STniche (mclust version): Detect functional spatial domains with model-based clustering

Unsupervised clustering of spots/cells in one or more ST samples using gene-set GSVA scores -> local Moran's I (Z) as features, then Gaussian mixture modeling via mclust with BIC-based model/cluster selection.

STniche(
  x = NULL,
  gene_sets = NULL,
  samples = NULL,
  ks = 2:10,
  knns = 6,
  min_genes = 1,
  out_i = FALSE,
  verbose = 1L,
  cores = NULL
)

Arguments

x

an STlist with normalized expression data in x@tr_counts[[sample]] (genes x spots)

gene_sets

a named list of gene sets (each an atomic character vector of gene symbols)

samples

vector of sample names or indices to run (default: all)

ks

numeric vector of candidate cluster numbers to evaluate with BIC (default: 2:10)

knns

integer k for kNN used in Local Moran's I (default: 6)

min_genes

minimum overlapping genes required to compute a GSVA score (default: 1)

out_i

logical; if TRUE return list(STlist, Z.Ii matrix), else STlist only

verbose

integer verbosity level (0=quiet, 1=basic, 2=chatty)

cores

integer number of workers for parallel scoring (Unix only; default chooses sensibly)

Value

If out_i=TRUE: list(updated_STlist, Zi_matrix) where Zi_matrix has rows = gene sets, columns = "sample&&libname" spots. Otherwise, the updated STlist.

Details

Workflow:

1. For each sample, compute GSVA scores for provided gene sets.

2. For each gene set within each sample, compute Local Moran's I Z-score over a kNN graph (k = knns).

3. Concatenate all samples into a single feature matrix (spots x gene-sets, entries are Z.Ii).

4. Run mclust::Mclust with candidate cluster counts ks (e.g., 2:10). Best model & G selected by BIC.

5. Write cluster labels back into x@spatial_meta[[sample]] under columns stniche_dsplFalse and STniche_DTC for plotting compatibility.