Help for package GPTCM

Title:

Generalized Promotion Time Cure Model with Bayesian Shrinkage Priors

Version:

1.1.3

Description:

Generalized promotion time cure model (GPTCM) via Bayesian hierarchical modeling for multiscale data integration (Zhao et al. (2025) <doi:10.48550/arXiv.2509.01001>). The Bayesian GPTCMs are applicable for both low- and high-dimensional data.

Maintainer:

Zhi Zhao <zhi.zhao@medisin.uio.no>

The code in src/arms.cpp is slightly modified based on the research paper implementation written by Wally Gilks.

URL:

https://github.com/ocbe-uio/GPTCM

BugReports:

https://github.com/ocbe-uio/GPTCM/issues

License:

GPL-3

VignetteBuilder:

knitr

Depends:

R (≥ 4.1.0)

Encoding:

UTF-8

RoxygenNote:

7.3.2

LinkingTo:

Rcpp, RcppArmadillo

Imports:

Rcpp, survival, riskRegression, ggplot2, ggridges, miCoPTCM, loo, mvnfast, Matrix, scales, utils, stats, graphics

Suggests:

knitr, survminer

NeedsCompilation:

yes

SystemRequirements:

C++17

Packaged:

2025-10-31 21:56:28 UTC; zhiz

Author:

Zhi Zhao [aut, cre]

Repository:

CRAN

Date/Publication:

2025-11-01 00:30:13 UTC

GPTCM: Generalized Promotion Time Cure Model with Bayesian Shrinkage Priors

Description

Generalized promotion time cure model (GPTCM) via Bayesian hierarchical modeling for multiscale data integration (Zhao et al. (2025) doi:10.48550/arXiv.2509.01001). The Bayesian GPTCMs are applicable for both low- and high-dimensional data.

Author(s)

Maintainer: Zhi Zhao zhi.zhao@medisin.uio.no

Fit Bayesian GPTCM Models

Description

This is the main function to fit the Bayesian GPTCMs (Zhao et al. 2025) with multiscale data for sparse identification of high-dimensional covariates. The core code for MCMC algorithm uses Rcpp (Eddelbuettel and François 2011) and RcppArmadillo (Eddelbuettel and Sanderson 2014)

Usage

GPTCM(
  dat,
  nIter = 500,
  burnin = 200,
  thin = 1,
  tick = 100,
  proportion.model = TRUE,
  dirichlet = TRUE,
  hyperpar = NULL,
  BVS = TRUE,
  threads = 1,
  kappaIGamma = FALSE,
  kappaSampler = "arms",
  gammaPrior = "bernoulli",
  gammaSampler = "MC3",
  etaPrior = "bernoulli",
  etaSampler = "MC3",
  w0IGamma = TRUE,
  initial = NULL,
  arms.list = NULL
)

Arguments

dat

input data as a list containing survival data sub-list survObj with two vectors (event and time), clinical variable matrix x0, cluster-specific covariates X, and proportions data matrix proportion

nIter

the number of iterations of the chain

burnin

number of iterations to discard at the start of the chain

thin

thinning MCMC intermediate results to be stored

tick

an integer used for printing the iteration index and some updated parameters every tick-th iteration. Default is 1

proportion.model

logical value; should the proportions be modeled or not. If (proportion.model = FALSE), the argument dirichlet will be invalid

dirichlet

logical value; should the proportions be modeled via the common (dirichlet = TRUE) or alternative (dirichlet = FALSE) parametrization of the Dirichlet regression model

hyperpar

a list of relevant hyperparameters

BVS

logical value for implementing Bayesian variable selection

threads

maximum threads used for parallelization. Default is 1

kappaIGamma

logical value for using inverse-gamma prior (TRUE) or gamma prior (FALSE) for Weibull's shape parameter shape parameter

kappaSampler

one of "arms", "slice" (slice not yet implemented)

gammaPrior

one of c("bernoulli", "MRF")

gammaSampler

one of c("mc3", "bandit")

etaPrior

one of c("bernoulli", "MRF")

etaSampler

one of c("mc3", "bandit")

w0IGamma

logical value; if FALSE, a common parameter is used for the intercept's prior variance and the coefficient's prior variance

initial

a list of initial values for parameters "kappa", "xi", "betas", and "zetas"

arms.list

a list of parameters for the ARMS method

Value

An object of a list including the following components:

