Run smoothclust spatial domain clustering

Smooth expression across spatial neighborhoods with the optional smoothclust package, then cluster the smoothed profiles into spatial domains with PCA and k-means.

Usage

RunSmoothClust(
  srt,
  assay = NULL,
  layer = "data",
  image = NULL,
  coord.cols = c("col", "row"),
  features = NULL,
  nfeatures = 2000,
  min_spots = 5,
  smooth_method = c("uniform", "kernel", "knn"),
  bandwidth = 0.05,
  k = 18,
  truncate = 0.05,
  cores = 1,
  n_threads = NULL,
  n_clusters,
  n_pcs = 15,
  center = TRUE,
  scale = TRUE,
  nstart = 10,
  iter.max = 100,
  algorithm = "Hartigan-Wong",
  cluster_colname = "SmoothClust_cluster",
  tool_name = "SmoothClust",
  store_results = TRUE,
  store_smoothed = FALSE,
  seed = 11,
  verbose = TRUE,
  coordinate_space = c("raw", "legacy_display"),
  ...
)

Arguments

srt: A Seurat object.
assay: Assay used for expression. If NULL, the default assay is used.
layer: Assay layer used for expression values.
image: Name of the Seurat spatial image. Required when multiple images are present; a single image is selected automatically when NULL.
coord.cols: Metadata coordinate columns used when no Seurat image is available.
features: Features to use. If NULL, current variable features are used; if no variable features are present, the top nfeatures by variance are used.
nfeatures: Number of variance-ranked features to use when features = NULL and no variable features are present.
min_spots: Minimum number of spots with non-zero expression required for a feature to be used.
smooth_method: Smoothing method passed to smoothclust::smoothclust().
bandwidth, k, truncate: Smoothing parameters passed to smoothclust::smoothclust(). bandwidth uses the selected coordinate units; truncate is the backend's dimensionless kernel cutoff, and k and are unitless counts.
cores: Number of workers passed to smoothclust::smoothclust().
n_threads: Deprecated alias for cores.
n_clusters: Number of spatial domains for k-means clustering. This must be supplied explicitly.
n_pcs: Number of principal components used for k-means.
center, scale: Whether to center and scale features before PCA.
nstart, iter.max, algorithm: Parameters passed to stats::kmeans().
cluster_colname: Metadata column used for smoothclust clusters.
tool_name: Name used to store detailed results in srt@tools.
store_results: Whether to store detailed results in srt@tools.
store_smoothed: Whether to store the smoothed expression matrix in srt@tools[[tool_name]]. This can be large.
seed: Random seed used for k-means.
verbose: Whether to print progress messages.
coordinate_space: Coordinate system used for distance-sensitive smoothing and smoothness calculations. The default is raw acquisition coordinates. Use "legacy_display" explicitly to reproduce the display-scaled behavior used before scop 0.9.0.
...: Additional arguments passed to smoothclust::smoothclust().

Value

A Seurat object with smoothclust clusters in metadata. When store_results = TRUE, detailed outputs are stored in srt@tools[[tool_name]].

Examples

data(visium_human_pancreas_sub)
spatial <- visium_human_pancreas_sub
spatial$SmoothClust_cluster <- factor(
  paste0("SmoothClust", (seq_len(ncol(spatial)) - 1) %% 3 + 1)
)

SpatialSpotPlot(
  spatial,
  group.by = "SmoothClust_cluster",
  overlay_image = FALSE,
  coord.cols = c("x", "y")
)


spatial <- Seurat::NormalizeData(spatial, assay = "Spatial", verbose = FALSE)
spatial <- Seurat::FindVariableFeatures(
  spatial,
  assay = "Spatial",
  nfeatures = 200,
  verbose = FALSE
)

spatial <- RunSmoothClust(
  spatial,
  assay = "Spatial",
  n_clusters = 3,
  smooth_method = "knn",
  coord.cols = c("x", "y"),
  k = 6,
  verbose = FALSE
)
#> ! [2026-08-30 05:38:30] Function "smoothness_metric" not found in smoothclust namespace
#> Warning: Function "smoothness_metric" not found in smoothclust namespace

table(spatial$SmoothClust_cluster)
#> 
#> SmoothClust1 SmoothClust2 SmoothClust3 
#>          886          534          566