| Type: | Package |
| Title: | Bayesian Profile Regression using Generalised Linear Mixed Models |
| Version: | 1.0.2 |
| Description: | Implements a Bayesian profile regression using a generalized linear mixed model as output model. The package allows for binary (probit mixed model) and continuous (linear mixed model) outcomes and both continuous and categorical clustering variables. The package utilizes 'RcppArmadillo' and 'RcppDist' for high-performance statistical computing in C++. For more details see Amestoy & al. (2025) <doi:10.48550/arXiv.2510.08304>. |
| License: | GPL-2 |
| Encoding: | UTF-8 |
| LazyData: | true |
| LazyDataCompression: | xz |
| RoxygenNote: | 7.3.2 |
| LinkingTo: | Rcpp, RcppArmadillo, RcppDist |
| Imports: | Rcpp, LaplacesDemon, MCMCpack, Matrix, Spectrum, mvtnorm |
| Depends: | R (≥ 3.5) |
| URL: | https://github.com/MatteoAmestoy/ProfileGLMM-package |
| BugReports: | https://github.com/MatteoAmestoy/ProfileGLMM-package/issues |
| NeedsCompilation: | yes |
| Packaged: | 2025-12-12 10:41:45 UTC; VNOB-0731 |
| Author: | Matteo Amestoy [aut, cre, cph], Mark van de Wiel [ths], Wessel van Wieringen [ths] |
| Maintainer: | Matteo Amestoy <m.amestoy@amsterdamumc.nl> |
| Repository: | CRAN |
| Date/Publication: | 2025-12-18 13:50:07 UTC |
One-Hot Encodes Factor Variables (FIRST Level as Reference)
Description
This function takes a dataframe, identifies all columns of class factor, and converts them into **dummy variables** using one-hot encoding via stats::model.matrix. For each factor, the function explicitly removes the first dummy variable generated, effectively making the **first level** of the factor the **reference level** (omitted category). Non-factor columns are retained as is.
Usage
encodeCat(dataframe)
Arguments
dataframe |
A |
Value
A data.frame where:
All original non-factor columns are present.
All original factor columns are replaced by a set of binary (0/1) dummy variables. The first level of the factor is excluded from the generated dummies, making the last level the reference.
Examples
data("exposure_data")
exp_data = exposure_data$df
covList = {}
covList$FE = c('X')
XFE = encodeCat(exp_data[,covList$FE, drop = FALSE])
List of the different outputs of the main function for examples
Description
A list of the different outputs of the main function for examples
Usage
examp
Format
A list with 4 components:
- dataProfile
Output of the profileGLMM_preprocess() function example
- MCMC_Obj
Output of the profileGLMM_Gibbs() function example
- post_Obj
Output of the profileGLMM_postprocess() function example
- pred_Obj
Output of the profileGLMM_predict() function example
Source
Generated synthetically by the package authors.
Simulated Data and Parameters for a exposure profile linear mixed model
Description
A list containing a simulated exposure dataset (df) and the ground-truth parameters
(theta0) used to generate it.
The dataset df contains N = 4500 observations across n_{Ind} = 1500
individuals, with $n_R = 3$ repeated measures per individual.
Usage
exposure_data
Format
A list with 2 components:
- df
A data frame with 4,500 rows and 6 variables (the simulated data).
- theta0
A list of 11 elements containing the true parameters used for simulation.
Details
The underlying model for the response \bold{Y} is:
\bold{Y} = \bold{X}_{Fe}\bold{\beta} + \bold{X}_{Int}\bold{\alpha}_{Lat} + \bold{X}_{Re}\bold{\alpha}_{RE} + \bold{\epsilon}
df Data Variables
- X
Continuous predictor (
\sim N(0, 1)).- t
Time-like variable (structured around 0, 1, 2).
- indiv
**Individual ID** (1 to 1500), the grouping factor.
- Exp1, Exp2
Exposure continuous predictors.
