The literature is replete with various approaches to extracting gene regulatory networks from microarray profiling data. Although many of these methods have produced networks which appear biologically plausible, based on circumstantial evidence from the literature, very little work has been done on validating the model networks experimentally. In this paper we present new results from a microarray time series study of adaptation to cold and successive re-adaptation to optimal temperatures in E. coli. Model networks were inferred from the data using the variational Bayesian state space modelling approach of Beal et al. (2005). Analysis of the biological implications of these network models is still on-going, but preliminary analysis has already revealed some promising novel biological hypotheses relating to the transcriptional response of bacterial cells adapting to the temperature shift. Our model places a number of genes at the higher level of the hierarchy (“hubs”) in the temperature shifted network, including hns and hybC. Encouragingly, the model network reveal some of the known regulatory interactions in the literature. The model also indicates that hns downregulates genes involved in aerobic metabolism and upregulates genes involved in anaerobic metabolism. This immediately suggests the hypothesis that hns plays a key role in regulating a switch between aerobic and anaerobic metabolism during the temperature adaptation. The experimental verification of this hypothesis is extremely simple. The hns- mutant exhibits the phenotype of growing at 10oC but stops growing if switched from 10oC to 37oC. Time series microarray data collected from this mutant strain should directly address the question of whether the expression of genes involved in aerobic/anaerobic metabolism during re-adaptation to 37oC is dependent on the expression of hns. Regulatory interactions which are either confirmed or not confirmed by this experiment will be used to define Bayesian priors for iterative retraining of the state space model by including the time series data collected from the perturbed system. We present results from a full cycle of this iterative procedure.