Fig. 1

3CPET workflow. a 3CPET overflow: three types of data are provided, including DNA–DNA interactions (ChIA-PET), DNA–protein interactions (ChIP-seq), and PPIs. The algorithm builds a network connecting each DNA–DNA interaction, then passes them to the HDP model to infer the set of enriched networks. b Networks connecting DNA–DNA interactions are built by connecting each TF on one side to all of the TFs on the other side of the interaction. These networks are then converted into a bag of edges, where the frequency of each edge is equal to the number of shortest paths in which it appears. c The HDP model: each β _k represents a CMN and is a distribution over all possible edges. θ _n represents the CMN-per-network distribution. For each edge e _jn , we associate a latent variable z _jn to indicate the membership to a CMN (B _k ). z _jn are drawn from the CMN-per-network distribution