Bayesian model-based tight clustering for time course data (Q626248)

The authors propose a new clustering algorithm aimed to detect small clusters of closely-related subjects (tight clusters). Longitudinal observations of each subject are modeled by smoothing splines with random coefficients. The distribution of the coefficients is different in different clusters. A Bayesian approach is proposed to the clustering. The Crowley's prior is used for the clustering partition of the sample. A Markov chain Monte Carlo (MCMC) optimization algorithm is constructed to derive the partition which maximizes the posterior density. Only one subject changes its cluster at each step of the algorithm. The authors propose to construct tight clusters from the subjects, which tend to stay together within a cluster over the course of MCMC iterations. This technique is applied to the analysis of DNA microarrays data of corneal wound healing.