Alzheimer’s disease (AD) is a clinically, anatomically and biologically heterogeneous disorder encompassing a wide spectrum of cognitive profiles, ranging from the typical amnestic syndrome to visuospatial changes in posterior cortical atrophy, language deficits in primary progressive aphasia and behavioural/executive dysfunctions in anterior variants. With the emergence of functional imaging and neural network analysis using graph theory for instance, some authors have hypothesized that this phenotypic variability is produced by the differential involvement of large-scale neural networks – a model called ‘molecular nexopathy’. At the moment, however, the hypothesized mechanisms underlying AD’s divergent network degeneration remain speculative and mostly involve selective premorbid network vulnerability. Herein we present an overview of AD’s clinicoanatomical variability, outline functional imaging and graph theory contributions to our understanding of the disease and discuss ongoing debates regarding the biological roots of its heterogeneity. We finally discuss the clinical promises of statistical signal processing disciplines (graph theory and information theory) in predicting the trajectory of AD variants. This paper aims to raise awareness about AD clinicoanatomical heterogeneity and outline how statistical signal processing methods could lead to a better understanding, diagnosis and treatment of AD variants in the future.