score {nem}R Documentation

Computes the marginal likelihood of phenotypic hierarchies

Description

Function to compute the marginal likelihood of a set of phenotypic hierarchies.

Usage

score(models, D, type="mLL", para=NULL, hyperpara=NULL, Pe=NULL, Pm=NULL, lambda=0, selEGenes=FALSE, verbose=TRUE, graphClass="graphNEL")

# S3 metehods for class 'score'
print.score(x, ...)

PhiDistr(Phi, Pm, a=1, b=ncol(Phi)^2)

Arguments

models a list of adjacency matrices with unit main diagonal
D data matrix. Columns correspond to the nodes in the silencing scheme. Rows are effect reporters.
type (1.) marginal likelihood "mLL" (only for cout matrix D), or (2.) full marginal likelihood "FULLmLL" integrated over a and b and depending on hyperparameters a0, a1, b0, b1 (only for count matrix D), or (3.) "CONTmLL" marginal likelihood for probability matrices, or (4.) "CONTmLLDens" marginal likelihood for probability density matrices, or (5.) "CONTmLLRatio" for log-odds ratio matrices
para Vector with parameters a and b (for "mLL" with count data)
hyperpara Vector with hyperparameters a0, b0, a1, b1 for "FULLmLL"
Pe prior position of effect reporters. Default: uniform over nodes in silencing scheme
Pm prior on model graph (n x n matrix) with entries 0 <= priorPhi[i,j] <= 1 describing the probability of an edge between gene i and gene j.
lambda regularization parameter to incorporate prior assumptions.
selEGenes optimize selection of E-genes for each model
verbose output while running or not
graphClass output inferred graph either as graphNEL or matrix
x nem object
... other arguments to pass
Phi adjacency matrix
a parameter of the inverse gamma prior for v=1/lambda
b parameter of the inverse gamma prior for v=1/lambda

Details

Scoring models by marginal log-likelihood is implemented in function score. Input consists of models and data, the type of the score ("mLL", "FULLmLL", "CONTmLL" or "CONTmLLDens"), the corresponding paramters (para) or hyperparameters (hyperpara), a prior for phenotype positions (Pe) and model structures Pm with regularization parameter lambda. If a structure prior Pm is provided, but no regularization parameter lambda, Bayesian model averaging with an inverse gamma prior on 1/lambda is performed.

score is usually called within function nem.

Value

graph the model with highest marginal likelihood (graphNEL object)
mLL vector of marginal likelihoods for all models
ppost vector of posterior probabilities for all models
pos a list of estimated positions of effect reporters for each model
mappos a list of maximum aposteriori estimates of effect positions for each model
type as used in function call
para as used in function call
hyperpara as used in function call
lambda as in function call

Author(s)

Florian Markowetz <URL: http://genomics.princeton.edu/~florian>

References

[1]
Markowetz F, Bloch J, Spang R, Non-transcriptional pathway features reconstructed from secondary effects of RNA interference, Bioinformatics, 2005
[2]
Markowetz F, Probabilistic Models for Gene Silencing Data, PhD thesis, Free University Berlin, 2006

See Also

nem, mLL, FULLmLL, enumerate.models

Examples

   # Drosophila RNAi and Microarray Data from Boutros et al, 2002
   data("BoutrosRNAi2002")
   D <- BoutrosRNAiDiscrete[,9:16]

   # enumerate all possible models for 4 genes
   models <- enumerate.models(unique(colnames(D)))

   # score models with marginal likelihood
   result <- score(models,D,type="mLL",para=c(.13,.05))
   
   # plot graph
   plot(result,what="graph")

   # plot scores
   plot(result,what="mLL") 
    
   # plot posterior of E-gene positions
   plot(result,what="pos")
   
   # MAP estimate of effect positions
   result$mappos[[which.max(result$mLL)]]


[Package nem version 2.2.1 Index]