A computational approach for interpreting aggregate signals in terms of neuronal mechanisms

The balance between excitatory (E) and inhibitory (I) neural activity is a fundamental principle of brain function. By stabilizing yet preserving flexibility in network dynamics, E–I balance supports perception, cognition, and adaptive behavior. Disruptions of this balance are implicated in several neurological and psychiatric disorders, including autism, schizophrenia, and Alzheimer’s disease. However, directly measuring E–I balance—especially in humans—remains challenging, motivating the development of computational and non-invasive approaches that infer circuit properties from aggregate neural signals. In this thesis, we developed biophysically grounded computational models linking synaptic and receptor-level mechanisms to macroscopic electrophysiological signals such as local field potentials (LFPs) and electroencephalography (EEG). Using biologically plausible E–I network models, we showed that changes in E–I balance produce distinct alterations in both aperiodic (Hurst exponent, H) and periodic (γ-band power) spectral features. Combining these measures yielded a robust estimator of network excitability. We validated this biomarker experimentally using chemogenetic manipulations in mice that systematically increased or decreased excitability, confirming that the model-derived measure accurately tracked firing rates. We then extended the framework to incorporate NMDA receptor–mediated excitation, revealing that NMDA strength strongly modulates the low-frequency spectral slope, particularly during persistent post-stimulus activity. This suggests that spectral slope may serve as a non-invasive marker of NMDA receptor function. Finally, we showed that NMDA- and AMPA-mediated currents differentially shape information encoding: NMDA enhances representation of temporal stimulus structure in the β band, whereas AMPA-dominated dynamics preferentially encode stimulus magnitude in the γ band. Strong γ oscillations, however, impair temporal encoding, revealing a trade-off between synchronization and sensitivity to input timescales. Together, these findings establish a mechanistic framework for interpreting macroscopic neural signals in terms of synaptic and receptor-level dynamics.