bicluster_network: Generate a bicluster network

In tdrose/mosbi: Molecular Signature identification using Biclustering

Generate a bicluster network

Description

The function computes a bicluster network based on a selected similarity metric. A similarity cut-off is calculated using randomized biclusters (the bicluster size distribution is kept).

Usage

bicluster_network(
  bics,
  mat,
  n_randomizations = 5,
  MARGIN = "both",
  metric = 4,
  n_steps = 100,
  plot_edge_dist = TRUE,
  sn_ratio = TRUE,
  error_threshold = 0.05,
  return_plot_data = FALSE,
  prob_scale = FALSE,
  prl = FALSE
)

Arguments

bics	A list of bicluster objects.
mat	The matrix used for biclustering.
n_randomizations	The number of randomizations for cut-off estimation. (The mean of all randomizations is used).
MARGIN	Margin over which the similarity is computed. Can be "row", "column", "mean" (In this case the mean of row and column similarity is used) or "both" (In this case the similarity between all the datapoints of biclusters is used).
metric	The similarity metric same as in similarity_matrix.
n_steps	Number of points where the difference between randomizations and the real data is evaluated.
plot_edge_dist	Show the plots for cut-off estimation with the error model.
sn_ratio	If TRUE, the signal to noise ratio is computed, otherwise the error_threshold is used to to estimate the cut-off at which only error_threshold*100 percent of the edges are estimated to be random overlaps.
error_threshold	If sn_ratio==FALSE this threshold is used to estimate the threshold at which only error_threshold*100 percent of the edges are estimated to be random overlaps.
return_plot_data	Please do not use outside of the package.
prob_scale	Scale similarity by the probability of an overlap equal of higher to the observed one. The scaling is done by multiplying the similarity with (1 - (1 / (1 - log(overlap_probability, base=100)))). The probability is computed using the f unction p_overlap_2d_higher for MARGIN =="both" and p_overlap_higher otherwise. Can be helpful for big imbalances of bicluster sizes.
prl	Compute the similarity matrix using multiple cores (works only for MARGIN="both"). The number of core can be defined by executing: RcppParallel::setThreadOptions(numThreads = 4) before running this function.

Value

An object of class bicluster_net.

Examples

m <- matrix(seq(1:16), nrow=4)
# m <- matrix(rnorm(10000), nrow=100)
# bics <- c(run_fabia(m), run_isa(m), run_plaid(m))
# bicluster_network(bics, m)

tdrose/mosbi documentation built on May 4, 2022, 3:22 p.m.