There is a growing recognition that psychiatric symptoms have the potential to causally interact with one another. Particularly in the earliest stages of psychopathology dynamic interactions between symptoms could contribute heterogeneous and cross-diagnostic clinical evolutions. Current clinical approaches attempt to merge clinical manifestations that co-occur across subjects and could therefore significantly hinder our understanding of clinical pathways connecting individual symptoms. Network approaches have the potential to shed light on the complex dynamics of early psychopathology. In the present manuscript we attempt to address 2 main limitations that have in our opinion hindered the application of network approaches in the clinical setting. The first limitation is that network analyses have mostly been applied to cross-sectional data, yielding results that often lack the intuitive interpretability of simpler categorical or dimensional approaches. Here we propose an approach based on multi-layer network analysis that offers an intuitive low-dimensional characterization of longitudinal pathways involved in the evolution of psychopathology, while conserving high-dimensional information on the role of specific symptoms. The second limitation is that network analyses typically characterize symptom connectivity at the level of a population, whereas clinical practice deals with symptom severity at the level of the individual. Here we propose an approach based on graph signal processing that exploits knowledge of network interactions between symptoms to predict longitudinal clinical evolution at the level of the individual. We test our approaches in two independent samples of individuals with genetic and clinical vulnerability for developing psychosis.