The pervasive impact of Alzheimer's disease on aging society represents one of the main challenges at this time. Current investigations highlight two specific misfolded proteins in its development: Amyloid-beta and tau. Previous studies focused on spreading for misfolded proteins exploited simulations, which required several parameters to be empirically estimated. Here, we provide an alternative view based on a machine learning approach. The proposed method applies an autoregressive model, constrained by structural connectivity, to predict concentrations of Amyloid-beta two years after the provided baseline. In experiments, the autoregressive model generally outperformed the state-of-art models yielding the lowest average prediction error (mean squared-error 0.0062). Moreover, we assess its effectiveness and suitability for real case scenarios, for which we provide a web service for physicians and researchers. Despite predicting amyloid pathology alone is not sufficient to clinical outcome, its prediction can be helpful to further plan therapies and other cures.