integratedEM: An iterative expectation-maximization algorithm for RIVER

In RIVER: R package for RIVER (RNA-Informed Variant Effect on Regulation)

Description

integratedEM iteratively executes e-step and m-step until it converges. This is a main function of RIVER.

Usage

integratedEM(Feat, Out, lambda, logistic.init, pseudoc, theta.init, costs,
  verbose = FALSE)

Arguments

Feat	Genomic features (G).
Out	Binary values of outlier status (E).
lambda	Selected lambda.
logistic.init	Smart initialization of beta (parameters between FR and G) from estimate of beta with E via multivariate logistic regression.
pseudoc	Pseudo count.
theta.init	Initial theta (parameters between FR (functionality of regulatory variant) and E).
costs	Candidate penalty parameter values for L2-regularization within logistic regression.
verbose	Logical option for showing extra information on progress.

Value

Best estimate of beta and theta, final multivariate logistic regression model, and posterior probabilities of FR.

Author(s)

Yungil Kim, [e-mail]

See Also

getFuncRvFeat, getFuncRvPosteriors, mleTheta, mleBeta, cv.glmnet, and https://github.com/ipw012/RIVER

Examples

dataInput <- getData(filename=system.file("extdata", "simulation_RIVER.gz",
        package = "RIVER"), ZscoreThrd=1.5)
Feat <- scale(t(Biobase::exprs(dataInput))) # genomic features (G)
Out <- as.vector(as.numeric(unlist(dataInput$Outlier))-1) # outlier status (E)
theta.init=matrix(c(.99, .01, .3, .7), nrow=2)
costs <- c(100, 10, 1, .1, .01, 1e-3, 1e-4)
logisticAllCV <- glmnet::cv.glmnet(Feat, Out, lambda=costs, family="binomial",
        alpha = 0, nfolds=10)
emModelAll <- integratedEM(Feat, Out, lambda=logisticAllCV$lambda.min,
        logistic.init=logisticAllCV$glmnet.fit, pseudoc=50, theta=theta.init,
        costs, verbose=FALSE)

RIVER documentation built on Nov. 8, 2020, 6:54 p.m.