Skip to contents

Expand knn graph to increase the number of neighbours

Source: R/expand_knn_graph.R
ExpandNeighbours.Rd

Expand a knn graph to increase the number of nearest neighbours using Dijkstra's algorithm, or a diffusion algorithm. Dijkstra's algorithm is used to prepare for LISI score, while diffusion is suited for preparing to compute the kBET score. Beware that diffusion is designed to work on connectivity matrices and is not adequate for distance-based networks.

Usage

ExpandNeighbours(
  object,
  graph.name = "RNA_nn",
  new.graph.name = NULL,
  graph.type = c("distances", "connectivities"),
  k.target = 90L,
  do.symmetrize = FALSE,
  algo = c("dijkstra", "diffusion"),
  which.dijkstra = c("auto", "igraph", "fast", "slow"),
  dijkstra.ncores = 1L,
  dijkstra.tol = 1L,
  diffusion.iter = 26L,
  assay = NULL,
  verbose = TRUE
)

# S4 method for class 'Seurat'
ExpandNeighbours(
  object,
  graph.name = "RNA_nn",
  new.graph.name = NULL,
  graph.type = c("distances", "connectivities"),
  k.target = 90L,
  do.symmetrize = FALSE,
  algo = c("dijkstra", "diffusion"),
  which.dijkstra = c("auto", "igraph", "fast", "slow"),
  dijkstra.ncores = 1L,
  dijkstra.tol = 1L,
  diffusion.iter = 26L,
  assay = NULL,
  verbose = TRUE
)

Arguments

object

a Seurat, Graph or Neighbor object

graph.name

name of a Graph or Neighbor instance stored in the Seurat object.

new.graph.name

name of the return Graph to store in the Seurat object.

graph.type

One of "distances" or "connectivities", indicating the type of metric stored in the graph object.

k.target

number of nearest neighbours to reach

do.symmetrize

whether to make the input graph symmetric if necessary. See Details section for further explanations

algo

One of "dijkstra" or "diffusion". "diffusion" is suited for connectivity matrices only

which.dijkstra

one of "igraph", "fast" or "slow". "auto" (default) chooses for you. See Details section

dijkstra.ncores

number of cores to use for Dijkstra's algorithm. Ignored when which.dijkstra = "igraph"

dijkstra.tol

number of sequential iterations with identical best neighbours found to consider that Dijkstra's algorithm should be stopped. Ignored when which.dijkstra = "igraph"

diffusion.iter

maximum number of iterations to reach k.target

assay

name of the assay to store in the output Graph

verbose

whether to print progress messages

Value

the Seurat object with a new Graph instance or a dgCMatrix representing the Graph itself

Details

The approximate nature of the nearest neighbour search algorithm used to compute the knn graph makes the resulting adjacency matrix asymmetric. It means that \(cell_i\) can be a nearest neighbour of \(cell_j\) while the reverse is not true even though the distance between \(cell_i\) and \(cell_j\) is lower than the distance between \(cell_j\) and some of its nearest neighbours.

One can choose to keep the graph as it is and consider it as a directed graph (do.symmetrize = FALSE). The alternative solution is to use all computed distances to extend the knn graph by making the matrix symmetric. Note that connectivity graphs are already symmetric, so the argument value should have no effect on the result.