Project Summary Emotions play a critical role in organizing human experience and behavior, and emotion dysregulation lies at the heart of psychopathology and functional impairment across disorders. To measure and understand emotion dysregulation, advances in understanding the fundamentals of how the brain generates and represents emotional states are vitally needed. This proposal develops and validates brain models underlying emotional states in naturalistic, narrative contexts. We combine Functional Magnetic Resonance Imaging (fMRI), multimodal latent factor models, natural language processing, and pattern recognition techniques to develop models of brain activity that characterize how individuals generate emotional experiences. Over the previous project period, we have developed several technical innovations to identify dynamic emotional states from multimodal data and how they vary moment-by-moment during fMRI scanning. These models allow different individuals to experience and express latent emotional states at different times, accounting for the idiosyncratic interpretations of events that are a hallmark of human emotional responses. We elicit emotional experiences using dynamic, naturalistic movies and autobiographical stories. In Experiment 1, we infer latent emotional states using a novel application of the Shared Response Model (SRM), a latent factor model that integrates multiple simultaneously acquired measurement modalities including: moment-by-moment subjective ratings inferred using an innovative collaborative filtering approach, automatically decoded facial expressions using computer vision techniques, and psychophysiological signals. We then use these emotion signals to identify distributed patterns of brain activity that track distinct emotional states. In Experiment 2, we characterize how cortical-subcortical circuits involved in appraisal–and particularly the ventromedial prefrontal cortex–generate interpretations of unfolding events that give rise to emotional experiences. We leverage high-dimensional semantic embeddings of participants’ appraisals as revealed by a think-aloud protocol. The resulting brain models of specific emotion categories afford several potentially transformative advantages. Such models can (a) provide insight into which systems are necessary and sufficient for emotion generation; (b) be shared and tested across studies, allowing us to evaluate their generalizability across contexts; and (c) provide targets for psychological and neurological interventions. Together, these studies will yield generalizable models of the dynamic brain patterns underlying specific emotional experiences. Such models could transform the study of emotion by providing ways of capturing the moment-by-moment dynamics of emotional states, and clinical research by allowing investigators to test effects of psychological interventions on brain targets related to specific emotions.