Compute Simpson index from a KNN graph

Given nearest-neighbor distances, nearest-neighbor indices, and a categorical label for each cell, compute the local Simpson index of each cell after matching the target perplexity.

Usage

compute_simpson_index(
  D,
  knn_idx,
  batch_labels,
  perplexity = 15,
  tol = 1e-05,
  max_iter = 50
)

Arguments

D

Numeric matrix of nearest-neighbor distances with neighbors in rows and cells in columns.

knn_idx

Integer matrix of nearest-neighbor indices with the same shape as D. One-based indices are expected; zero-based indices are also accepted and converted automatically.

batch_labels

Integer-like label vector of length equal to the number of cells.

perplexity

Effective neighborhood size. Defaults to 15.

tol

Tolerance used in the binary search for the target perplexity. Defaults to 1e-5.

max_iter

Maximum number of binary-search iterations. Defaults to 50.

Value

A numeric vector containing the local Simpson index for each cell.

Examples

D <- matrix(
  c(0.1, 0.2, 0.2, 0.1,
    0.3, 0.4, 0.4, 0.3),
  nrow = 2,
  byrow = TRUE
)
knn_idx <- matrix(
  c(2, 1, 4, 3,
    3, 4, 2, 1),
  nrow = 2,
  byrow = TRUE
)
batch_labels <- c(1, 1, 2, 2)
compute_simpson_index(D, knn_idx, batch_labels, perplexity = 2)
#> [1] 0.5000049 0.5000049 0.5000049 0.5000049