Characterizing Complex Patterns of Neural Signals: From Modeling of Neuronal Networks to Data-Driven Analysis of Epileptic High Frequency Oscillations

Abstract

The brain's extraordinary complexity, characterized by intricate networks and dynamic processes, presents significant challenges in unraveling its underlying mechanisms. Addressing these complexities naturally gives way to the integration of computational neuroscience, which serves as a vital bridge between theoretical models and empirical data to enhance our understanding of neural function and dysfunction. Here, we employ both model-driven and data- driven approaches, integrating techniques taken from multiple disciplines including physics, statistics, data science and engineering to investigate fundamental and clinical aspects of neural dynamics, offering insights into neurological phenomena and disorders. Starting with the first topic, we adopted a model-driven approach to investigate how neuromodulatory changes in acetylcholine (ACh) during sleep-like states could reorganize large- scale neuronal networks. By implementing Hodgkin–Huxley-like neurons arranged in a scale- free topology, we reveal that low ACh levels, analogous to slow-wave sleep, cause synchronous, lower-frequency firing patterns and drive selective synaptic reconfiguration via spike-timing- dependent plasticity (STDP). These results suggest a plausible mechanistic route by which sleep down-regulates global synaptic strengths while preserving key hub-to-nonhub connections that may be critical for learning. Shifting to a clinical context, the next topic employs data-driven analyses of high- frequency oscillations (HFOs) (80–500 Hz) recorded from intracranial EEG in patients undergoing epilepsy surgery. Traditionally treated as isolated events in a single location, we view HFOs instead as manifestations of network discharges spanning multiple electrodes. By characterizing the functional connectivity of HFO events, this chapter shows that network-based HFO measurements predict surgical outcome more reliably than simpler metrics, particularly in “definitive” surgeries where most of the clinically defined seizure onset zone (SOZ) has been resected. Such findings underscore HFO networks as a robust biomarker for epileptogenic tissue and open new opportunities for predictive models of post-surgical outcome. Building on these insights, the last topic tackles the longstanding challenge of distinguishing “pathological” HFOs—indicative of epileptic hyperexcitability—from benign “physiological” HFOs linked to normal processes such as memory consolidation. By integrating EEG signals, spectral power distributions, and a set of morphometric features, we train a suite of machine-learning classifiers, including convolutional neural networks and logistic regression, to discriminate between HFOs that appear within resected epileptogenic zones and those that arise in non-epileptic tissue. The models achieve robust separation under leave-patient-out cross- validation, revealing that key features such as overall signal power and lower-frequency dominance are strongly associated with pathological HFOs. Overall, we demonstrate how advanced computational techniques effectively address diverse research questions spanning fundamental neurobiology and clinical neuroscience. The model-driven simulations in the first topic provided mechanistic insights into how neuromodulatory changes orchestrate large-scale network dynamics and synaptic plasticity, thereby enhancing our understanding of sleep mechanisms. Concurrently, data-driven analyses in the latter topics leveraged signal processing, complex network theory, and machine learning to unravel the complexities of high-frequency oscillations (HFOs) in epilepsy, enabling predictions of surgical outcomes and distinguishing pathological from physiological activity. By integrating multidisciplinary computational approaches, this dissertation successfully bridges theoretical models with empirical data, thereby advancing both basic scientific knowledge and clinical applications in the field of neuroscience.