input - a list of all input parameters by the user
output - a list of the all mcmc output estimates:
- "xi" - a matrix with MCMC intermediate estimates of effects on clinical variables
- "kappa" - a vector with MCMC intermediate estimates of the Weibull's shape parameter
- "betas" - a matrix with MCMC intermediate estimates of effects on cluster-specific survival
- "zetas" - a matrix with MCMC intermediate estimates of effects on cluster-specific proportions
- "gammas" - a matrix with MCMC intermediate estimates of inclusion indicators of variables for cluster-specific survival
- "gamma_acc_rate" - acceptance rate of the M-H sampling for gammas
- "etas" - a matrix with MCMC intermediate estimates of inclusion indicators of variables for cluster-specific proportions
- "eta_acc_rate" - acceptance rate of the M-H sampling for etas
- "loglikelihood" - a matrix with MCMC intermediate estimates of individuals' likelihoods
- "tauSq" - a vector with MCMC intermediate estimates of tauSq
- "wSq" - a matrix with MCMC intermediate estimates of wSq
- "vSq" - a matrix with MCMC intermediate estimates of vSq
- "post" - a list with posterior means of "xi", "kappa", "betas", "zetas", "gammas", "etas"
call - the matched call

References

Eddelbuettel D, Sanderson C (2014). RcppArmadillo: Accelerating R with high-performance C++ linear algebra. Computational Statistics and Data Analysis, 71, 1054–1063

Zhao Z, Kızılaslan F, Wang S, Zucknick M (2025). Generalized promotion time cure model: A new modeling framework to identify cell-type-specific genes and improve survival prognosis. arXiv:2509.01001

Examples


# simulate data
set.seed(123)
n <- 200 # subjects
p <- 10 # variable selection predictors
L <- 3 # cell types
dat <- simData(n, p, L)

# run a Bayesian GPTCM model: GPTCM-Ber2
fit <- GPTCM(dat, nIter = 10, burnin = 0)

Extract the posterior estimate of parameters

Description

Extract the posterior estimate of the parameters of a GPTCM class object.

Usage

getEstimator(object, estimator = "gamma", Pmax = 0, type = "marginal")

Arguments

object

an object of class GPTCM

estimator

the name of one estimator. Default is the latent indicator estimator "gamma". Other options are among "c('beta', 'zeta', 'eta', 'xi', 'elpd', 'logP')"

Pmax

threshold that truncate the estimator "gamma" or "eta". Default is 0. If Pmax=0.5 and type="conditional", it gives median probability model betas

type

the type of output beta. Default is marginal, giving marginal beta estimation. If type="conditional", it gives beta estimation conditional on gamma=1

Value

Return the estimator from an object of class GPTCM. It is a matrix or vector

References

Examples


# simulate data
set.seed(123)
n <- 200 # subjects
p <- 10 # variable selection predictors
L <- 3 # cell types
dat <- simData(n, p, L)

# run a Bayesian GPTCM model: GPTCM-Ber2
fit <- GPTCM(dat, nIter = 10, burnin = 0)

gamma.hat <- getEstimator(fit, estimator = "gamma")

Metropolis sampler for a target density

Description

Random number generator via Metropolis-Hastings algorithm.

Usage

metropolis_sampler(
  initial_value,
  n = n,
  proposal_shape = 1,
  proposal_scale = 1,
  theta = 1,
  proportion = 0.5,
  mu = 1,
  kappas = 0.9,
  burnin = 0,
  lag = 1
)

Arguments

initial_value

initial values

n

number of draws

proposal_shape

Weibull's shape parameter in the proposal

proposal_scale

Weibull's scale parameter in the proposal

theta

cure rate parameter (log scale)

proportion

proportions data

mu

mean survival time

kappas

Weibull's true shape parameter

burnin

length of burn-in period

lag

discarding lag-1 values in the Metropolis step

Value

A dataframe consisting of the sampled values and acceptance rate

Examples


times <- metropolis_sampler(10, 5)

Plot curves of time-dependent Brier score

Description

Predict time-dependent Brier scores based on different survival models

Usage

plotBrier(
  dat,
  datMCMC,
  dat.new = NULL,
  time.star = NULL,
  xlab = "Time",
  ylab = "Brier score",
  PTCM = TRUE,
  ...
)

Arguments

dat

datMCMC

returned object from the main function GPTCM()

dat.new

input data for out-sample prediction, with the same format as dat

time.star

largest time for survival prediction

xlab

a title for the x axis

ylab

a title for the y axis

PTCM

logical value for adding survival prediction by the PTCM

...

other parameters

Value

A ggplot2::ggplot object. See ?ggplot2::ggplot for more details of the object.

