A population Monte Carlo method for Bayesian inference and its application to stochastic kinetic models

We introduce an extension of the population Monte Carlo (PMC) methodology to address the problem of Bayesian inference in high dimensional models. Specifically, we introduce a technique for the selection and update of importance functions based on the construction of Gaussian Bayesian networks. The structure of the latter graphical model enables a sequential sampling procedure that requires drawing only from unidimensional conditional distributions and leads to very efficient PMC algorithms. In order to illustrate the potential of the new technique we have considered the estimation of rate parameters in stochastic kinetic models (SKMs). SKMs are multivariate systems that model molecular interactions in biological and chemical problems. We present some numerical results based on a simple SKM known as predator-prey model.