- Y
The **Simulated Response Variable** calculated as:
\bold{Y} = y_{Fe} + y_{Int} + y_{Re} + \epsilon, where\epsilon ~ N(0, 1).
theta0 Parameters
The list theta0 holds the true values used to generate Y, including:
-
Lat: **Categorical Factor** (9 levels), defining the clusters for interaction effects. -
beta: True fixed effects for the global intercept and\bold{X}(i.e., $(3, 2)$). -
alphaLat: Vector of 18 coefficients defining the cluster-specific intercepts and slopes for\bold{X}within the 9Latcategories. -
alphaRE: Vector of 1500 random slopes for the time variable\bold{t}, drawn from $N(0, 1)$. -
sigma: Residual standard deviation (1).
Source
Generated synthetically by the package authors.
Simulated Data and Parameters for a Piecewise Example
Description
A list containing a second simulated dataset (df) and its ground-truth
parameters (theta0). This dataset is generated from a **piecewise linear
model**, where the continuous predictor x is segmented into 6 bins, and
different intercept and slope coefficients are applied to each segment.
The dataset df contains $N = 3000$ observations.
Usage
piecewise_data
Format
A list with 2 components:
- df
A data frame with 3,000 rows and 2 variables (the simulated data).
- theta0
A list of 5 elements containing the true parameters used for simulation.
Details
The underlying model for the response \bold{Y} is:
\bold{Y} = \bold{X}_{Fe}\bold{\beta} + \bold{X}_{Lat}\bold{\alpha}_{Lat} + \bold{\epsilon}
where \bold{X}_{Fe} is the global intercept, and \bold{X}_{Lat}\bold{\alpha}_{Lat} models the piecewise relationship of x across the 6 categories defined in theta0$Lat. The error term \bold{\epsilon} ~ N(0, 1).
df Data Variables
- x
A continuous predictor, uniformly distributed between -3 and 3.
- Y
The **Simulated Response Variable** defined by the piecewise linear model.
theta0 Parameters
The list theta0 holds the true values used for simulation, including:
-
beta: True global intercept (i.e., (0.5)). -
Lat: The categorical factor (1 to 6) derived from segmentingx. -
alphaLat: Vector of $2 * 6 = 12$ coefficients defining the specific intercept and slope forxwithin each of the 6 segments.
Source
Generated synthetically by the package authors.
Initialize the prior hyperparameters for the Profile GLMM
Description
This function establishes the prior distributions for all parameters
in the Profile GLMM. It sets up vague, non-informative priors (often using small
precision/large variance or conjugate forms like Wishart/Dirichlet) for the fixed effects (beta_{FE}),
residual variance (\sigma^2), random effects covariance (\Sigma_{RE}), latent effects covariance (\Sigma_{Lat}),
cluster parameters (means and covariances), and the Dirichlet Process parameters (\alpha).
Usage
prior_init(params)
Arguments
params |
A list containing dimensional parameters of the model (often the output of
|
Value
A list (prior) containing the hyperparameter values structured by the parameter block they govern:
FE:Priors for fixed effects and residual variance (e.g.,
lambda,a,bfor conjugate Normal-Gamma).RE:Inverse-Wishart priors for random effects covariance (
\Sigma_{RE}) (e.g.,Phi,eta).assign:Priors for the cluster assignment parameters, nested under
Cont(Normal-Inverse-Wishart for continuous) andCat(Dirichlet for categorical).Lat:Inverse-Wishart prior for the latent effects covariance (
\Sigma_{Lat}) (e.g.,Phi,eta).DP:Parameters for the Dirichlet Process prior (e.g.,
scale,shape).
