Think of the murmuration of starlings, the schooling of fish, or even hypes on social media or the spreading of power outs: the behaviour of a network is critically determined by its structure. This structure induces collective emergent behaviour that can only be understood by analysing the whole network in relation to its constituent parts. The relation between network structure and information processing capacity is essential at every scale: from molecules and genes to large neural networks to populations of behaving agents, on every level nodes form complex networks underlying the most essential functions of the brain, body and society. Only recently, with high-throughput techniques, have we begun to collect the vast amounts of data needed to study the structure and functioning of these networks. However, analysing these data is still a challenge and the nature of complex network processes are still poorly understood.
In order to compare networks, simulated or physical ones, or healthy versus diseased, network analysis and visualization tools are needed across three analysis dimensions: structure, activity and information processing. Many quantitative tools exist for analysing networks, but they are mostly restricted to one application domain or network type and often only address one of these dimensions. The novelty of our contributions lies in the combination of existing and new techniques, from different research domains, and across the three analysis dimensions. By transcending specific data sets or domains, we open up new insights into the properties, behaviour and dynamic evolution of biological networks.
In order to understand the relation between network connectivity, plasticity (how networks change) and activity, the researchers have obtained, through experiments and simulations, high quality and rich datasets, across levels and species: from molecular networks to whole brain neural activity of the fish Danionella. In order to analyse such rich datasets, the researchers developed new tools, based on amongst others information theory and topology, and expanded the use of existing tools into new domains, for instance from genetic to neural networks. Next, to assess the computations performed by biological networks, the researchers used computational models and experiments using virtual reality, acquiring datasets that show the relation between networks properties and their function. The analysis of these datasets required new tools, for instance to analyse body posture and movement form videos, but also task performance. All in all, these new datasets and analyses have resulted in new insights into the roles node non-linearity and heterogeneity, delays, the type of input a network receives and the dimensionality of the network’s activity play with respect to network functioning in different tasks and contexts. The obtained data and developed analyses methods form valuable resources for the community as a whole.
We trained a cohort of data scientists of the future, that will be able to analyse (biological) networks across levels, domains and types.
