BAREB

BAREB performs Bayesian repulsive biclustering to simultaneously cluster periodontal disease (PD) patients and tooth sites while incorporating patient- and site-level covariates to characterize disease heterogeneity.

Key Features:

• Bayesian Repulsive Biclustering: Employs a Bayesian framework to bicluster patients and tooth sites jointly, producing distinct and non-overlapping biclusters.
• Determinantal Point Process (DPP) Prior: Utilizes a DPP prior to induce diversity among biclusters and promote interpretable, well-separated groupings.
• Spatial Dependence Modeling: Incorporates spatial dependence among tooth sites to account for spatially referenced PD progression.
• Handling Nonrandom Missingness: Integrates mechanisms to account for nonrandom (nonignorable) missingness due to tooth loss in PD datasets.
• Reversible Jump MCMC for Posterior Inference: Performs posterior inference using a reversible jump Markov chain Monte Carlo (MCMC) sampler for model estimation and comparison.

Scientific Applications:

• PD Pattern Identification: Identifies distinct periodontal disease patterns across patients and tooth sites for stratified analysis.
• Spatial PD Analysis: Enables analysis of spatially structured PD progression across oral sites.
• Robust Analysis with Missing Data: Supports robust inference in clinical PD datasets with nonrandom missingness due to tooth loss.

Methodology:

Bayesian framework with repulsive biclustering using a Determinantal Point Process (DPP) prior, modeling of spatial dependence among tooth sites, explicit handling of nonrandom (nonignorable) missingness, posterior inference via reversible jump MCMC, and implementation in R with Rcpp.