BAREB: A Bayesian repulsive biclustering model for periodontal data.

Area of Science:

• Biostatistics
• Computational Biology
• Oral Health Research

Background:

• Periodontal diseases (PD) present heterogeneous patterns in patients, complicating oral care and treatment.
• Understanding this heterogeneity is crucial for effective prevention and maintaining tooth structure and function.

Purpose of the Study:

• To develop a novel Bayesian repulsive biclustering method (BAREB) for simultaneous clustering of PD patients and their tooth sites.
• To incorporate patient- and site-level covariates, spatial dependence, and nonrandom missingness into the clustering framework.

Main Methods:

• Developed BAREB, a Bayesian biclustering approach utilizing a determinantal point process prior for cluster diversity.
• Accounted for spatial dependence among tooth sites and nonignorable missing data mechanisms inherent in PD studies.
• Employed an efficient reversible jump Markov chain Monte Carlo sampler for posterior inference.

Main Results:

• Simulation studies demonstrated BAREB's accuracy in estimating biclusters and favorable performance against alternative methods.
• Application of BAREB to a clinical PD dataset yielded desirable and interpretable results.
• An Rcpp implementation of BAREB is provided, enhancing accessibility for researchers.

Conclusions:

• BAREB offers a robust and interpretable framework for analyzing complex periodontal disease patterns.
• The method effectively handles spatial dependencies and missing data, advancing the statistical analysis of oral health data.
• The availability of the BAREB R package facilitates its application in future periodontal disease research.