Property-Driven Statistics of Biological Networks

An analysis of heterogeneous biological networks based on randomizations that preserve the structure of component subgraphs is introduced and applied to the yeast protein-protein interaction and transcriptional regulation network. Shuffling this network, under the constraint that the transcriptional and protein-protein interaction subnetworks are preserved reveals statistically significant properties with potential biological relevance. Within the population of networks which embed the same two original component networks, the real one exhibits simultaneously higher bi-connectivity (the number of pairs of nodes which are connected using both subnetworks), and higher distances. Moreover, using restricted forms of shuffling that preserve the interface between component networks, we show that these two properties are independent: restricted shuffles tend to be more compact, yet do not lose any bi-connectivity.

Finally, we propose an interpretation of the above properties in terms of the signalling capabilities of the underlying network.