Brain activity lies on a very complex network. This complexity is multiple but a prominent one is the non-linear and multi-factorial interaction between its nodes. In the specific case of Parkinson's disease (PD), the brain loses dopaminergic neurons in the substantia nigra compacta (SNc). Highly involved in the basal ganglia-cortex-thalamus-cerebellum (BG-Ctx-T-Cb) network, the main role of dopamine (DA) is in learning, decision-making and motor control, as manifested in the PD symptoms such as rigidity, slowness of movements, tremor and poor behavioural flexibility. Following DA loss, the underlying changes in the cortical activity are both poorly characterised and understood. We are interested here in the changes of the brain region dynamics at the network level. As brain signals we use MEG which has both a good temporal and spatial resolution. Instead of relying on previous techniques that require manual parameter tuning, we extract brain activity networks using the graph Laplacian, a method grounded in graph theory. We then use the recently developed graph signal processing (GSP) theory, see [1], to analyse how MEG brain signals rely on its functional connectivity (FC). Structural connectivity (SC) is more commonly used than FC. Surprisingly, the FC graph Laplacian decomposition is very similar to the one obtained by SC. Indeed, the first eigenvectors show low-frequency spatial patterns as illustrated in Fig. 1. When looking at higher order eigenvectors, they show more and more complex spatial patterns. Granted this similarity, we compared signal distributions across low-, medium-, and high-frequency components in PD and HC groups using this approach. Our analysis reveal significant differences, particularly in sensory regions, indicating a stronger dependence on the underlying graph in PD patients compared to HC.