Discovering Condition-Dependent Bayesian Networks for Gene Regulation

Among the main interests in many biological studies are the structure of gene regulatory network, and in particular differences in the regulatory interactions between different conditions. However, since the number of available samples is always very small and estimating the network structure is extremely hard, most current algorithms have to assume that the gene regulation does not change between conditions. We propose a new Bayesian network algorithm which (i) utilizes all the samples for estimating regulatory relations that remain the same across conditions, and (ii) explicitly searches for regulatory relationships that are active only in one of the conditions. The result is an easily interpretable map of changes in regulation in several conditions.