Multiscale Modelling of the Mouse Cerebellar Microcircuit: Biophysically Realistic Neural and Synaptic Properties, Ataxic Alterations, and Integration into Whole-Brain Models Coupled with Neurostimulation

The dissertation focuses on multiscale modeling of the mouse cerebellum, with particular emphasis on the development of biologically-grounded spiking neural networks in physiological and pathological conditions, and on the derivation of corresponding mean-field models for mesoscale and whole-brain applications. It also contributes to the development of a whole-brain mouse framework to study how cerebellar dynamics influence large-scale brain activity and modulate the network response to externally induced electric fields in the cerebellum. First, this work presents a detailed cerebellar spiking neural network, enriched with new features and automatically tuned against cell-specific multicompartmental models, including refined point-neuron dynamics and receptor-specific synaptic and short-term plasticity mechanisms. Local field potentials of the granular layer and population-specific firing patterns were compared between simulations and experimental data in healthy mice, providing a fundamental validation step. Then, one application of this model focused on its customization for cerebellar ataxias by incorporating specific structural alterations into the spiking network. The detailed spiking neural network was then used as a generative model for the derivation of a mesoscale description of the cerebellar dynamics, namely a cerebellar mean-field model, by remapping micro-to-mesoscale parameters and estimating input-output relationships specific to each network's neuronal population. This workflow can be generalized to derive proxies of region-specific models, suitable for integration within whole-brain frameworks. To investigate how specific cerebellar dynamics shape macroscopic brain activity in both healthy and ataxic conditions, this work also introduces a virtual mouse brain framework fitted to mouse fMRI data. The goal is to progress toward a co-simulation architecture in which cerebellar nodes are represented by detailed spiking neural networks rather than neural mass models, thereby preserving a fine-grained description of neuron-level ataxic alterations. Finally, within this virtual mouse brain framework, a key methodological step was the development of a dosimetry-informed perturbation map to simulate whole-brain dynamics under cerebellar TMS, derived from region-specific estimates of the induced electric field. This enabled a first integration of cerebellar transcranial magnetic stimulation into the virtual mouse brain framework, making it possible to simulate the propagation of stimulation across the network and to assess its large-scale effects in a mechanistic way.