Skip to contents

Computes a 2-D UMAP embedding via the uwot package and stores it under "umap". By default the embedding is computed on the top principal components (see use_pca).

Usage

RunUMAP(
  object,
  dims = 30L,
  use_pca = TRUE,
  n_neighbors = 15L,
  densmap = FALSE,
  slot = "data",
  markers = NULL,
  seed = NULL,
  n_threads = 1L,
  pca_method = "svd",
  ...
)

Arguments

object

A SpatialCellData-class object.

dims

Integer. Number of PCs (or markers) to embed. Default 30.

use_pca

Logical. Embed on PCA space (default TRUE) or directly on the marker matrix.

n_neighbors

Integer. UMAP neighbourhood size. Default 15.

densmap

Logical. Preserve local densities with the approximation of densMAP in uwot (dens_scale = 1 in uwot::umap()) instead of plain UMAP. Default FALSE. For partial density preservation, pass dens_scale through ... instead.

slot

Character. "data" (default) or "counts".

markers

Character vector or NULL.

seed

Integer or NULL. Random seed for reproducibility. The random number stream of the caller is restored afterwards.

n_threads

Integer. Number of worker threads for UMAP neighbour search. Defaults to 1; increase explicitly for larger analyses.

pca_method

Character. PCA solver used by uwot for initialisation or optional dimensionality reduction. Defaults to "svd", using base R for the small marker/PC panels typical of spatial data. Pass "auto" or "irlba" to use an optional truncated solver for larger panels.

...

Passed to uwot::umap().

Value

The object with a "umap" entry in its reductions slot.

Examples

# \donttest{
if (requireNamespace("uwot", quietly = TRUE)) {
  counts <- matrix(rnorm(500), nrow = 50,
                   dimnames = list(NULL, paste0("M", 1:10)))
  coords <- data.frame(x = runif(50), y = runif(50))
  obj <- CreateSpatialObject(counts, coords)
  obj <- RunUMAP(obj, seed = 1)
}
# }