References

Examples


# simulate data
set.seed(123)
n <- 200 # subjects
p <- 10 # variable selection predictors
L <- 3 # cell types
dat <- simData(n, p, L)

# run a Bayesian GPTCM model: GPTCM-Ber2
fit <- GPTCM(dat, nIter = 5, burnin = 0)


plotBrier(dat, datMCMC = fit, PTCM = FALSE)

Plot posterior estimates of regression coefficients

Description

create nice plots for estimated coefficients and 95

Usage

plotCoeff(
  dat,
  datMCMC,
  estimator = "beta",
  intercept = FALSE,
  bandwidth = NULL,
  xlim = NULL,
  xlab = NULL,
  label.y = NULL,
  first.coef = NULL,
  y.axis.size = 8,
  ...
)

Arguments

dat

datMCMC

returned object from the main function GPTCM()

estimator

print estimators, one of c("beta", "zeta", "gamma", "eta")

intercept

logical value to print intercepts

bandwidth

a value of bandwidth used for the ridgeplot

xlim

numeric vectors of length 2, giving the x-coordinate range.

xlab

a title for the x axis

label.y

a title for the y axis

first.coef

number of the first variables. Default NULL for all variables

y.axis.size

text size in pts

...

others

Value

A ggplot2::ggplot object. See ?ggplot2::ggplot for more details of the object.

References

Examples


# simulate data
set.seed(123)
n <- 200 # subjects
p <- 10 # variable selection predictors
L <- 3 # cell types
dat <- simData(n, p, L)

# run a Bayesian GPTCM model: GPTCM-Ber2
fit <- GPTCM(dat, nIter = 10, burnin = 0)

plotCoeff(dat, datMCMC = fit, estimator = "beta")

MCMC trace-plots

Description

Trace-plots of regression coefficients over MCMC iterations

Usage

plotMCMC(dat, datMCMC, estimator = "xi")

Arguments

dat

datMCMC

returned object from the main function GPTCM()

estimator

print estimators, one of c("beta", "zeta", "gamma", "eta")

Value

A ggplot2::ggplot object. See ?ggplot2::ggplot for more details of the object.

References

Examples


# simulate data
set.seed(123)
n <- 200 # subjects
p <- 10 # variable selection predictors
L <- 3 # cell types
dat <- simData(n, p, L)

# run a Bayesian GPTCM model: GPTCM-Ber2
fit <- GPTCM(dat, nIter = 10, burnin = 0)

plotMCMC(dat, datMCMC = fit, estimator = "xi")

Prediction of survival probability

Description

Compute predicted survival probability for a GPTCM

Usage

## S3 method for class 'GPTCM'
predict(object, dat, newdata = NULL, type = "survival", times = NULL, ...)

Arguments

object

the results of a GPTCM fit

dat

the dataset used in GPTCM()

newdata

optional new data at which to do predictions. If missing, the prediction will be based on the training data

type

the type of predicted value. Currently it is only valid with 'survival'

times

evaluation time points for survival prediction. Default NULL for predicting all time points in the newdata set

...

for future methods

Value

A matrix object

References

Examples


# simulate data
set.seed(123)
n <- 200 # subjects
p <- 10 # variable selection predictors
L <- 3 # cell types
dat <- simData(n, p, L)

# run a Bayesian GPTCM model: GPTCM-Ber2
fit <- GPTCM(dat, nIter = 10, burnin = 0)

pred.survival <- predict(fit, dat, newdata = dat, times = c(1, 3, 5))

Main function implemented in C++ for the MCMC loop

Description

Main function implemented in C++ for the MCMC loop

Usage

run_mcmc(
  nIter,
  burnin,
  thin,
  n,
  nsamp,
  ninit,
  convex,
  npoint,
  dirichlet,
  proportion_model,
  BVS,
  threads,
  gamma_prior,
  gamma_sampler,
  eta_prior,
  eta_sampler,
  initList,
  rangeList,
  hyperparList,
  datEvent,
  datTime,
  datX,
  datX0,
  datProportionConst
)

