Markov chain Monte Carlo methods for Bayesian network inference, with applications in systems biology

Getting stuck in local maxima is a problem that arises while inferring Bayesian network (BN) structures. This thesis proposes Markov chain Monte Carlo (MCMC) samplers that have been known to substantially resolve this problem but have not been modified to fit simulating BN structures from discrete search spaces. The proposed MCMC samplers are new instances of the Neighbourhood sampler, Hit-and-Run sampler and Metropolis-Hastings sampler. Two adaptive techniques have been also developed to reduce the time-complexity required for inference. A new software has been designed to facilitate using the samplers. The proposed samplers have demonstrated efficient performance in practice.