The connectivity map of a living neuronal network, together with the intrinsic neuronal dynamics, shapes in great measure the collective behavior and functionality of a neuronal circuit [1,2]. While neuronal dynamics can be well analyzed and modeled, accessing the connectivity blueprint of a neuronal network is a highly difficult task, particularly in living circuits. Hence, a major effort has been devoted to seek for strategies to bring to light such information, either as a detailed, neuron-to-neuron connectivity map or through statistical descriptors.
Combining experiments and modeling, we here present an innovative framework in the context of an Ising model to unveil key topological features of neuronal networks in vitro, specifically their average connectivity and distribution of connections. Experimentally, our networks consist of cultures of rat cortical neurons at 14 days of development and with diverse density and spatial distributions. Through electrical stimulation and calcium imaging (Fig. 1), we measure the transition from an ensemble of connected neurons to a disconnected population as the connectivity probability between neurons is gradually reduced. We considered as order parameter the size of the largest group of connected neurons (giant component), and investigate the characteristics of both the transition curve and point [3,4,5]. Theoretically, we use Stastical Physics to map the behavior of the neuronal network into a Random Field Ising model (RFIM) [6,7,8] in a mean-field approximation. Percolation is then introduced by reducing the average connectivity strength between neurons. Our model can be solved analytically and the solutions depend just on few parameters that can be ascribed to experimental details. The fitting of the theoretical model to experimental data uncovers two key topological aspects, namely the average connectivity of the network and the degree of heterogeneity of the connectivity map, quantified through the discordance between the mean-field theoretical curves and the actual data.