Arguments

nIter

number of MCMC iterations

burnin

length of MCMC burn-in period

thin

number of thinning

n

number of samples to draw

nsamp

how many samples to draw for generating each sample; only the last draw will be kept

ninit

number of initials as meshgrid values for envelop search

convex

adjustment for convexity (non-negative value, default 1.0)

npoint

maximum number of envelope points

dirichlet

not yet implemented

proportion_model

logical value for modeling the proportions data

BVS

logical value for implementing Bayesian variable selection

threads

maximum threads used for parallelization. Default is 1

gamma_prior

one of c("bernoulli", "MRF")

gamma_sampler

one of c("mc3", "bandit")

eta_prior

one of c("bernoulli", "MRF")

eta_sampler

one of c("mc3", "bandit")

initList

a list of initial values for parameters "kappa", "xi", "betas", and "zetas"

rangeList

a list of ranges of initial values for parameters "kappa", "xi", "betas", and "zetas"

hyperparList

a list of relevant hyperparameters

datEvent

a vector of survival status

datTime

a vector of survival times

datX

an array of cluster-specific covariates

datX0

a matrix of mandatory variables

datProportionConst

an array of cluster-specific proportions

Simulate data

Description

Simulate survival data based on a GPTCM or Cox model

Usage

simData(
  n = 200,
  p = 10,
  L = 3,
  Sigma = NULL,
  kappas = 2,
  proportion.model = "dirichlet",
  model = "GPTCM"
)

Arguments

n

number of subjects

p

number of covariates in each cluster

L

number of clusters

Sigma

NULL (for a default covariance matrix) or "independent" (i.e. Sigma=diag(p*L)) or a self-defined matrix

kappas

value of the Weibull's shape parameter

proportion.model

One of c("alr", "cloglog", "log", "dirichlet")

model

one of c("GPTCM", "Cox")

Value

An object of a list with 12 components

"survObj" - a list including events and times
"accepted" - a vector with acceptance rates to generate each time-to-event data point by Metropolis-Hastings algorithm.
"proportion.model" - value to indicate the model type for simulating proportions data.
"proportion" - a matrix with simulated proportions data.
"kappas" - value of the Weibull's shape parameter.
"x0" - a matrix with the data of clinical variables
"X" - an array with cluster-specific covariates
"xi" - effects of clinical variables
"beta0" - intercepts related to cluster-specific-survival.
"betas" - effects related to cluster-specific-survival.
"zetas" - effects related to cluster-specific-proportions.
"mrfG" - a graph corresponding to the precision matrix of cluster-specific covariates
"mrfG2" - a graph corresponding to every second edge in "mrfG"

References

Examples


# simulate data
set.seed(123)
n <- 200 # subjects
p <- 10 # variable selection predictors
L <- 3 # cell types
dat <- simData(n, p, L)
str(dat)

Target density

Description

Predefined target density corresponding to the population survival function of GPTCM

Usage

target(x, theta, proportion, mu, kappas)

Arguments

x

value generated from the proposal distribution

theta

cure rate parameter (log scale)

proportion

proportions data

mu

mean survival time

kappas

Weibull's true shape parameter

Value

value of the targeted (improper) probability density function

Examples


time1 <- target(1.2, 0.1, c(0.2, 0.3, 0.5), c(0.2, 0.1, 0.4), 2)

GPTCM: Generalized Promotion Time Cure Model with Bayesian Shrinkage Priors

Description

Author(s)

See Also

Fit Bayesian GPTCM Models

Description

Usage

Arguments

Value

References

Examples

Extract the posterior estimate of parameters

Description

Usage

Arguments

Value

References

Examples

Metropolis sampler for a target density

Description

Usage

Arguments

Value

Examples

Plot curves of time-dependent Brier score

Description

Usage

Arguments

Value

References

Examples

Plot posterior estimates of regression coefficients

Description

Usage

Arguments

Value

References

Examples

MCMC trace-plots

Description

Usage

Arguments

Value

References

Examples

Prediction of survival probability

Description

Usage

Arguments

Value

References

Examples

Main function implemented in C++ for the MCMC loop

Description

Usage

Arguments

Simulate data

Description

Usage

Arguments

Value

References

Examples

Target density

Description

Usage

Arguments

Value

Examples