Create a network — createNetwork • GiottoClass

Create networks from node values. This is a hub function for many different methods of finding nearest neighbors. See details for additional params important for generating specific types of networks.

Usage

createNetwork(
  x,
  type = c("sNN", "kNN", "delaunay"),
  method = c("dbscan", "geometry", "RTriangle", "deldir"),
  node_ids = NULL,
  include_distance = TRUE,
  include_weight = TRUE,
  as.igraph = TRUE,
  verbose = NULL,
  ...
)

Arguments

x: matrix. Data to treat as nodes. Examples include expression information, PCA matrix, spatial locations xy(z) coordinates.
type: type of network to create. Currently: "sNN", "kNN", or "delaunay"
method: method used to create the type of network requested. One of "dbscan" for sNN and kNN or "geometry", "RTriangle", or "deldir" for delaunay.
node_ids: character. Node ID values to assign. If NULL, integer indices will be used as node IDs.
include_distance: logical. include edge distance attribute in output
include_weight: logical. include edge weight attribute in output
as.igraph: logical. Whether to return as igraph. Otherwise returns as data.table
verbose: be verbose. Default = NULL (uses giotto verbosity option)
...: additional params to pass. See details section

Value

Either igraph if as.igraph = TRUE and data.table otherwise.

Details

Additional params are described below. Items in parenthesis refer to which network types and/or methods the params are specific to.

[k] numeric. (sNN, kNN) number of neighbors to find. Default is 30
[minimum_shared] numeric. (sNN) minimum shared neighbors allowed per edge
[top_shared] numeric. (sNN) keep at least this many edges per node, where kept edges are top ranked in terms of number of shared neighbors.
[filter] logical. (kNN) whether to filter for only unique edges and apply minimum_k and maximum_distance filters. Should be set TRUE when generating a spatial kNN network. Default is FALSE.
[minimum_k] (delaunay, kNN) minimum nearest neighbours if maximum_distance != NULL
[maximum_distance] (delaunay, kNN) edge maximum euclidean distance allowed
[Y] (RTriangle) If TRUE prohibits the insertion of Steiner points on the mesh boundary. Default is TRUE
[j] (RTriangle) If TRUE jettisons vertices that are not part of the final triangulation from the output. Default is TRUE
[S] (RTriangle) Specifies the maximum number of added Steiner points. Default is 0
[options] (geometry) default is "Pp". See geometry::delaunayn
[weight_fun] function to calculate weights based on distance if include_weight = TRUE. Default is \(weight = 1 / (1 + distance)\) for "kNN" and "sNN" types and \(weight = 1 / distance\) for delaunay type networks

Examples

pca <- GiottoData::loadSubObjectMini("dimObj")[]
sl <- GiottoData::loadSubObjectMini("spatLocsObj")[]

# Delaunay via geometry::delaunayn()
del_geom <- createNetwork(
    x = as.matrix(sl[, .(sdimx, sdimy)]),
    type = "delaunay",
    method = "geometry",
    include_weight = TRUE,
    weight_fun = function(d) 1 / d,
    as.igraph = FALSE,
    node_ids = sl$cell_ID
)

# Delaunay via RTriangle::triangulate()
del_rt <- createNetwork(
    x = as.matrix(sl[, .(sdimx, sdimy)]),
    type = "delaunay",
    method = "RTriangle",
    include_weight = TRUE,
    weight_fun = function(d) 1 / d,
    as.igraph = FALSE,
    node_ids = sl$cell_ID
)

# Delaunay via deldir::deldir()
del_dd <- createNetwork(
    x = as.matrix(sl[, .(sdimx, sdimy)]),
    type = "delaunay",
    method = "deldir",
    include_weight = TRUE,
    weight_fun = function(d) 1 / d,
    as.igraph = FALSE,
    node_ids = sl$cell_ID
)

# kNN spatial network
kNN_spat <- createNetwork(
    x = as.matrix(sl[, .(sdimx, sdimy)]),
    type = "kNN",
    method = "dbscan",
    include_weight = TRUE,
    weight_fun = function(d) 1 / d, # not the default
    as.igraph = FALSE,
    node_ids = sl$cell_ID,
    k = 4L,
    maximum_distance = NULL,
    minimum_k = 0L
)

# kNN NN network
kNN <- createNetwork(
    pca[, seq_len(10)],
    type = "kNN",
    method = "dbscan",
    node_ids = rownames(pca),
    as.igraph = TRUE
)

# sNN NN network
sNN <- createNetwork(
    pca[, seq_len(10)],
    type = "sNN",
    method = "dbscan",
    node_ids = rownames(pca),
    as.igraph = TRUE
)

# using defaults for sNN with index IDs to create igraph
sNN_idx <- createNetwork(pca[, seq_len(10)])