score {nem} | R Documentation |
Function to compute the marginal likelihood of a set of phenotypic hierarchies.
score(models, D, type="mLL", para=NULL, hyperpara=NULL, Pe=NULL, Pm=NULL, lambda=0, delta=1, verbose=TRUE, graphClass="graphNEL") ## S3 method for class 'score': print(x, ...) PhiDistr(Phi, Pm, a=1, b=ncol(Phi)^2)
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 log-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. |
delta |
regularization parameter for automated E-gene subset selection (CONTmLLRatio only) |
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 |
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"
or "CONTmLLRatio"
), 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.
With type "CONTmLLRatio" usually an automated selection of most relevant E-genes is performed by introducing a "null" S-gene. The corresponding prior probability of leaving out an E-gene is set to delta/no. S-genes.
score
is usually called within function nem
.
graph |
the inferred directed graph (graphNEL object) |
mLL |
marginal likelihood of final model |
pos |
posterior over effect positions |
mappos |
MAP estimate of effect positions |
type |
as used in function call |
para |
as used in function call |
hyperpara |
as used in function call |
lambda |
as in function call |
selected |
selected E-gene subset |
Holger Froehlich, Florian Markowetz <URL: http://genomics.princeton.edu/~florian>
nem
, mLL
, FULLmLL
, enumerate.models
# 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)]]