Predictive Functional Connectivity of Real-World Systems
We are flooded with large-scale dynamic networked data. Analyses requiring exact comparisons between networks are computationally intractable, so new methodologies are sought. We extend the graphlet-based statistics to directed networks and demonstrate that they are superior to other measures. We predict a country's gross domestic product (GDP) solely from its wiring patterns in the world trade network (WTN) that could inform policy makers on benefits of trade agreements. Surprisingly, we find that it is not enough for a country to be in a densely connected core in the WTN to have a high GDP, as was previously believed. In addition to being in the core, a country must also trade with peripheral countries, while only being in the core and not trading with peripheral economies makes a country prone to debt. Furthermore, by tracking the dynamics of a country's positioning in the WTN over years, we predict success or failure of an emerging economy. We validate our methodology on metabolic networks, yielding insights into preservation of enzyme function from the network wiring patterns rather than sequence data. Overall, our methodology enables advanced analyses of directed networked data from any area of science, allowing domain-specific interpretation of a directed network's topology.