Frässle S, Marquand AF, Schmaal L, et al. - Given heterogeneous treatment response and highly variable clinical trajectories have been reported among patients with major depressive disorder (MDD), and a key challenge for psychiatry is predicting individual clinical trajectories at an early stage, which might aid individually tailored interventions, but there is a lack of reliable predictors at the single-patient level, so, researchers assessed the efficacy of a machine learning strategy – generative embedding (GE) – which integrates interpretable generative models with discriminative classifiers. The NEtherlands Study of Depression and Anxiety was used to obtain functional magnetic resonance imaging data of emotional face perception from 85 MDD patients. The use of a generative model of effective (directed) connectivity combined with support vector machines made it possible to predict if a given patient would face chronic depression vs fast remission with a balanced precision of 79%. The findings were preliminary but may serve as proof-of-concept, demonstrating that GE has the potential to afford clinical predictions that are interpretable in terms of network mechanisms. A higher risk of developing a less favorable clinical course might exist in correlation with abnormal dynamic alterations of connections involved in emotional face processing.