Poisson/gamma random field models: a door opener to nonparametric Bayesian modelling

Katja Ickstadt. Fakultat Statistik, Technical University of Dortmund

The talk will start with briefly reviewing joint work with Robert Wolpert on Poisson/gamma random field models, including their history, the sampling scheme and applications. These models are nonparametric Bayesian models, and their impact on further non- and semiparametric Bayesian modelling approaches developed in my research group at the TU Dortmund University will comprise the remainder of this talk. Two specific topics will be discussed in more detail. First, nonparametric Bayesian network models in systems biology are studied with respect to their ability to uncover distinct protein-network topologies in heterogeneous cell populations. Second, shape constraint regression models based on Bernstein-Schoenberg splines are introduced that allow for penalizing against general parametric functions instead of using difference penalties.