SIMoNe algorithm for network inference
Description
The simone
function offers an interface to infer networks based
on partial correlation coefficients in various contexts and methods
(steadystate data, timecourse data, multiple sample setup,
clustering prior)
Usage
1 2 3 4 5 
Arguments
X 
a n x p matrix of data, typically n
expression levels associated to the same p genes. Can also
be a 
type 
a character string indicating the data specification (either

clustering 
a logical indicating if the network inference should be perfomed by
penalizing the edges according to a latent clustering discovered
during the network structure recovery. Default is 
tasks 
A factor with n entries indicating the task belonging
for each observation in the multiple sample framework. Default is

control 
A list that is used to specify lowlevel options for the
algorithm, defined through the 
Details
Any inference method available ("neighborhood selection",
"graphicalLasso", "VAR(1) inference" and "multitask learning"  see
simonepackage
) relies on an optimization problem under
the general form
clustering
is set to TRUE
.
The model and the penalty function
pen_{l1} differ according to the context
(steadystate/timecourse data, multitask learning and its associated
coupling effect). For further details on the models, please check the
papers listed in the reference section of
simonepackage
.
The criterion displayed during a SIMoNe run is the value of the penalized likelihood for the current values of the estimor Θ_{hat}(λ) corresponding to a given value of the overall penalty level λ.
The following information criteria are also computed for any value of
λ
and part of the output of simone
. The BIC (Bayesian
Information Criterion)
and the AIC (Akaike Information Criterion)
Value
Returns an object of class simone
, which is listlike and
contains the following:
networks 
a list with all the inferred networks stocked as adjacency matrices (the successive values of
Θ
controled by the penalty level
λ). In
the multiple sample setup, each element of the list is a list with
as many entries as samples or levels in 
penalties 
a vector of the same length as 
n.edges 
a vector of the same length as 
BIC 
a vector of the same length as 
AIC 
a vector of the same length as 
clusters 
a sizep factor indicating the class of each variable. 
weights 
a pxp matrix of weigths used to adapt the penalty to
each entry of the 
control 
a list describing all the posterior values of the parameters used by
the algorithm, to compare with the one set by the

Note
If nothing particular is specified about the penalty through the
control
list (see setOptions
), the default is to
start from a value of
λ
that ensures an empty network. Then
λ is
progressively shrinked, as close to zero as possible. Along the
shrinkage of
λ,
only networks with different numbers of edges are kept in the final
output.
Author(s)
J. Chiquet
See Also
setOptions
, plot.simone
,
cancer
and demo(package="simone")
.
Examples
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17  ## load the breast cancer data set
data(cancer)
attach(cancer)
## launch simone with the default parameters and plot results
plot(simone(expr))
## try with clustering now (clustering is achieved on a 30edges network)
plot(simone(expr, clustering=TRUE, control=setOptions(clusters.crit=30)))
## Not run:
## try the multiple sample
plot(simone(expr, tasks=status))
## End(Not run)
detach(cancer)
