generate_expression {graphsim}

Generate Simulated Expression

Description

Compute simulated continuous expression data from a graph network structure. Requires an igraph pathway structure and a matrix of states (1 for activating and -1 for inhibiting) for link signed correlations, from a vector of edge states to a signed adjacency matrix for use in generate_expression. Uses graph structure to pass a sigma covariance matrix from make_sigma_mat_graph or make_sigma_mat_dist_graph on to rmvnorm. By default data is generated with a mean of 0 and standard deviation of 1 for each gene (with correlations between derived from the graph structure).

Usage

generate_expression(
  n,
  graph,
  state = NULL,
  cor = 0.8,
  mean = 0,
  sd = 1,
  comm = FALSE,
  dist = FALSE,
  absolute = FALSE,
  laplacian = FALSE
)

generate_expression_mat(
  n,
  mat,
  state = NULL,
  cor = 0.8,
  mean = 0,
  sd = 1,
  comm = FALSE,
  dist = FALSE,
  absolute = FALSE,
  laplacian = FALSE
)

Arguments

n

number of observations (simulated samples).

graph

An igraph object. May must be directed if states are used.

state

numeric vector. Vector of length E(graph). Sign used to calculate state matrix, may be an integer state or inferred directly from expected correlations for each edge. May be applied a scalar across all edges or as a vector for each edge respectively. May also be entered as text for "activating" or "inhibiting" or as integers for activating (0,1) or inhibiting (-1,2). Compatible with inputs for plot_directed. Also takes a pre-computed state matrix from make_state if applied to the same graph multiple times.

cor

numeric. Simulated maximum correlation/covariance of two adjacent nodes. Default to 0.8.

mean

mean value of each simulated gene. Defaults to 0. May be entered as a scalar applying to all genes or a vector with a separate value for each.

sd

standard deviations of each gene. Defaults to 1. May be entered as a scalar applying to all genes or a vector with a separate value for each.

comm, absolute, laplacian

logical. Parameters for Sigma matrix generation. Passed on to make_sigma or make_sigma.

dist

logical. Whether a graph distance make_sigma_mat_graph or derived matrix make_sigma_mat_dist_graph is used to compute the sigma matrix (using make_distance).

mat

precomputed adjacency, laplacian, commonlink, or scaled distance matrix (generated by make_distance).

Value

numeric matrix of simulated data (log-normalised counts)

Author(s)

Tom Kelly [e-mail]

See Also

See also make_sigma for computing the Sigma (\Sigma) matrix, make_distance for computing distance from a graph object, and make_state for resolving inhibiting states.

See also plot_directed for plotting graphs or heatmap.2 for plotting matrices.

See also make_laplacian, make_commonlink, or make_adjmatrix for computing input matrices.

See also igraph for handling graph objects.

Other graphsim functions: make_adjmatrix, make_commonlink, make_distance, make_laplacian, make_sigma, make_state, plot_directed()

Other generate simulated expression functions: make_distance, make_sigma, make_state

Examples

# construct a synthetic graph module
library("igraph")
graph_test_edges <- rbind(c("A", "B"), c("B", "C"), c("B", "D"))
graph_test <- graph.edgelist(graph_test_edges, directed = TRUE)

# compute a simulated dataset for toy example
# n = 100 samples
# cor = 0.8 max correlation between samples
# absolute = FALSE (geometric distance by default)
test_data <- generate_expression(100, graph_test, cor = 0.8)
##' # visualise matrix
library("gplots")
# expression data
heatmap.2(test_data, scale = "none", trace = "none",
          col = colorpanel(50, "blue", "white", "red"))
# correlations
heatmap.2(cor(t(test_data)), scale = "none", trace = "none",
          col = colorpanel(50, "white", "red"))
# expected correlations (\eqn{\Sigma})
sigma_matrix <- make_sigma_mat_graph(graph_test, cor = 0.8)
heatmap.2(make_sigma_mat_graph(graph_test, cor = 0.8),
          scale = "none", trace = "none", 
          col = colorpanel(50, "white", "red"))

# compute adjacency matrix for toy example
adjacency_matrix <- make_adjmatrix_graph(graph_test)
# generate simulated data from adjacency matrix input
test_data <- generate_expression_mat(100, adjacency_matrix, cor = 0.8)

# compute a simulated dataset for toy example
# n = 100 samples
# cor = 0.8 max correlation between samples
# absolute = TRUE (arithmetic distance)
test_data <- generate_expression(100, graph_test, cor = 0.8, absolute = TRUE)
##' # visualise matrix
library("gplots")
# expression data
heatmap.2(test_data, scale = "none", trace = "none",
          col = colorpanel(50, "blue", "white", "red"))
# correlations
heatmap.2(cor(t(test_data)),
          scale = "none", trace = "none",
          col = colorpanel(50, "white", "red"))
# expected correlations (\eqn{\Sigma})
sigma_matrix <- make_sigma_mat_graph(graph_test, cor = 0.8)
heatmap.2(make_sigma_mat_graph(graph_test, cor = 0.8),
          scale = "none", trace = "none",
          col = colorpanel(50, "white", "red"))

