Filters all cell-indexed slots ($cell.list, $cell.data,
$cell.meta, the three neighborhood slots and every table in
$aggregations) to the cells in i, and all edge-indexed slots
($edge.list, $edge.data, $edge.meta) to keep only
edges where both sending and receiving cells are in i.
If $segmentation is populated, it is filtered to retain only rows
whose cell is in the retained set. Slot-level invariants are
re-validated by the internal constructor before the subset is returned.
Value
A new NICHESObject restricted to the retained cells and
their fully-internal edges, with their neighborhood entries and
aggregation rows unchanged. Gene and LRM slots are unchanged.
Details
Cell barcodes in i that are not present in x$cell.list are
silently dropped. If the intersection is empty the function errors rather
than returning an unusable empty object.
Neighborhoods and aggregations after subsetting
A retained cell keeps its neighborhood entries and aggregation rows
unchanged, so they still describe its whole neighborhood in the full
object, including neighbors outside i. Aggregate the full object
first, then subset. Summing a cell's edges in $edge.data can give
less than its aggregation row, because the edge slots hold only edges
between retained cells; a export_to_TissuePlex export of the
subset shows those edges only.
Aggregations cannot be recomputed on a subset whose neighborhoods include
removed edges: add_neighborhood_edge_agg and
add_neighborhood_composition stop with an error. To analyze
the subset as a separate tissue instead, call
add_neighborhoods(obj, overwrite = TRUE) first. This rebuilds the
neighborhoods from the retained edges and clears the carried-over
aggregation tables.
See also
is_NICHESObject to validate the result.
Examples
# \donttest{
set.seed(1L)
count.mtx <- Matrix::sparseMatrix(
i = sample(6L, 30L, replace = TRUE),
j = sample(8L, 30L, replace = TRUE),
x = as.numeric(rpois(30L, lambda = 3) + 1L),
dims = c(6L, 8L),
dimnames = list(paste0("Gene", 1:6), paste0("Cell", 1:8))
)
meta.data <- data.frame(
x = runif(8L), y = runif(8L),
row.names = paste0("Cell", 1:8)
)
LRM.db <- data.frame(
ligand = c("Gene1", "Gene3"),
receptor = c("Gene2", "Gene4")
)
obj <- create_NICHESObject(
count.mtx = count.mtx, meta.data = meta.data, LRM.db = LRM.db,
mode = "spatial", ligand.col = "ligand", receptor.col = "receptor",
k = 3L, n.cores = 1L, verbose = FALSE
)
#> [create_Edgelist_Spatial] Mode: kNN | Param: 3 | Cells: 8 | Inter-cell: 18 | Autocrine: 8
#> [compute_CellToCell] 26 edges x 2 LR pairs | method = 'product' | cores = 1
#> [compute_CellToCell] Done. 9 non-zero edge signals.
# Subset to the first four cells; only fully-internal edges are kept
sub <- obj[paste0("Cell", 1:4)]
dim(sub)
#> cells genes edges LRMs
#> 4 6 6 2
# }