The local topology of dynamical network models for biology

Abstract

The search for motifs -recurrent patterns in the topology of a network- has allowed the identification of universal classes of complex systems from very diverse fields, and has been used as a quantitative tool to highlight unifying properties of evolved and designed networks. This work explores if, and to what extent, network superfamilies previously identified through census of triadic motifs, are represented in the largest data set of dynamic, biological network models. This work presents triad significance profiles of 71 existing biological, and experimentally inspired, network models. The data generated is treated agnostically, and consistently clustered to two classes using several unsupervised techniques. The more populated class correlates with the previously identified superfamily of sensory transmission networks, characterized by the feedforward loop motif typical of signal-processing systems. The other class, surprisingly, better correlates to the superfamily of word-adjacency network. The result is analyzed for varying network size thresholds, and connected to the effect of the model building activity. It is shown that the topology of biological subnetworks starts resembling the topology of "sentences" in word-adjacency networks when the model focuses on smaller portions of the network, coarse-graining the boundary dynamics.