Optimise UMAP hyper-parameters for a Seurat object — umap

Sweeps min.dist and n.neighbors, saving a UMAP plot PDF for each combination so the best-looking embedding can be chosen.

Usage

umap_optimise(
  input.seurat,
  output.dir = "results/optimising_UMAP/",
  init.min.dist = 0.001,
  init.neigh = 5,
  min.dist.step = 2,
  neigh.step = 20,
  dims = 1:20,
  max.neigh = 50,
  seed = 42
)

Arguments

input.seurat: A Seurat object with a pca reduction.
output.dir: Directory to write the PDFs into (created if needed).
init.min.dist: Starting min.dist (sensible range 0.001-0.5).
init.neigh: Starting n.neighbors (sensible range 5-50).
min.dist.step: Multiplicative step for min.dist.
neigh.step: Additive step for n.neighbors.
dims: Principal components used for the embedding.
max.neigh: Largest n.neighbors to try.
seed: Random seed for reproducible embeddings.

Value

Invisibly, the paths of the PDFs written.

Examples

if (FALSE) { # \dontrun{
umap_optimise(pbmc_small)
} # }