Smooth Doublet Scores — smoothDoubletScores • scDblFinder

This function applies a distance-weighted kNN smoothing to single-cell doublet probabilities, amplifying high-probability doublet clusters while suppressing isolated noisy predictions.

Usage

smoothDoubletScores(
  x,
  knn = NULL,
  coords = "PCA",
  k = 20,
  alpha = 0.5,
  gamma = 1,
  weights = TRUE,
  decayByDistance = c("linear", "exp", "none"),
  scoreColumn = "scDblFinder.score",
  outColumn = "smoothedDoubletScore",
  ...
)

Arguments

x: A SingleCellExperiment object (containing the 'scDblFinder.score' colData column), or a numeric vector of doublet probabilities.
knn: Optional k nearest neighbors. A list containing index and distance matrices, typically the output of findKNN. If NULL, the kNN graph will be computed.
coords: Either a character scalar indicating the name of the reduced dimension to use for kNN computation, or a matrix of such reduced dimensions, with cells as rows and dimensions as columns. Ignored if knn is given. Default "PCA".
k: Integer. The number of nearest neighbors to compute if knn is NULL. Defaults to 30.
alpha: Numeric [0, 1]. The blending factor between the cell's own score (0) and the neighborhood consensus (1). Defaults to 0.5.
gamma: Numeric >= 1.0. The non-linear amplification exponent. Values above 1 exponentially amplify clusters of high-probability cells. Defaults to 2.0.
weights: Logical. If TRUE (default), uses adaptive Gaussian kernel weighting based on continuous distance. If FALSE, applies uniform weights across all neighbors.
decayByDistance: Whether (and how) to reduce the importance of the neighborhood for isolated cells (based on the first neighbor distance), forcing them to rely primarily on their own raw score. Default 'linear'.
scoreColumn: Character. The column name in colData(x) containing the raw doublet scores. Defaults to "scDblFinder.score".
outColumn: Character. The column name to store the smoothed scores if x is a SingleCellExperiment. Defaults to "smoothedDoubletScore".
...: Passed to findKNN.

Value

If x is a SingleCellExperiment, returns the object with an added colData column containing the smoothed scores. If x is a numeric vector, returns a numeric vector of smoothed scores.