Usage
detect.responses(datamatrix, network = NULL, initial.responses = 1,
max.responses = 10, max.subnet.size = 10, verbose = TRUE,
prior.alpha = 1, prior.alphaKsi = 0.01, prior.betaKsi = 0.01,
update.hyperparams = 0, implicit.noise = 0, vdp.threshold = 1e-05,
merging.threshold = 0, ite = Inf, information.criterion = "BIC",
speedup = TRUE, speedup.max.edges = 10, positive.edges = FALSE,
mc.cores = 1, mixture.method = "vdp", bic.threshold = 0,
pca.basis = FALSE, ...)
Arguments
datamatrix
Matrix of samples x features. For example, gene expression
matrix with conditions on the rows, and genes on the columns. The matrix
contains same features than the 'network' object, characterizing the network
states across the different samples.
network
Binary network describing undirected pairwise interactions between
features of 'datamatrix'. The following formats are supported: binary
matrix, graphNEL, igraph, graphAM, Matrix, dgCMatrix, dgeMatrix
initial.responses
Initial number of components for each subnetwork
model. Used to initialize calculations.
max.responses
Maximum number of responses for each subnetwork. Can be
used to limit the potential number of network states.
max.subnet.size
Numeric. Maximum allowed subnetwork size.
verbose
Logical. Verbose parameter.
prior.alpha, prior.alphaKsi, prior.betaKsi
Prior parameters for
Gaussian mixture model that is calculated for each subnetwork
(normal-inverse-Gamma prior). alpha tunes the mean; alphaKsi and betaKsi are
the shape and scale parameters of the inverse Gamma function, respectively.
update.hyperparams
Logical. Indicate whether to update
hyperparameters during modeling.
implicit.noise
Implicit noise parameter. Add implicit noise to vdp
mixture model. Can help to avoid overfitting to local optima, if this
appears to be a problem.
vdp.threshold
Minimal free energy improvement after which the
variational Gaussian mixture algorithm is deemed converged.
merging.threshold
Minimal cost value improvement required for merging
two subnetworks.
ite
Defines maximum number of iterations on posterior update
(updatePosterior). Increasing this can potentially lead to more accurate
results, but computation may take longer.
information.criterion
Information criterion for model selection.
Default is BIC (Bayesian Information Criterion); other options include AIC
and AICc.
speedup
Takes advantage of approximations to PCA, mutual information
etc in various places to speed up calculations. Particularly useful with
large and densely connected networks and/or large sample size.
speedup.max.edges
Used if speedup = TRUE. Applies prefiltering of
edges for calculating new joint models between subnetwork pairs when
potential cost changes (delta) are updated for a newly merged subnetwork and
its neighborghs. Empirical mutual information between each such subnetwork
pair is calculated based on their first principal components, and joint
models will be calculated only for the top candidates up to the number
specified by speedup.max.edges. It is expected that the subnetwork pair that
will benefit most from joint modeling will be among the top mutual
infomation candidates. This way it is possible to avoid calculating
exhaustive many models on the network hubs.
positive.edges
Consider only the edges with positive association. Currently measured with Spearman correlation.
mc.cores
Number of cores to be used in parallelization. See
help(mclapply) for details.
mixture.method
Specify the approach to use in mixture modeling.
Options. vdp (nonparametric Variational Dirichlet process mixture model);
bic (based on Gaussian mixture modeling with EM, using BIC to select the
optimal number of components)
bic.threshold
BIC threshold which needs to be exceeded before a new mode is added to the mixture with mixture.method = "bic"
pca.basis
Transform data first onto PCA basis to try to avoid problems with non-diagonal covariances.
...
Further optional arguments to be passed.