Modelling causal and multiscale network dynamics to understand complex brain functions: Application to health and disease

The human brain operates as a complex, multiscale system in which coordinated activity emerges from interactions across distributed networks. Understanding this complexity requires investigating connectivity at multiple levels such as structural, functional, and effective connectivity, which capture the brain’s architecture and function across different dimensions. Advances in computational neuroimaging have made it possible to model these dimensions using biophysically informed frameworks. Among these, Dynamic Causal Modelling (DCM) can be used to estimate directed, so called effective connectivity between brain pairs of regions, while The Virtual Brain (TVB) simulates large-scale neural activity constructed by individual structural and physiological parameters. This thesis applies these frameworks to investigate how brain networks maintain, adapt, and reorganize their connectivity across motor, developmental, and neuroinflammatory conditions. Using DCM and TVB, this thesis explores how directed communication and excitation/inhibition balance shape healthy brain function and how their disruption contributes to pathology. In the first study, DCM was used to examine action execution (AE) and action observation (AO) to test the hypothesis that how cerebro-cerebellar loops implement sensorimotor prediction. A visuomotor network including primary motor, parietal, and cerebellar regions showed shared architecture across AE and AO, with cerebellar output switching from excitatory to inhibitory depending on task context. This revealed the cerebellum as a dynamic regulator of cortical communication and predictive control. Building on this, the second study extended the modelling framework to developmental dyslexia (DD), integrating imaging and genetic data to investigate how reading proficiency and the DCDC2-related READ1 deletion variant influence cortico-cerebellar coupling. Results showed that children with DD carrying READ1d exhibited reduced effective connectivity between the cerebellum and dorsal magnocellular and ventral attention regions, while reading proficiency was linked to stronger cerebellar modulation of visuo-attentional processes; highlighting the cerebellum’s integrative role in linking perceptual and cognitive functions. Finally, the third study combined DCM and TVB information to study the relevance of alterations in directed connectivity and excitation/inhibition dynamics for clinical measures in a cohort including people with multiple sclerosis (pwMS); This final study examined interactions using task-based, resting-state, and diffusion imaging data. Results revealed preserved overall network architecture but altered effective connectivity in pwMS compared to healthy volunteers. TVB-derived inhibitory and excitatory parameters further correlated with disability and motor performance, suggesting that maladaptive regulation of inhibitory control contributes to functional impairment. Collectively, these studies demonstrate how computational models such as DCM and TVB can bridge local causal interactions with large-scale network dynamics, offering mechanistic insights into brain reorganization across distinct domains. By pointing towards an integration of multimodal imaging with biologically interpretable modelling, this thesis contributes to the development of multiscale, model-informed biomarkers that may guide personalized diagnosis, prognosis, and possible treatment strategies in complex brain disorders.