Detect local outliers in spatial transcriptomics data

localOutliers() compares each spot's quality-control metric with the same metric in its nearest neighbors. It supports SpatialExperiment and Seurat objects and stores the resulting robust z-scores and outlier calls in the object's column metadata.

Usage

localOutliers(
  spe,
  metric = "detected",
  direction = "lower",
  n_neighbors = 36,
  samples = "sample_id",
  log = TRUE,
  cutoff = 3,
  workers = 1,
  coords = NULL
)

Arguments

spe: A SpatialExperiment or Seurat object.
metric: A single column name in the object's column metadata containing a numeric QC metric.
direction: Direction of outlier detection: "higher", "lower", or "both".
n_neighbors: Number of nearest neighbors used as the local reference.
samples: A single metadata column name containing sample identifiers.
log: Whether to apply log1p() to metric before scoring.
cutoff: Non-negative absolute robust z-score cutoff.
workers: Number of workers passed to BiocNeighbors::findKNN().
coords: Optional numeric coordinate matrix with spots as rows. When NULL, spatial coordinates are obtained from spe.

Value

The input object with <metric>_z and <metric>_outliers metadata columns. When log = TRUE, <metric>_log is also added.

Details

The robust z-score for spot i is qnorm(0.75) * (x_i - median(x_neighbors)) / MAD_unscaled(x_neighbors). The focal spot is not included when estimating the neighborhood median or median absolute deviation. A score of zero is returned when the neighborhood MAD is zero or non-finite.

Examples

library(SpotSweeper)
library(SpatialExperiment)

spe <- STexampleData::Visium_humanDLPFC()
#> see ?STexampleData and browseVignettes('STexampleData') for documentation
#> loading from cache
rownames(spe) <- rowData(spe)$gene_name
spe <- spe[, spe$in_tissue == 1]
spe <- spe[, !is.na(spe$ground_truth)]

is.mito <- grepl("^MT-", rownames(spe))
spe <- scuttle::addPerCellQCMetrics(
    spe,
    subsets = list(Mito = is.mito)
)
#> Warning: 'scuttle::addPerCellQCMetrics' is deprecated.
#> Use 'scrapper::quickRnaQc.se' instead.
#> See help("Deprecated")
#> Warning: 'perCellQCMetrics' is deprecated.
#> Use 'scrapper::computeRnaQcMetrics' instead.
#> See help("Deprecated")
#> using unknown matrix fallback for 'dgTMatrix'

spe <- localOutliers(
    spe,
    metric = "sum",
    direction = "lower",
    log = TRUE
)

# Seurat objects use the same interface. Custom coordinates can also be
# supplied for either object type.
if (requireNamespace("SeuratObject", quietly = TRUE)) {
    counts <- matrix(
        seq_len(24),
        nrow = 4,
        dimnames = list(paste0("gene", 1:4), paste0("spot", 1:6))
    )
    seurat <- SeuratObject::CreateSeuratObject(counts)
    seurat$sample_id <- "sample"
    seurat$detected <- c(100, 10, 11, 9, 12, 10)
    example_coords <- cbind(x = seq_len(6), y = 0)
    seurat <- localOutliers(
        seurat,
        n_neighbors = 4,
        coords = example_coords
    )
}
#> Warning: Data is of class matrix. Coercing to dgCMatrix.