Developing algorithmic psychiatry via multi-level spanning computational models

Modern psychiatry faces challenges in translating neurobiological insights into treatments for severe illnesses. The mid-20th century witnessed the rise of molecular mechanisms as pathophysiological and treatment models, with recent holistic proposals keeping this focus unaltered. In this perspective, we explore how psychiatry can utilize systems neuroscience to develop a vertically integrated understanding of brain function to inform treatment. Using schizophrenia as a case study, we discuss scale-related challenges faced by researchers studying molecules, circuits, networks, and cognition and clinicians operating within existing frameworks. We emphasize computation as a bridging language, with algorithmic models like hierarchical predictive processing offering explanatory potential for targeted interventions. Developing such models will not only facilitate new interventions but also optimize combining existing treatments by predicting their multi-level effects. We conclude with the prognosis that the future is bright, but that continued investment in research closely driven by clinical realities will be critical.