The eukaryotic cell is a dynamic machinery responding to both internal and external stimuli. It is essential to understand its operation in a dynamic context in order to capture the existing inherent complexity. One way to tackle this problem is through the elucidation of the dynamic behaviour of cellular networks.
Current technologies allow us to identify the differences in the collective internal state of cellular networks by acquiring snapshots of the global system through the course of time. However, defining the relationship between the entities, i.e. the edges, in a dynamic biological network, which can also potentially involve a time component, is challenging. We can monitor some time-dependent changes in the network topology, for example, in the event that one or more nodes lose their functional or physical links with other nodes, by acquiring mutations over time. These irreversible modifications can then be simplified into a static representation of the final state of the network. However, it is challenging to account for the reversible changes, which represent a substantial fraction of all interactions in the cell. These are frequently attributed to temporary interactions and/or couplings; those that tend to be established conditionally, satisfying certain temporal, spatial, or state-specific constraints, as opposed to permanent interactions, that are constitutively present across the complete spatio-temporal range. Although interactions can be classified as permanent and/or transient, we lack information on the co-occurrence of specific temporary interactions on a global scale across the eukaryotic network. Therefore, most of the cellular interaction networks represent all possible interactions that might occur rather than offering a realistic snapshot view of the topology of the cell at any given instance.
Dynamic networks, mostly due to their temporal nature, are generally larger, more multi-modal, more multi-plex and have higher dimensionality than their static counterparts, with a huge potential to develop into meta-networks. Such “networks of networks” can allow the integration of multiple types of data and can potentiate the study of the whole cell in light of a novel integrative perspective. Statistical methods that are able to handle heterogeneous and complex structures and uncertainties are routinely employed in the analysis of such networks. These methods also allow the nodes in a dynamic network to be treated as probabilistic features, which can gain the ability to learn in order to adapt to a condition over time, facilitating difficult problems in cellular biology to be addressed. These statistical analyses need to be evaluated in reference to a null model. A network with randomised information embedded in the nodes, the edges and/or the timings can act as a suitable null model. Alternatively, the disruption of the causality structure implied by the timings allows temporal sequential randomisation of events and states. A reference network model of the expected transient behaviour of the system under investigation can also act as a null model.
The most pertinent aspect of the temporal-topological structures is the involvement of timings as an additional parameter. The existing connectivity metrics of network analysis such as path lengths and shortest path lengths as well as centrality measures need to be modified in order to become “time-respecting” measures. “wait-times” can be used to describe timed events that occur with delays. The rate of reaching a node starting from another node, i.e. the fastest paths, the average velocity of the information transferred through the network; i.e. the latency, and the inter-contact times in a dynamic network and refers to the non-uniformity of the timings of communication events, i.e. the burstiness are novel metrics that provide a measure of how fast information is relayed across the network on average.
A number of elements need to be considered in designing a study for investigating the dynamics of eukaryotic networks: (i) physiological and functional characteristics of the system including its temporal behaviour/nature (periodic, cyclic, pulse-like or impulse-like response), which in turn, affects the frequency and the duration of sampling, (ii) the limitations on the type of data that can be collected and the implementation of the experimental method due to the sample volume requirements of the downstream analytics, the sampling time resolution suitable for investigating system dynamics, or to the capability of the experimental techniques to address the research question, and (iii) suitability and availability of methods and tools for the modelling and analysis of time-series data.
The complexity of bio-meta-networks where multiple layers of information represented by different types of biological data would be embedded will necessitate the standardisation of the representation and exchange of meta-bio-network information. Therefore, the network community will soon face the need to build and establish their own standardised language of information exchange, which complies with the “Minimum Information Standards”, to allow ease of verification, exchange and interpretation of network-based information in biological systems. Despite its extensive utilisation, network science is yet to have its spotlight in the analysis of biological systems, and such efforts in standardisation will no doubt expedite the process.