knn_graph {speakeasyR}

K-nearest neighbors graph

Description

Create a directed sparse graph with edges to each nodes k nearest neighbors. Nearness is calculated as the inverse of the euclidean distance between two columns.

Usage

knn_graph(mat, k, weighted = FALSE)

Arguments

mat

A matrix to be compared column-by-column.

k

How many nearest neighbors to collect.

weighted

By default, a binary edge is made between a node and each of it's k closest nodes. Set weighted to TRUE to weigh each edge by the similarity (inverse of euclidean distance).

Value

A directed sparse adjacency matrix with k * ncol(mat) nonzero edges. Each column has k edges connected to the k closest columns (not including itself).

Examples

# Simple random graph
mat <- matrix(runif(100) > 0.75, nrow = 5)
knn_graph(mat, 3)

## Don't run because loading data is slow.

if (requireNamespace("scRNAseq") &&
  requireNamespace("SummarizedExperiment")) {
  # Single Cell RNA data
  library(Matrix)

  expression <- scRNAseq::FletcherOlfactoryData()
  cell_types <- expression$cluster_id

  ## Filter genes with low expression. Remove any genes with less than 10
  ## cells with with any reads.
  counts <- SummarizedExperiment::assay(expression, "counts")
  indices <- rowSums(counts > 0) > 10
  counts <- counts[indices, ]

  ## Normalize by shifted logarithm
  target <- median(colSums(counts))
  size_factors <- colSums(counts) / target
  counts_norm <- log(t(t(counts) / size_factors + 1))

  ## Dimension reduction
  counts_norm <- t(prcomp(t(counts_norm), scale. = FALSE)$x)[1:50, ]

  adj <- knn_graph(counts_norm, 10)
}