Hannah Zoé Kerger

Flash talk \ Manfred Eigen lecture theatre

Understanding the dynamical behaviour of neural networks is a fundamental challenge in neuroscience. Theoretical approaches usually provide complete knowledge of the phase space, facilitating a thorough characterisation of the dynamical system. Characterising experimental systems, however, presents a far greater challenge. Neuronal cultures possess a large number of degrees of freedom, many of which are unknown, which makes direct dynamical measures extremely difficult [1, 2]. Furthermore, conventional information-theoretic approaches such as mutual information or entropy involve multiple processing steps, strong statistical assumptions and free parameters [3].
Here, we propose reproducibility as a more direct and assumption-free alternative measure of network dynamics.
Using optogenetically driven primary rodent neuronal cultures, we deliver the same patterned stimulus repeatedly across trials and quantify the degree to which the network reproduces its response. In a purely chaotic system, the sensitivity to initial conditions would lead to a divergence of the responses across trials [4]. Thus, a high reproducibility indicates that the network is not in a purely chaotic regime. The degree of reproducibility can therefore provide an empirical proxy for the degree of chaos in the network dynamics. Furthermore, this approach is parameter-free and directly interpretable, making it robust to analytical biases and applicable across experimental preparations.
Crucially, this approach is not limited to mouse neuronal cultures, but is directly applicable to iPSC-derived neuronal networks and organoids [5]. It could provide an assumption-free characterisation of network dynamics in patient-derived models, enabling the detection of network-level dysfunction in diseased relative to healthy controls.