The simone
function offers an interface to infer networks based
on partial correlation coefficients in various contexts and methods
(steady-state data, time-course data, multiple sample setup,
clustering prior)
simone(X,
type = "steady-state",
clustering = FALSE,
tasks = factor(rep(1, nrow(X))),
control = setOptions())
a \(n\times p\) matrix of data, typically \(n\)
expression levels associated to the same \(p\) genes. Can also
be a data.frame
with \(n\) entries, each column
corresponding to a variable (a gene). Specifying colnames
to
X
may be convenient in view of results analysis, since it
will be used to annotate the plots. Note that this is the only
required argument.
a character string indicating the data specification (either
"steady-state"
or "time-course"
data). Default is
"steady-state"
.
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 FALSE
.
A factor with \(n\) entries indicating the task belonging
for each observation in the multiple sample framework. Default is
factor(rep(1, nrow(X)))
, that is, all observations come from a
unique homogeneous sample.
A list that is used to specify low-level options for the
algorithm, defined through the setOptions
function.
Returns an object of class simone
, which is list-like and
contains the following:
a list with all the inferred networks stocked as adjacency matrices (the successive values of
controled by the penalty level
). In
the multiple sample setup, each element of the list is a list with
as many entries as samples or levels in tasks
.
a vector of the same length as networks
, containing the
successive values of the penalty level.
a vector of the same length as networks
, containing the
successive numbers of edges in the inferred networks. In the multiple
sample setup, n.edges
is a matrix with as many columns as
levels in tasks
.
a vector of the same length as networks
, containing the value
of the BIC for the successively estimated networks.
a vector of the same length as networks
, containing the value
of the AIC for the successively estimated networks.
a size-\(p\) factor indicating the class of each variable.
a \(p\times p\) matrix of weigths used to adapt the penalty to
each entry of the Theta
matrix. It is inferred through the
algorithm according to the latent clustering of the network. When
clustering
is set to FALSE
, all the weights are equal
to "1", which mean no adaptive penalization.
a list describing all the posterior values of the parameters used by
the algorithm, to compare with the one set by the
setOptions
function. As a matter of fact, many of the options
are defined depending on the nature of the data and can be automatically corrected during internal checks of the coherence of desired options to the characteristics of the data.
Any inference method available ("neighborhood selection",
"graphical-Lasso", "VAR(1) inference" and "multitask learning" - see
simone-package
) relies on an optimization problem under
the general form
where \(\mathcal{L}\) is the log-likelihood of the model (pseudo
log-likelihood for "neighborhood selection") and
is a penalty parameter which controls the sparsity level of the
network. The \(p\times p\) matrix
describes the parameters (basically, the
edges) of the model, while \(\mathbf{Z}\) represents a latent
clustering which is also estimated when the argument clustering
is set to TRUE
.
The model and the penalty function
differ according to the context
(steady-state/time-course data, multitask learning and its associated
coupling effect). For further details on the models, please check the
papers listed in the reference section of
simone-package
.
The criterion displayed during a SIMoNe run is the value of the penalized likelihood for the current values of the estimor corresponding to a given value of the overall penalty level .
The following information criteria are also computed for any value of
and part of the output of simone
. The BIC (Bayesian
Information Criterion)
and the AIC (Akaike Information Criterion)
setOptions
, plot.simone
,
cancer
and demo(package="simone")
.
# NOT RUN {
## load the breast cancer data set
data(cancer)
attach(cancer)
## launch simone with the default parameters and plot results
plot(simone(expr))
# }
# NOT RUN {
## try with clustering now (clustering is achieved on a 30-edges network)
plot(simone(expr, clustering=TRUE, control=setOptions(clusters.crit=30)))
## try the multiple sample
plot(simone(expr, tasks=status))
# }
# NOT RUN {
detach(cancer)
# }
Run the code above in your browser using DataLab