The emerging field of computational psychiatry uses computational modeling to improve the understanding, prediction and treatment of mental disorders. This symposium will highlight recent developments in theory-driven computational neuroscience, which uses formal models of brain function to elucidate the mechanisms of psychopathology. Focusing on psychosis, we will put forward the notion of psychotic symptoms as maladaptive inferences resulting from altered predictive brain mechanisms. Researchers from different disciplines and backgrounds will present recent advances in relating psychosis to neural computations within a Bayesian framework. Katharina Schmack will discuss how cross-species computational psychiatry approaches can provide insights into neural circuits relevant to psychosis. She will introduce a behavioral task and computational model to capture hallucination-like inferences in humans and mice alike, and present results elucidating the causal role of striatal dopamine in such hallucination-like inferences. Veith Weilnhammer will discuss psychosis as a result of altered perceptual inference. Findings from neuroimaging, brain stimulation and computational modeling indicate a key role for prefrontal cortex in perceptual inference. Patients with schizophrenia show alterations in such inference, suggesting non-invasive stimulation of prefrontal cortex as a promising new treatment approach. Andreea Diaconescu will focus on social inference and present work combining probabilistic reward learning tasks with Bayesian modeling to compare social inference mechanisms in early psychosis and schizophrenia to those in depression and borderline personality disorder. Results reveal a pattern shared between patients with schizophrenia and borderline personality disorder, who both show over-reliance on predictions about social information. Finally, all presenters will discuss outstanding questions in computational psychiatry and ways to address them in future research.