A critical review of computational modelling in psychiatry

Computational psychiatry attempts to translate advances in computational neuroscience and machine learning into improved outcomes for patients. This talk will critically review progress in mechanistic, theory-driven approaches. It will start by introducing the core motivation behind the application computational work, namely the notion of the brain as a fundamentally computational organ. The talk will then outline the more established combination of cognitive-computational concepts and methods to richly elucidate psychiatric symptoms. These approaches have enabled profound insights into the structure and neurobiology of phenomena ranging across the diagnostic categories. However, they have often suffered from uncertain cross-sectional correlations and difficulties translating into improved treatments. The talk will then focus on more recent causal work linking treatments to changes in specific computational processes. This work has opened concrete causal paths towards clinical translation, including through the combination of experimental medicine with precision computational approaches, repurposing, novel computational intervention design and large language models. Finally, the talk will outline remaining barriers, including measurements, outcome definition, the interdisciplinary people gap, and regulation.