Bayesian Clustering of Multivariate Immunological Data.

Given a dataset of B cell subpopulation quantities, for about six thousand patients, that is a cross-sectional immunological dataset, here we detect clusters representing models of immune system states in an unsupervised way (i.e., according only to their different statistical properties). Two time-evolving B cell networks are also generated from data-driven hidden Markov models, with four and five hidden states, respectively. Our interpretation from a biomedical viewpoint of the statistical parameters of the Bayesian models confirms an age related decline of some types of B cell functions and finds out a class of old patients with unexpected B cell values.