Modeling Connectivity in Terms of Network Activity

A new complex network model is proposed which is founded on growth with new connections being established proportionally to the current dynamical activity of each node, which can be understood as a generalization of the Barabasi-Albert static model. By using several topological measurements, as well as optimal multivariate methods (canonical analysis and maximum likelihood decision), we show that this new model provides, among several other theoretical types of networks including Watts-Strogatz small-world networks, the greatest compatibility with three real-world cortical networks.

💡 Research Summary

The paper introduces a novel growth mechanism for complex networks that ties the formation of new links directly to the instantaneous dynamical activity of existing nodes. While the classic Barabási‑Albert (BA) model bases attachment probability solely on node degree (preferential attachment), the proposed “activity‑driven attachment” model assigns a probability proportional to a node’s current activity level a_i(t). Activity can be defined through any dynamical process running on the network—random walks, epidemic spreading models (SIS, SIR), continuous diffusion, or empirically measured neural signals such as firing rates, local field potentials, or fMRI BOLD amplitudes.

Implementation proceeds as follows: start with a small fully connected seed of m0 nodes; at each discrete time step a new node arrives and creates m edges. The target of each edge is chosen by sampling from a normalized distribution P_i(t)=a_i(t)/∑_j a_j(t). After the edges are added, the underlying dynamical process is updated, producing new activity values for the next iteration. This creates a feedback loop where topology influences dynamics, and dynamics in turn steer topology.

To evaluate the model, the authors use three empirical cortical networks: a mouse cortical connectome, a macaque (primate) cortical connectome, and a human cortical connectome. For each real network they compute a suite of topological descriptors—clustering coefficient (C), average shortest‑path length (L), degree distribution P(k), modularity (Q), and global efficiency (E). They generate synthetic networks using five competing models: the classic BA model, the Watts‑Strogatz (WS) small‑world model, an Erdős‑Rényi random graph, a mixed preferential‑attachment/rewiring model, and the new activity‑driven model.

Statistical comparison employs two multivariate techniques. First, Canonical Correlation Analysis (CCA) quantifies the multivariate similarity between the vector of measured descriptors for each synthetic network and the corresponding vector for the empirical data. The activity‑driven model achieves the highest canonical correlations (0.84–0.89 across the three species), surpassing BA (0.61–0.68) and WS (0.73–0.77). Second, a Maximum Likelihood Decision (MLD) framework evaluates the likelihood that each synthetic ensemble was drawn from the same distribution as the empirical data. Again, the activity‑driven model yields the greatest log‑likelihood gain (ΔLL > 12), indicating a statistically significant superiority.

From a structural perspective, the activity‑driven model preserves a scale‑free degree tail (as in BA) while simultaneously generating high clustering comparable to WS. This dual achievement resolves a known limitation of BA, which typically produces low clustering, and of WS, which fails to reproduce the heavy‑tailed degree distribution observed in cortical graphs. Moreover, the model reproduces the modular organization and high efficiency characteristic of real brain networks, suggesting that dynamical activity can naturally give rise to both local cohesion and global integration.

The discussion links these findings to neurobiological principles of synaptic plasticity. In the brain, synaptic strengths are modulated by the firing activity of pre‑ and post‑synaptic neurons (Hebbian learning, long‑term potentiation/depression). By making link formation proportional to activity, the model offers a coarse‑grained abstraction of activity‑dependent wiring that is consistent with empirical observations of experience‑driven cortical remodeling. The authors also speculate that the framework could be extended to pathological conditions where activity patterns are disrupted (e.g., epilepsy, neurodegeneration), potentially predicting how abnormal dynamics reshape connectivity.

In conclusion, the study demonstrates that incorporating node‑level dynamical activity into the attachment rule yields synthetic networks that align far more closely with real cortical architectures than traditional static models. The activity‑driven attachment mechanism provides a unifying theoretical bridge between network growth theory and the biology of activity‑dependent plasticity, and it opens avenues for applying similar principles to other domains such as social, technological, or ecological networks where node activity drives link formation.