Dynamical Bayesian networks have been extensively applied to the reconstruction of gene regulatory networks from gene expression time series. However, the standard approach is based on a homogeneous Markov chain, which fails to allow for changes in the regulatory processes with time. The objective of my presentation is to discuss a non-homogeneous generalization of the conventional approach. The method is based on a multiple change-point process and a mixture model, using latent variables to assign individual measurements to different components. The practical inference follows the Bayesian paradigm and samples the network structure, the number of components and the assignment of latent variables from the posterior distribution with RJMCMC. I will demonstrate the application of this scheme to gene expression time series from Arabidopsis thaliana and Drosophila melanogaster, with the objective to infer the gene regulatory network structures related to two important biological processes: circadian regulation in plants, and morphogenesis in insects. I will conclude the evaluation with an application to synthetic biology, where the objective is to learn a known in vivo regulatory network of five genes in Saccharomyces cerevisiae.