pmixcox: IRLS-EM algorithm

In scrcss/pfmc_pkg: IRLS-EM algorithm for the penalized finite-mixture Cox PH model

Description

IRLS-EM algorithm for the optimization of penalized maximum likelihood of finite-mixture Cox PH model using the penalty of LASSO, adaptive LASSO, SCAD or elastic net.

Usage

pmixcox(Time, Delta, X, K = 3, iter.max = 1000, u.init = NULL, 
        tparm = 0, alpha = 1, scad = FALSE, adpcoef = NULL, 
		abstol = 0.01, reltol = 1e-04, seed = 1)

Arguments

Time	observed time
Delta	survival status
X	a data matrix of explanatory variables, where each colomn correponds to one predictor and each row indicates one sample.
K	number of components in the finite-mixture Cox model
iter.max	maximum number of EM iterations
u.init	initial value of U, a data matrix that gives the probability that each sample belongs to each component. Each row corresponds to one sample and each column indicates one component.
tparm	value of tuning parameter for variable selection
alpha	the elastic net mixing parameter
scad	is the SCAD penalty applied?
adpcoef	the adaptive weights in the adaptive LASSO method
abstol	absolute tolerance of EM algorithm
reltol	relative tolerance of EM algorithm
seed	random seed for initialing U if it is not given

Value

a list, where

U	posterior probability matrix
fit	a list with K components, each of which is an object with class glmnet or ncvsurv, output by the last EM iteration
pi	estimate for the mixing probability
class	subtype classification for each sample
ploglik	partial log-likelihood
mixloglik	observed log-likelihood
iter	number of EM iterations
convergence	indicator to show the algorithm converged or not

Author(s)

Shijie Quan, Shun He

References

Subtype classification and heterogeneous prognosis model construction in precision medicine. Na You, Shun He, Xueqin Wang, Junxian Zhu and Heping Zhang

Examples

require(PFMC)
data(example.data)
nopenalty.fit = pmixcox(Time = example.data$time,
  Delta = example.data$status,
  X = example.data[,-1:-2],
  K = 2,
  tparm = 0,
  seed = 1)

adpcoef = lapply(nopenalty.fit$fit, function(x) 1/abs(x))
alasso.cv = calcvcox(Time = example.data$time,
  Delta = example.data$status,
  X = example.data[,-1:-2],
  K = 2,
  nopenaltyresult = nopenalty.fit,
  adpcoef = adpcoef)

tparmchoice = alasso.cv[1,]
alasso.fit = pmixcox(Time = example.data$time,
  Delta = example.data$status,
  X = example.data[,-1:-2],
  K = 2,
  u.init = nopenalty.fit$U,
  tparm = tparmchoice,
  adpcoef = adpcoef)

scrcss/pfmc_pkg documentation built on May 8, 2019, 2:33 a.m.