Parametric gene expression profile models and optimal Bayesian clustering

 

We present a broad class of models for nonparametric Bayesian clustering of parametric gene expression profiles. Under a surprisingly wide range of prior assumptions, 'incremental' Gibbs samplers for the unknown partition are available for posterior simulation. The main focus of the talk is on optimal clustering under pairwise coincidence loss functions, and we present a novel heuristic algorithm for the resulting optimisation problem.

Joint work with John Lau.