Skip to contents

Leiden clustering of landmarks

Source: R/lm.graph.embed.R
lm.cluster.Rd

Applies Leiden community detection to partition landmarks into clusters based on either the SNN graph or UMAP fuzzy graph. Clusters are used for visualization and as a coarser grouping for statistical testing. This is a wrapper around leiden.cluster.

Usage

lm.cluster(x, ...)

# S3 method for class 'TDRObj'
lm.cluster(
  x,
  .cl.method = "snn",
  .cl.resolution.parameter = 0.8,
  .seed = 123,
  .verbose = TRUE,
  .small.size = 3,
  .column.name = NULL,
  ...
)

Arguments

x

A TDRObj, Seurat, SingleCellExperiment, or HDF5AnnData (anndataR) object.

...

Additional arguments passed to methods.

.cl.method

Character specifying clustering method: "snn" (shared nearest neighbors) or "fgraph" (fuzzy graph from UMAP). Default "snn".

.cl.resolution.parameter

Numeric resolution for Leiden clustering (default 0.8). Internally scaled by 1e-3 for CPM objective. Higher values yield finer/more clusters.

.seed

Integer seed for reproducibility (default 123).

.verbose

Logical for progress messages (default TRUE).

.small.size

Integer threshold for straggler absorption (default 3). Clusters smaller than this are merged into most connected neighbors.

.column.name

Character name for storing the clustering solution (default NULL, which auto-generates paste0("leiden.res.", .cl.resolution.parameter)). Cannot be "ids", which is reserved for the active solution.

Value

Updated .tdr.obj with @landmark.annot$clustering updated:

  • $ids: Factor of cluster assignments for each landmark (active solution)

  • $[[.column.name]]: Same factor, stored as a named solution for multi-solution workflows

Details

The fuzzy graph method ("fgraph") uses probabilistic UMAP edges (pruned with tolerance 1/20 to remove weak connections) and may produce more visually coherent clusters in UMAP space. The SNN method uses Jaccard similarity of k-NN overlaps and often gives more stable results for downstream statistics.

See also

leiden.cluster, get.graph

Examples

if (FALSE) { # \dontrun{
# After graph construction
lm.cells <- setup.tdr.obj(.cells = .cells, .meta = .meta) |>
  get.landmarks() |>
  get.graph()

# Extract clustering (called internally by get.graph)
lm.cells <- lm.cluster(lm.cells)

# Higher resolution clustering
lm.cells <- lm.cluster(lm.cells, .cl.resolution.parameter = 200)

# Use fuzzy graph instead of SNN
lm.cells <- lm.cluster(lm.cells, .cl.method = "fgraph")
} # }