netdis_one_to_one: Netdis between two graphs

In alan-turing-institute/network-comparison: An implementation of the NetEMD alignment-free network distance measure

Netdis between two graphs

Description

Calculates the different variants of the network dissimilarity statistic Netdis between two graphs. The variants currently supported are Netdis using a gold-standard network, Netdis using no expecations (ref_graph = 0), and Netdis using a Geometric Poisson approximation for the expectation (ref_graph = NULL).

Usage

netdis_one_to_one(
  graph_1 = NULL,
  graph_2 = NULL,
  ref_graph = 0,
  max_graphlet_size = 4,
  neighbourhood_size = 2,
  min_ego_nodes = 3,
  min_ego_edges = 1,
  binning_fn = NULL,
  bin_counts_fn = NULL,
  exp_counts_fn = NULL,
  graphlet_counts_1 = NULL,
  graphlet_counts_2 = NULL,
  graphlet_counts_ref = NULL
)

Arguments

graph_1	A simple graph object from the igraph package. graph_1 can be set to NULL (default) if graphlet_counts_1 is provided. If both graph_1 and graphlet_counts_1 are not NULL, then only graphlet_counts_1 will be considered.
graph_2	A simple graph object from the igraph package. graph_2 can be set to NULL (default) if graphlet_counts_2 is provided. If both graph_2 and graphlet_counts_2 are not NULL, then only graphlet_counts_2 will be considered.
ref_graph	Controls how expected counts are calculated. Either: 1) A numeric value - used as a constant expected counts value for all query graphs . 2) A simplified igraph object - used as a reference graph from which expected counts are calculated for all query graphs. 3) NULL (Default) - Used for Netdis-GP, where the expected counts will be calculated based on the properties of the query graphs themselves. (Geometric-Poisson approximation).
max_graphlet_size	Generate graphlets up to this size. Currently only 4 (default) and 5 are supported.
neighbourhood_size	Ego network neighborhood size (default: 2).
min_ego_nodes	Filter ego networks which have fewer than min_ego_nodes nodes (default: 3).
min_ego_edges	Filter ego networks which have fewer than min_ego_edges edges (default: 1).
binning_fn	Function used to bin ego network densities. Takes edge densities as its single argument, and returns a named list including, the input densities, the resulting bin breaks (vector of density bin limits), and the vector interval_indexes which states to what bin each of the individual elements in densities belongs to. ego network). If NULL, then the method binned_densities_adaptive with min_counts_per_interval = 5 and num_intervals = 100 is used (Default: NULL).
bin_counts_fn	Function used to calculate expected graphlet counts in each density bin. Takes graphlet_counts, interval_indexes (bin indexes) and max_graphlet_size as arguments. If bin_counts_fn is NULL, (default), it will apply either the approach from the original Netdis paper, or the respective Geometric-Poisson approximation; depending on the values of ref_graph and graphlet_counts_ref.
exp_counts_fn	Function used to map from binned reference counts to expected counts for each graphlet in each ego network of the query graphs. Takes ego_networks, density_bin_breaks, binned_graphlet_counts, and max_graphlet_size as arguments. If exp_counts_fn is NULL, (default), it will apply either the approach from the original Netdis paper, or the respective Geometric-Poisson approximation; depending on the values of ref_graph and graphlet_counts_ref.
graphlet_counts_1	Pre-generated graphlet counts for the first query graph. Matrix containing counts of each graphlet (columns) for each ego-network (rows) in the input graph. Columns are labelled with graphlet IDs and rows are labelled with the ID of the central node in each ego-network. As well as graphlet counts, each matrix must contain an additional column labelled "N" including the node count for each ego network. (default: NULL). If the graphlet_counts_1 argument is defined then graph_1 will not be used. These counts can be obtained with count_graphlets_ego.
graphlet_counts_2	Pre-generated graphlet counts for the second query graph. Matrix containing counts of each graphlet (columns) for each ego-network (rows) in the input graph. Columns are labelled with graphlet IDs and rows are labelled with the ID of the central node in each ego-network. As well as graphlet counts, each matrix must contain an additional column labelled "N" including the node count for each ego network. (default: NULL). If the graphlet_counts_2 argument is defined then graph_2 will not be used. These counts can be obtained with count_graphlets_ego.
graphlet_counts_ref	Pre-generated reference graphlet counts. Matrix containing counts of each graphlet (columns) for each ego-network (rows) in the reference graph. Columns are labelled with graphlet IDs and rows are labelled with the ID of the central node in each ego-network. As well as graphlet counts, each matrix must contain an additional column labelled "N" including the node count for each ego network. (default: NULL). If the graphlet_counts_ref argument is defined then ref_graph will not be used.

Value

Netdis statistics between graph_1 and graph_2 for graphlet sizes up to and including max_graphlet_size.

Examples

require(netdist)
require(igraph)
#Set source directory for Virus PPI graph edge files stored in the netdist package.
source_dir <- system.file(file.path("extdata", "VRPINS"), package = "netdist")
# Load query graphs as igraph objects
graph_1 <- read_simple_graph(file.path(source_dir, "EBV.txt"),format = "ncol")
graph_2 <- read_simple_graph(file.path(source_dir, "ECL.txt"),format = "ncol")

#Netdis variant using the Geometric Poisson approximation to remove the background expectation of each network.
netdis_one_to_one(graph_1= graph_1, graph_2= graph_2,  ref_graph = NULL) #This option will focus on detecting more general and global discrepancies between the ego-network structures.

#Comparing the networks via their observed ego counts without centering them (equivalent to using expectation equal to zero). This option, will focus on detecting small discrepancies.
netdis_one_to_one(graph_1= graph_1, graph_2= graph_2,  ref_graph = 0)

# Example of the use of netdis with a reference graph.This option will focus on detecting discrepancies between the networks relative to the ego-network structure of the reference network / gold-standard.
# Two lattice networks of different sizes are used for this example. 
 goldstd_1 <- graph.lattice(c(8,8)) #A reference net
 goldstd_2 <- graph.lattice(c(44,44)) #A reference net
 
 netdis_one_to_one(graph_1= graph_1, graph_2= graph_2,  ref_graph = goldstd_1)
 netdis_one_to_one(graph_1= graph_1, graph_2= graph_2,  ref_graph = goldstd_2)
 
 
 #Providing pre-calculated subgraph counts.
 
 props_1 <- count_graphlets_ego(graph = graph_1)
 props_2 <- count_graphlets_ego(graph = graph_2)
 props_goldstd_1 <- count_graphlets_ego(graph = goldstd_1)
 props_goldstd_2 <- count_graphlets_ego(graph = goldstd_2)
 
#Netdis Geometric-Poisson.
netdis_one_to_one(graphlet_counts_1= props_1,graphlet_counts_2= props_2, ref_graph = NULL)

#Netdis Zero Expectation.
netdis_one_to_one(graphlet_counts_1= props_1,graphlet_counts_2= props_2, ref_graph = 0)

#Netdis using gold-standard network
netdis_one_to_one(graphlet_counts_1= props_1,graphlet_counts_2= props_2, graphlet_counts_ref = props_goldstd_1)
netdis_one_to_one(graphlet_counts_1= props_1,graphlet_counts_2= props_2, graphlet_counts_ref = props_goldstd_2)

alan-turing-institute/network-comparison documentation built on June 7, 2022, 10:41 p.m.