nem (version 2.46.0)

set.default.parameters: Get/set hyperparameters

Description

Allows to set and retrieve various hyperparameters for different inference methods.

Usage

set.default.parameters(Sgenes, ...)

Arguments

Sgenes

character vector of S-gene identifiers

...

parameters to set (see details)

Value

A list containing all parameters described above.

Details

Since version 2.5.4 functions in the nem package do not have any more a large amount of individual parameters. Instead there is just one hyperparameter, which is passed to all functions. Parameter values with the hyperparameter can be set with this function.

type

mLL or FULLmLL or CONTmLL or CONTmLLBayes or CONTmLLMAP or depn. CONTmLLDens and CONTmLLRatio are identical to CONTmLLBayes and CONTmLLMAP and are still supported for compatibility reasons. mLL and FULLmLL are used for binary data (see BoutrosRNAiDiscrete) and CONTmLL for a matrix of effect probabilities. CONTmLLBayes and CONTmLLMAP are used, if log-odds ratios, p-value densities or any other model specifies effect likelihoods. CONTmLLBayes refers to an inference scheme, were the linking positions of effect reporters to network nodes are integrated out, and CONTmLLMAP to an inference scheme, were a MAP estimate for the linking positions is calculated. depn indicates Deterministic Effects Propagation Networks (DEPNs).

para

vector of length two: false positive rate and false negative rate for binary data. Used by mLL

hyperpara

vector of length four: used by FULLmLL() for binary data

Pe

prior of effect reporter positions in the phenotypic hierarchy (same dimension as D). Not used type depn. Default: NULL

Pm

prior over models (n x n matrix). Default: NULL

Pm.frac_edges

expected fraction of edges in the true S-gene graph

Pmlocal

local model prior for pairwise and triple learning. For pairwise learning generated by local.model.prior according to arguments local.prior.size and local.prior.bias

local.prior.size

prior expected number of edges in the graph (for pairwise learning). Default: no. nodes

local.prior.bias

bias towards double-headed edges. Default: 1 (no bias; for pairwise learning)

triples.thrsh

threshold for model averaging to combine triple models for each edge. Default: 0.5

lambda

regularization parameter to incorporate prior assumptions. May also be a vector of possible values, if nemModelSelection is used, Default: 0 (no regularization)

delta

regularization parameter for automated subset selection of effect reporters. If no E-gene selection is wanted, set delta to 0. Default: 1/ (no. S-genes + 1)

selEGenes.method

If "regularization", E-gene selection is performed by introducing a "null" S-gene to which E-genes are attached with probability delta/ (no. S-genes + 1). If "iterative" and selEGenes=TRUE, getRelevantEGenes is called and a new model is trained on the selected E-genes. The process is then repeated until convergence. Default: "regularization"

selEGenes

Tune parameter delta for automated selection of E-genes. Default: FALSE. NOTE: Since version > 2.18.0 E-gene selection is now performed per default by using the regularization mechanism with parameter delta. If no E-gene selection is wanted, set delta to 0.

trans.close

Should always transitive closed graphs be computed? Default: TRUE. NOTE: This has only an impact for type nem.greedyMAP and depn. Default: TRUE

backward.elimination

For module networks and greedy hillclimbing inference: Try to eliminate edges increasing the likelihood. Works only, if trans.close=FALSE. Default: FALSE

mode

For Bayesian network inference and DEPNs: binary_ML: effects come from a binomial distribution - ML learning of parameters (Bayesian networks only); binary_Bayesian: effects come from a binomial distribution - Bayesian learning of parameters (Bayesian networks only); continuous_ML: effects come from a normal distribution - ML learning of parameters; continuous_Bayesian: effects come from a normal distribution - Bayesian learning of parameters.