# construct a synthetic graph network
graph_structure_edges <- rbind(c("A", "C"), c("B", "C"), c("C", "D"), c("D", "E"),
                               c("D", "F"), c("F", "G"), c("F", "I"), c("H", "I"))
graph_structure <- graph.edgelist(graph_structure_edges, directed = TRUE)

# compute a simulated dataset for toy network
# n = 250 samples
# state = edge_state (properties of each edge)
# cor = 0.95 max correlation between samples
# absolute = FALSE (geometric distance by default)
edge_state <- c(1, 1, -1, 1, 1, 1, 1, -1)
structure_data <- generate_expression(250, graph_structure,
                                      state = edge_state, cor = 0.95)
##' # visualise matrix
library("gplots")
# expression data
heatmap.2(structure_data, scale = "none", trace = "none",
          col = colorpanel(50, "blue", "white", "red"))
# correlations
heatmap.2(cor(t(structure_data)), scale = "none", trace = "none",
          col = colorpanel(50, "blue", "white", "red"))
# expected correlations (\eqn{\Sigma})
sigma_matrix <- make_sigma_mat_graph(graph_structure,
                                     state = edge_state, cor = 0.8)
heatmap.2(make_sigma_mat_graph(graph_structure,
                               state = edge_state, cor = 0.8),
          scale = "none", trace = "none",
          col = colorpanel(50, "blue", "white", "red"))

# compute adjacency matrix for toy network
graph_structure_adjacency_matrix <- make_adjmatrix_graph(graph_structure)
# define states for for each edge
edge_state <- c(1, 1, -1, 1, 1, 1, 1, -1)
# generate simulated data from adjacency matrix input
structure_data <- generate_expression_mat(250, graph_structure_adjacency_matrix,
                                          state = edge_state, cor = 0.8)

# compute a simulated dataset for toy network
# n = 1000 samples
# state = TGFBeta_Smad_state (properties of each edge)
# cor = 0.75 max correlation between samples
# absolute = FALSE (geometric distance by default)
 # compute states directly from graph attributes for TGF-\eqn{\Beta} pathway
TGFBeta_Smad_state <- E(TGFBeta_Smad_graph)$state
table(TGFBeta_Smad_state)
# generate simulated data
TGFBeta_Smad_data <- generate_expression(1000, TGFBeta_Smad_graph, cor = 0.75)
##' # visualise matrix
library("gplots")
# expression data
heatmap.2(TGFBeta_Smad_data, scale = "none", trace = "none",
          col = colorpanel(50, "blue", "white", "red"))
# correlations
heatmap.2(cor(t(TGFBeta_Smad_data)), scale = "none", trace = "none",
          dendrogram = "none", Rowv = FALSE, Colv = FALSE,
          col = colorpanel(50, "blue", "white", "red"))
# expected correlations (\eqn{\Sigma})
sigma_matrix <- make_sigma_mat_dist_graph(TGFBeta_Smad_graph, cor = 0.75)
heatmap.2(make_sigma_mat_dist_graph(TGFBeta_Smad_graph, cor = 0.75),
          scale = "none", trace = "none",
          dendrogram = "none", Rowv = FALSE, Colv = FALSE,
          col = colorpanel(50, "blue", "white", "red"))


# generate simulated data (absolute distance and shared edges)
TGFBeta_Smad_data <- generate_expression(1000, TGFBeta_Smad_graph,
                                         cor = 0.75, absolute = TRUE, comm = TRUE)
##' # visualise matrix
library("gplots")
# expression data
heatmap.2(TGFBeta_Smad_data, scale = "none", trace = "none",
          col = colorpanel(50, "blue", "white", "red"))
# correlations
heatmap.2(cor(t(TGFBeta_Smad_data)), scale = "none", trace = "none",
          dendrogram = "none", Rowv = FALSE, Colv = FALSE,
          col = colorpanel(50, "blue", "white", "red"))
# expected correlations (\eqn{\Sigma})
sigma_matrix <- make_sigma_mat_graph(TGFBeta_Smad_graph,
                                     cor = 0.75, comm = TRUE)
heatmap.2(make_sigma_mat_graph(TGFBeta_Smad_graph, cor = 0.75, comm = TRUE),
          scale = "none", trace = "none",
          dendrogram = "none", Rowv = FALSE, Colv = FALSE,
          col = colorpanel(50, "blue", "white", "red"))