Early identification can improve outcomes for individuals at high risk of developing psychosis, and multimodal machine-learning models may be able to help. In a study comparing participants with clinical high risk syndromes or recent-onset depression to healthy volunteers, researchers used a machine-learning model incorporating clinical and neurocognitive data, structural magnetic resonance imaging (sMRI), and polygenic risk scores for schizophrenia to predict development of psychosis. The model accurately predicted transition to psychosis in 85.9% of cases, compared to 73.2% of cases predicted by clinicians. The model showed greater predictive accuracy than any individual predictive factor alone. To learn more, see the study in JAMA Psychiatry.