Examples
# Load dataProfile, the result of profileGLMM_preProcess()
data("examp")
dataProfile = examp$dataProfile
prior_config <- prior_init(dataProfile$params)
R Wrapper for Profile GLMM Gibbs Sampler (C++ backend)
Description
This is the main function for fitting the Profile Generalized Linear Mixed Model (Profile GLMM) using a blocked Gibbs sampling algorithm. It acts as an R wrapper, passing pre-processed data, initial values, and prior hyperparameters contained in the model object directly to the C++ implementation GSLoopCPP. The function simulates the posterior distribution of all model parameters, including fixed effects, random effects variance, profile cluster parameters, latent effects, and cluster assignments.
Usage
profileGLMM_Gibbs(model, nIt, nBurnIn)
Arguments
model |
A list object containing all data, initial parameter values, model dimensions, prior hyperparameters, and model configuration (e.g., regression type). This object is typically the output of a data processing function like
|
nIt |
Integer, the total number of MCMC iterations *counting* the burn-in period. The sampler will run for |
nBurnIn |
Integer, the number of initial MCMC iterations that are discarded (not saved) to allow the chain to converge. |
Value
A list containing the saved Gibbs-sampled MCMC chains for all model parameters (e.g., beta, Z, gamma, pvec, muClus, PhiClus, etc.) and the variable names from the original data. This output is ready for post-processing with profileGLMM_postProcess.
Examples
# Load dataProfile, the result of profileGLMM_pREProcess()
data("examp")
dataProfile = examp$dataProfile
MCMC_Obj = profileGLMM_Gibbs(model = dataProfile,
nIt = 100,
nBurnIn = 10
)
Post-process the MCMC chain from profileGLMM_Gibbs
Description
This function performs essential post-processing of the MCMC output generated by the profileGLMM_Gibbs function. It calculates the posterior means and credible intervals for the fixed effects (population parameters) and, optionally, computes a representative cluster partition using methods like Least Squares (LS) or Ng's spectral clustering (NG) on the co-occurrence matrix. It also provides estimated cluster characteristics (centroids, probability vectors, and outcome effects) for the representative partition.
Usage
profileGLMM_postProcess(
MCMC_Obj,
modeClus = "NG",
comp_cooc = TRUE,
alpha = 0.05
)
Arguments
MCMC_Obj |
Profile GLMM MCMC output of the |
modeClus |
A character string specifying the clustering method to determine the representative partition. Options are |
comp_cooc |
A logical value. If |
alpha |
A numeric value between 0 and 1, specifying the significance level for calculating the posterior credible intervals (CIs) of the fixed effects. Defaults to |
Value
A list with three elements:
coocMat:The co-occurrence matrix of the MCMC cluster assignments (
MCMC_Obj$Z).clust:A list containing the results of the representative clustering (if
comp_cooc = TRUE), including the optimal partition (Zstar), number of clusters (Kstar), representative cluster parameters (cen,pvec,gamma), and full posterior samples for the cluster characteristics.pop:A list containing the posterior mean and
(1-alpha)credible intervals for the fixed effects (betaFE).
Examples
# Load MCMC_Obj, the result of profileGLMM_Gibbs()
data("examp")
MCMC_Obj = examp$MCMC_Obj
post_Obj = profileGLMM_postProcess(MCMC_Obj, modeClus='LS')
print(post_Obj$pop$betaFE)
Prediction of cluster memberships and outcomes
Description
This function uses the results of the post-processed Profile GLMM MCMC chain to predict cluster memberships and outcomes for new or existing data. It first calculates the fixed effect (FE) contribution and then, if a representative clustering is available in post_Obj, computes the predicted cluster membership and the corresponding latent effect (Lat) contribution to the outcome.
Usage
profileGLMM_predict(post_Obj, XFE, XLat, UCont, UCat)
Arguments
post_Obj |
The post-processed output from the |
XFE |
A numeric matrix of fixed effects covariates for the prediction data. |
XLat |
A numeric matrix of latent effect covariates. This matrix is used for the interaction term with the predicted cluster membership. |
UCont |
A numeric matrix or vector of continuous profile variables (used for predicting cluster membership). Set to |
UCat |
A numeric matrix or vector of categorical profile variables (used for predicting cluster membership). Set to |
Value
A list with the following elements:
FE:A numeric vector of the predicted fixed effects contribution to the outcome.
