build_knn_graph: Build kNN graph

Description

Build kNN graph either from distance (from == "dist") or from coordinates (from == "coordinates")

Usage

build_knn_graph(
  X,
  k = 5,
  from = c("dist", "coordinates"),
  use.nn2 = TRUE,
  return_neighbors_order = FALSE,
  dist_method = "euclidean",
  cor_method = "pearson",
  p = 2,
  directed = FALSE,
  DoSNN = FALSE,
  which.snn = c("bluster", "dbscan"),
  pruning = NULL,
  kmin = 0,
  ...
)

Value

a list with components

graph.knn - igraph object
order - Nxk matrix with indices of k nearest neighbors ordered by relevance (from 1st to k-th)

Arguments

X: either distance or matrix of coordinates (rows are samples and cols are coordinates)
k: kNN parameter
from: from which data type to build kNN network: "dist" if X is a distance (dissimilarity) or "coordinates" if X is a matrix with coordinates as cols and cells as rows
use.nn2: whether use nn2 method to buid kNN network faster (available only for "coordinates" option)
return_neighbors_order: whether return order of neighbors (not available for nn2 option)
dist_method: method to compute dist (if X is a matrix of coordinates) available: c("cor", "euclidean", "maximum", "manhattan", "canberra", "binary", "minkowski")
cor_method: if distance is computed as correlation (dist_method == "cor), which type of correlation to use (available: "pearson", "kendall", "spearman")
p: p param in "dist" function
directed: whether to build a directed graph
DoSNN: whether to apply shared nearest neighbors (default is FALSE)
which.snn: whether to use neighborsToSNNGraph or sNN for sNN graph construction
pruning: quantile to perform edge pruning (default is NULL - no pruning applied) based on PCA distance distribution
kmin: keep at least kmin edges in single-cell graph when pruning applied (idnored if is.null(pruning))
...: other parameters of neighborsToSNNGraph or sNN