Fig. 1: Study overview.

Schematic of study procedures. All participants received the same testing procedures. Depressed participants were recruited from a national Australian database, and the sample was enriched for a higher likelihood of melancholia. Participants completed a clinical assessment with a psychiatrist, during which they were classified as melancholic or non-melancholic using a validated semi-structured interview (the Sydney Melancholia Prototype Index). Dynamic changes in positive facial expressivity were analysed using a convolutional neural network trained on the participant’s response to a movie clip of a stand-up comedian. We modelled brain activity in a set of brain regions associated with the positive evaluation of stimuli (see section ‘Materials and Methods’) during exposure to an emotionally evocative movie in the MRI scanner. Markers of facial affect and differences in patterns of brain activity were compared between melancholic and non-melancholic groups.