nu.intervention, lambda.intervention

For depn: For any perturbed node we suppose the unknown mean mu given its unknown variance sigma2 to be drawn from N(nu.intervention, sigma2/lambda.intervention). Default: nu.intervention=0.6, lambda.intervention=4

nu.no\_intervention, lambda.no\_intervention

The same parameters for unperturbed nodes. Default: nu.no\_intervention=0.95, lambda.no\_intervention=4

df.intervention, scale.intervention

For depn: The unknown variance sigma2 for perturbed nodes is supposed to be drawn from Inv-$\chi^2$(df.intervention, scale.intervention). Default: df.intervention=4.4, scale.intervention=4.4

df.no\_intervention, scale.no\_intervention

The same parameters for unperturbed nodes. Default: df.no\_intervention=4.4, scale.no\_intervention=0.023

map

For depn: Mapping of interventions to network nodes. The format is a named list of strings with names being the interventions and entries being the network nodes. Default: Entries and names are the network nodes.

outputdir

Directory where to put diagnostic plots. Default: folder "QualityControl" in current working directory

debug

Print out or plot diagnostic information. Default: FALSE

mc.cores

number of cores to be used on a multicore processor. Default: 8

mcmc.nsamples

Number of MCMC samples to take. Default: 1e6

mcmc.nburnin

Number of additional samples for burnin phase. Default: 1e6

mcmc.seed

random seed. Default: 1234

mcmc.hyperprior

Parameter for exponential distribution hyperprior for regularization parameter 1/lambda. Default: 1

eminem.maxsteps

Maximum number of iterations for the EM algorithm (MC.EMINEM). Default: 1000

eminem.sdVal

positive number, between 1 and ncol(D)*(ncol(D)-1): number of edges to change in one MCMC step; see paper for the author's choice. Default: 1

eminem.changeHfreq

positive number, mcmc.nsamples must be a multiple: the Empirical Bayes step is performed every steps (see paper for the author's choice); set >= mcmc.nsamples (or leave to default) to exclude this step. Default: NULL

prob.cutoff

Only edges with probability > prob.cutoff are assumed to be present. Default: 0.5

References

Markowetz, F.; Bloch, J. & Spang, R., Non-transcriptional Pathway Features Reconstructed from Secondary Effects of RNA interference. Bioinformatics, 2005, 21, 4026 - 4032\

Markowetz, F.; Kostka, D.; Troyanskaya, O. & Spang, R., Nested Effects Models for High-dimensional Phenotyping Screens. Bioinformatics, 2007, 23, i305 - i312\

Fr\"ohlich, H.; Fellmann, M.; S\"ultmann, H.; Poustka, A. & Beissbarth, T. Large Scale Statistical Inference of Signaling Pathways from RNAi and Microarray Data. BMC Bioinformatics, 2007, 8, 386\

Fr\"ohlich, H.; Fellmann, M.; S\"ultmann, H.; Poustka, A. & Beissbarth, T. Estimating Large Scale Signaling Networks through Nested Effect Models with Intervention Effects from Microarray Data. Bioinformatics, 2008, 24, 2650-2656\

Tresch, A. & Markowetz, F., Structure Learning in Nested Effects Models Statistical Applications in Genetics and Molecular Biology, 2008, 7\

Zeller, C.; Fr\"ohlich, H. & Tresch, A., A Bayesian Network View on Nested Effects Models EURASIP Journal on Bioinformatics and Systems Biology, 2009, 195272\

Fr\"ohlich, H.; Tresch, A. & Beissbarth, T., Nested Effects Models for Learning Signaling Networks from Perturbation Data. Biometrical Journal, 2009, 2, 304 - 323\

Fr\"ohlich, H.; Sahin, \"O.; Arlt, D.; Bender, C. & Beissbarth, T. Deterministic Effects Propagation Networks for Reconstructing Protein Signaling Networks from Multiple Interventions. BMC Bioinformatics, 2009, 10, 322\

Fr\"ohlich, H.; Praveen, P. & Tresch, A., Fast and Efficient Dynamic Nested Effects Models. Bioinformatics, 2011, 27, 238-244\

Niederberger, T.; Etzold, S.; Lidschreiber, M; Maier, K.; Martin, D.; Fr\"ohlich, H.; Cramer, P.; Tresch, A., MC Eminem Maps the Interaction Landscape of the Mediator, PLoS Comp. Biol., 2012, submitted.

Examples

control = set.default.parameters(LETTERS[1:5], type="CONTmLLBayes", selEGenes=TRUE) # set inference type and whether to use automatic E-gene selection for a network with nodes "A"-"E".