Modeling population dynamics of neural circuits: from in vitro neural systems to structured cortical networks

Abstract

The collective dynamics of neuronal populations form the basis of complex physiological processes and adaptive behavior in animals. Computational modeling is one of the key tools that facilitate the understanding of population dynamics and its functions. Recent progress in machine learning and brain recording techniques allowed a close integration of experimental recordings and computational modeling. This, on the one hand, enables detailed quantitative fits of experimental data that improve our understanding of basic physiology, network organization, and variability. On the other hand, modern methods help to better characterize neuronal activity and learn the details of the computations these systems perform. In the first part of the thesis, I demonstrate how modeling approaches and simulation-based parameter inference, integrated closely with experiments, can enhance our understanding of network organization principles. In the second part, I apply statistical modeling methods to characterize the neuronal dynamics underlying complex self-correction behavior. Chapter 2 leverages simulation-based inference (SBI) to integrate single-cell properties, network structure, and population dynamics in a model of networks of dissociated neurons in vitro. This approach allowed us to discover that networks adjust the inhibitory connectivity and maintain excitation/inhibition balance under chronic changes in the cellular excitatory/inhibitory composition. Chapter 3 identifies the dynamical states underlying the self-organization of networks of cultured neurons towards collective bursting activity. I show how a reduced model of network activity can explain the difference between networks of rodent and human pluripotent stem cell-derived neurons in vitro that exhibit seemingly identical dynamics. Chapter 4 focuses on the functional implications of the excitation/inhibition connectivity in the spiking networks model and examines how recurrent network structure allows the development of excitation and inhibition balance and stimulus tuning. Chapter 5 turns to the more complex dynamics of cortical networks underlying self-correction behavior in rats. There, statistical modeling and data analysis techniques helped us to uncover the complex organization of neural responses in the Anterior Cingulate Cortex. In the final Chapter 6, I discuss the main limitations of this work and the main future direction.

Die Dissertation ist gesperrt bis zum 20. März 2026 !