est.IDR.discrete: EM algorithm to estimate the copula mixture model for...

In qunhualilab/gIDR: Generalized irreproducibile discovery rate

Description

EM algorithm to estimate the copula mixture model for discrete data

Usage

est.IDR.discrete(x, y, mu, sigma, rho, p, eps, n.missing, miss.sym,
  as.single.loglik, as.single.em, common.only = TRUE, labels = NULL)

Arguments

x	replicate 1 : a vector of values from original observation
y	replicate 2 : paired with x, a repeated measure having same length as x.
mu	a starting value for the mean of the reproducible component.
sigma	a starting value for the standard deviation of the reproducible component.
rho	a starting value for the correlation coefficient of the reproducible component.
p	a starting value for the proportion of reproducible component.
eps	a small number to control convergence. For example, esp=0.0001.
n.missing	n.missing=0 indicates that we run the program as if there is no missing values; n.missing=positive number: the program estimates missing values.
miss.sym	symbol for missing value. For example, miss.sym=0.
as.single.loglik	an integer n to control the precision of numberical integration. If n=1, the computation is exact. If n>1, only integrate the bins with counts more than n, and treat bins with counts <= n as singletons in the likelihood computation
as.single.em	similar to as.single.loglik, except that it does the approximation in EM
common.only	If TRUE, the computation will only use the nonmissing entries. In this case, n.missing should be set to 0. If FALSE, it will use all the entries.
labels	the cdf on the left side of the lower boundary of the bins

Details

EM to compute the latent structure steps: 1. raw values are first transformed into pseudovalues 2. EM is used to compute the underlining structure, which is a mixture of two normals

Value

• para estimated parameters: p, rho, mu, sigma, estimated n.missing.

• loglik can be used for mapping category format back to original e.g. x.cat[level.factor[1]] == x[1], y.cat[level.factor[1]]==y[1].

• loglik.trace trajectory of log-likelihood.

• idr.cat a numeric vector of the local idr for each category (i.e. estimated conditional probablility for each observation to belong to the irreproducible component).

• IDR.cat a numerical vector of the expected irreproducible discovery rate for categories that are as irreproducible or more irreproducible than the given categories.

• idr.obs a numeric vector of the local idr for each observation.

• IDR.obs a numerical vector of the expected irreproducible discovery rate for observations that are as irreproducible or more irreproducible than the given observations.

Examples

#load chip_seq data
data(chip_seq)
x = chip_seq[,1]
y = chip_seq[,2]

# Initiation
mu <- 2.6
sigma <- 1.3
rho <- 0.8
p <- 0.7
eps <- 0.001
n.missing <- 0

# Estimate parameters of mixture model
gidr.out <- est.IDR.discrete(x, y, mu, sigma, rho, p, eps, n.missing,
                            miss.sym = 0, as.single.loglik = 1,
                            as.single.em = 1, common.only=TRUE, labels=NULL)

names(gidr.out)

qunhualilab/gIDR documentation built on May 14, 2019, 10:38 a.m.