

cluster_col

String. The obs column holding the cluster assignments. Empty factor levels are retained.

modality

String. One of c("rna", "adt", "wnn"). Which kNN graph to run over. Anything but "rna" requires a SingleCellsMultiModal object.

.verbose

Boolean or integer. Controls verbosity and returns run times. FALSE -> quiet, TRUE or 1L -> normal verbosity, 2L -> detailed verbosity.

Value

A PagaRes S3 object with:

  • connectivities - Symmetric sparse matrix of clusters x clusters with a zero diagonal. Values lie in (0, 1].

  • connectivities_tree - Maximum spanning forest of connectivities, carrying the original connectivity values on the retained edges.

  • sizes - data.table with cluster and n_cells.

  • params - List with cluster_col and modality.

References

Wolf, et al., Genome Biol., 2019.

Examples

# cluster level graph abstraction of the Leiden clusters
sc <- demo_single_cells()
sc <- find_clusters_sc(sc, res = 1.0)
res <- run_paga_sc(sc, cluster_col = "leiden_clustering", .verbose = FALSE)
res$sizes
#>    cluster n_cells
#>     <char>   <int>
#> 1:       0     169
#> 2:       1     166
#> 3:       2     165

unlink(sc@dir_data, recursive = TRUE, force = TRUE)