Fig. 2

Optimization and evaluation of the model: First, using only HC subjects, the global coupling parameter G is found, and then the model is adjusted to minimize the distance between the empirical and simulated fMRI data, taking into account the regional burden distributions. A Minimization of G between 0 and 5.5, for functional connectivity (FC), sliding-window functional connectivity dynamics (swFCD), and phase FCD (phFCD). Given their strong similarity in the results, phFCD was used for all subsequent computations. B, C Shows the normalized (in [0, 1]) FCD distributions for the empirical data (top) and the simulated model at the optimal result (bottom). D, E, F Analysis of the impact (smaller values are better) of the different burdens with respect to their impact on the phFCD (KS distance) when optimized together and in isolation, with the homogeneous state as a reference. Clearly, in all cases, the combined burden outperforms any other model. However, as can be seen, the results for AD clearly show that tau alone accounts for the vast majority of the weight of the impact on brain activity (F), while for MCI patients it is A\(\beta\) who dominates (E). For HC patients we also see a predominance of A\(\beta\), although with less difference between the model incorporating A\(\beta\) and tau vs. A\(\beta\) in isolation (D). Average distributions of A\(\beta\) (G) and tau burdens (H) over each cohort (using ADNI’s database). Colors correspond to the normalized burden of each protein. The increase in A\(\beta\) and tau can be clearly seen