Y:A numeric vector of the total predicted outcome (FE + Lat).
classPred:A factor vector of the predicted cluster membership for each observation.
NULLif no representative clustering was provided inpost_Obj.Int:A numeric vector of the predicted latent effect contribution to the outcome.
NULLif no representative clustering was provided.
Examples
# Load post_Obj, the result of profileGLMM_postProcess()
data("examp")
post_Obj = examp$post_Obj
# Load dataProfile, the result of profileGLMM_preProcess()
dataProfile = examp$dataProfile
pred_Obj = profileGLMM_predict(post_Obj,
dataProfile$d$XFE,
dataProfile$d$XLat,
dataProfile$d$UCont,
dataProfile$d$UCat)
Preprocess the data from a list describing the profile LMM model
Description
Preprocess the data from a list describing the profile LMM model
Usage
profileGLMM_preprocess(
regType,
covList,
dataframe,
nC,
intercept = list(FE = TRUE, RE = TRUE, Lat = TRUE)
)
Arguments
regType |
A string, current possibilities: linear or probit |
covList |
A list with fields:
|
dataframe |
A dataframe containing outcome anf covariates |
nC |
int: maximal number of cluster for the DP truncation |
intercept |
(optionnal): A list with fields
|
Value
A list with
d dictionary with [XFE,XRE,XLat,UCont,UCat,ZRE] design matrices
[[params]] list of the parameters of the data
n int nb of obs
qFE lint, number of covariates of FE
nRE int, number of stat units of RE
qRE int, number of covariates of RE
qLat int, number of covariates interacting with the latent clusters
qUCont int, number of continuous clustering covariates
qUCat int, number of categorical clustering covariates
nC int, maximal number of clusters
prior a list with all the specification of the default prior used
theta a list with a default set of parameters to start the chain, drawn from the prior
regType an int. Currently 0 for linear, 1 for probit
Examples
data("exposure_data")
exp_data = exposure_data$df
theta0 = exposure_data$theta0
covList = {}
covList$FE = c('X')
covList$RE = c('t')
covList$REunit = c('indiv')
covList$Lat = c('X')
covList$Assign$Cont = c('Exp1','Exp2')
covList$Assign$Cat = NULL
covList$Y = c('Y')
dataProfile = profileGLMM_preprocess(regType = 'linear',
covList = covList,
dataframe = exp_data,
nC = 30,
intercept = list(FE = TRUE, RE = FALSE, Lat = TRUE))
Initialize the variables for the Gibbs sampler chain
Description
This function generates initial values (theta) for all parameters in the Profile GLMM Gibbs sampler by drawing from the specified prior distributions. These initial values are crucial for starting the MCMC chain in profileGLMM_Gibbs. The initialization includes parameters for fixed effects, random effects variance, latent effects, and the profile cluster parameters (centroids, covariances, and categorical probability vectors).
Usage
theta_init(prior, params)
Arguments
prior |
A list containing the prior configuration to draw initialization from. This list should match the structure produced by the |
params |
A list containing the problem's dimensional parameters and indices (e.g., number of observations, number of covariates). This list should match the structure of the output from |
Value
A list (theta) containing the sampled initialization values for the Gibbs sampler. Key elements include:
sig2:Initial residual variance.
betaFE:Initial fixed effects coefficients.
SigRE:Initial random effects covariance matrix.
SigLat:Initial latent effects covariance matrix.
gammaLat:Initial latent effects coefficients, organized by cluster.
ClusCont:List containing initial continuous cluster parameters (
muandSigma).ClusCat:List containing initial categorical cluster parameters (
pvecClus).
Examples
# Load dataProfile, the result of profileGLMM_preProcess()
data("examp")
dataProfile = examp$dataProfile
theta = theta_init(dataProfile$prior,dataProfile$params)