A generalized Markov sampler (Q1431344)

The authors present a model-switching MCMC (Markov chain Monte Carlo) sampler that generalizes the Gibbs sampler and enables Gibbs-like sampling in more general spaces. In particular, it allows one sampling from probability spaces with uncertainty about not only the model parameters but also about the model itself. Therefore, the new sampler enables one to use the Gibbs sampling approach for model determination, evaluation, and averaging. Moreover, the authors show that the new sampler encompasses, as particular cases, the well-known Metropolis, Barker, and Hastings MCMC samplers and the reversible jump MCMC. The sampler is illustrated by three examples of its applications in computational biology.