Noise-driven multistability versus deterministic chaos in phenomenological semi-empirical models of whole brain activity

Abstract

An outstanding open problem in neuroscience is to understand how neural systems are capable of producing and sustaining complex spatiotemporal dynamics. Computational models that combine local dynamics with in vivo measurements of anatomical and functional connectivity can be used to test potential mechanisms underlying this complexity. We compared two conceptually different mechanisms: noise-driven switching between equilibrium solutions (modeled by coupled Stuart-Landau oscillators) and deterministic chaos (modeled by coupled Rossler oscillators). We found that both models struggled to simultaneously reproduce multiple observables computed from the empirical data. This issue was especially manifest in the case of noise-driven dynamics close to a bifurcation, which imposed overly strong constraints on the optimal model parameters. In contrast, the chaotic model could produce complex behavior over an ampler range of parameters, thus being capable of capturing multiple observables at the same time with good performance. Our observations support the view of the brain as a non-equilibrium system able to produce endogenous variability. We presented a simple model capable of jointly reproducing functional connectivity computed at different temporal scales. Besides adding to our conceptual understanding of brain complexity, our results inform and constraint the future development of biophysically realistic large-scale models. The quote "What I cannot create, I do not understand" was found written in the blackboard of celebrated physicist Richard Feynman at the time of his death. This sentence suggests a way forward for neuroscientists interested in unravelling the principles behind the richness and complexity of spontaneous brain dynamics. Over the last decades, tremendous advances in neuroimaging enabled the construction of whole-brain activity models with real predictive power in the statistical sense. It is now possible to create realistic complex dynamics, instead of passively screening for their presence in neuroimaging data. We contrasted two different types of building blocks (i.e. two choices of local dynamics) and tested their capacity to reproduce the empirical data, with the purpose of increasing our conceptual understanding of the mechanisms behind large-scale spontaneous activity in the human brain.