fitBM: Fit a three-beta mixture model to densities of functional...
In PANR: Posterior association networks and functional modules inferred from rich phenotypes of gene perturbations

Description Usage Arguments Details Value Author(s) References See Also Examples

The function fits a three-beta mixture model to densities of functional gene associations computed from rich phenotyping screens.

1
2
3

fitBM(object, para=list(zInit=NULL, thetaInit=c(alphaNeg=2, betaNeg=4,
alphaNULL=4, betaNULL=4, alphaPos=4, betaPos=2), gamma=NULL), 
ctrl=list(fitNULL=FALSE, tol=1e-3, maxIter=NULL), verbose=TRUE, ...)

`object`	an object of S4 class `BetaMixture`.
`para`	a list of initial values for parameter estimation in fitting a three-beta mixture model (see 'details').
`ctrl`	a list of control parameters for the mixture model fitting (see 'details').
`verbose`	a logical value to switch on (if `TRUE`) or off if `FALSE` detailed run-time message.
`...`	other arguments of the function `nlm`.

This function fits a beta-mixture model to functional gene associations using the Expectation-Maximization algorithm. The function allows various parameter settings to perform fitting by the original (if model='global') or stratified (if model='stratified') beta-mixture model (the model should be specified when creating a new object of BetaMixture).

The initial values of the beta distributions can be set by thetaInit, is a numeric vector including the two shape parameters for the '-' (negative), 'x' (NULL) and '+' (positive) distributions. Please note that if ctrl$NULL is TRUE, meaning that the NULL distribution has already been fitted, then para$alphaNULL and para$betaNULL are supposed to be filled in the estimated NULL parameters by the function fitNULL).

zInit is a matrix of posterior probabilities for gene associations following the three mixture components.

The hyper-parameters for the dirichlet priors for the mixture components can also be set by para$gamma, which is a numeric matrix with rows and columns correponding to association partitions and the three beta mixture components.

The other arguments to control the fitting algorithm are tol and maxIter, which are covergence tolerence and the maximal iterations.

Since the estimation of shape parameters of beta distributions are realized by the function nlm numerically, additional arguments for nlm are allowed by ....

This function will return an updated object of class BetaMixture.

Xin Wang xw264@cam.ac.uk

Xin Wang, Mauro Castro, Klaas W. Mulder and Florian Markowetz, Posterior association networks and enriched functional gene modules inferred from rich phenotypic perturbation screens, in preparation.

fitNULL

## Not run: 
data(Bakal2007)
bm1<-new("BetaMixture", pheno=Bakal2007, model="global", order=1)
bm1<-fitNULL(bm1, nPerm=10, thetaNULL=c(alphaNULL=4, betaNULL=4),
	sumMethod="median", permMethod="all", verbose=TRUE)
bm1<-fitBM(bm1, para=list(zInit=NULL, thetaInit=c(alphaNeg=2, betaNeg=4, 
	alphaNULL=bm1@result$fitNULL$thetaNULL[["alphaNULL"]], 
	betaNULL=bm1@result$fitNULL$thetaNULL[["betaNULL"]], 
	alphaPos=4, betaPos=2), gamma=NULL), 
	ctrl=list(fitNULL=FALSE, tol=1e-1), verbose=TRUE, gradtol=1e-3)

## End(Not run)