multivarNetwork: Fit a collection of networks from multivariate experiments

In jchiquet/multivarNetwork: Infer Multivariate Gaussian Graphical Models from Multiattribute Experiments

Description

The inference procedure is multivariate neighborhood selection with group Lasso penalty. If an univariate experiment is provided, the usual neighborhood selection with lasso penalty applies.

Usage

multivarNetwork(X, sym.rule = "AND", select = c("none", "1se", "min",
  "stabsel"), nlambda = 50, min.ratio = 0.001, mc.cores = 1, nfold = 5,
  cutoff = 0.75)

Arguments

X	a list of matrices with the same dimension, corresponding to multiple experiments related to the same individuals and variables, e.g., proteomics and transcriptomics data.
sym.rule	a character, either "AND" or "OR" to define the post-symmetrization rule applied to the coefficients in neighborhood selection
select	a character for defining the cross-validation rule used to choose the penalty level (i.e., the number of edges in the the network). If "none", the whole regulariazation paths is sent back. If "min" or "1se", penalties are cross-validation for each variable with the corresponding rule.
nlambda	integer defining the number of penalty levels used on the grid
min.ratio	the grid of penalties starts by lambda.max, that is, minimal penalty level for selecting no edge in the network, then decrease on a log scale until min.ratio* lambda.max.
mc.cores	for distributing the computation over the variables. Default is 1 (no parallelization).

jchiquet/multivarNetwork documentation built on May 7, 2019, 9:37 a.m.