h1

TY  - THES
AU  - Gutzen, Robin
TI  - Analysis and quantitative comparison of neural network dynamics on a neuron-wise and population level
VL  - 102
PB  - RWTH Aachen University
VL  - Dissertation
CY  - Jülich
M1  - RWTH-2024-03073
SN  - 978-3-95806-738-7
T2  - Schriften des Forschungszentrums Jülich. Reihe Information/information
SP  - 1 Online-Ressource : Illustrationen
PY  - 2024
N1  - Druckausgabe: 2024. - Onlineausgabe: 2024. - Auch veröffentlicht auf dem Publikationsserver der RWTH Aachen University
N1  - Dissertation, RWTH Aachen University, 2023
AB  - Our goal is to better understand the working mechanisms of biological neural systems. To this end, describing neural systems as networks provides a powerful and widely-used analysis approach. The network syntax of interacting nodes exhibiting joint dynamics facilitates the quantitative characterizations of neural systems across scales. Moreover, this approach enables us to construct systematic comparisons of neural network descriptions across domains. We aim to identify characterizations of neural network activity that reflect the underlying connectivity and relate to the network's ability to process information. In this context, we explore characteristic measures from experimental and simulated data sources. Concretely, we look at cortical activity data from mice and monkeys, from different recording techniques like implanted electrode arrays, laminar probes, ECoG, and calcium imaging, and further from simulations of stochastic processes, spiking, and mean-field network models. We investigate activity measures of different complexity, including measures on the level of individual neurons, higher-order measures of coordinated spiking activity, and population-level field potential measures describing spatial wave patterns. Such activity characterizations always represent an abstraction, and the right level of detail depends on the data type and the question of interest. For a given context, the appropriate abstraction level allows us to integrate and compare data and models from heterogeneous sources. Evaluating the similarity between such different network descriptions is a common demand in computational neuroscience. Extending the concept of validation, we formalize and apply cross-domain comparisons in model vs. experiment, model vs. model, and experiment vs. experiment scenarios. In this framework, we further evaluate and extend existing statistical testing approaches and look at reproducibility, sources of variability, and technical limitations. Through our exploration of network activity characterizations and their comparability, we evaluate the relationship between network connectivity, activity, and function. Concretely, over the course of five research projects, we implement and demonstrate systematic approaches to validate model simulators, statistically evaluate network organization, infer network connectivity from activity data, combine data sources of wave activity, and relate wave activity to external influences and behavior. With a focus on open and collaborative science practices, we implement our methodologies as reusable open-source tools while building upon existing open-source tools and standards.
LB  - PUB:(DE-HGF)11 ; PUB:(DE-HGF)3
DO  - DOI:10.18154/RWTH-2024-03073
UR  - https://publications.rwth-aachen.de/record/981487
ER  -