Maintaining an adaptive balance of emotions is central to well-being, and dysregulated emotions contribute broadly to clinical disorders that impart high personal and societal burdens. Recognizing the transdiagnostic importance of emotion to mental health, the National Institute of Health's Research Domain Criteria (RDoC) matrix contains overarching domains of Negative Valence, Positive Valence, and Arousal. However, the matrix underspecifies how specific affective states like sadness, anxiety, or craving are organized within and across these domains, in part because it is unknown whether representations of discrete emotions are reliably differentiated. Other RDoC constructs, such as rumination and worry, modify the temporal parameters of emotions that confer psychopathology risk and exacerbate symptom maintenance. Nonetheless, it is unknown how these processes interface with emotional brain circuits to impact affect dynamics, particularly as they often occur spontaneously during mind wandering. The proposed research promises to improve the RDoC depiction of these emotion-related constructs by taking an affective computing approach. During combined recording of psychophysiology and functional magnetic resonance imaging (fMRI), adult participants will experience emotions to vignettes and movie clips spanning the arousal and valence dimensions, and will report on their spontaneous emotions during resting-state fMRI scans. Machine learning algorithms will decode emotion- specific signals across the levels of analysis, which will be integrated using Bayesian state-space modeling. An analysis of classifier errors will test competing predictions from emotion theories regarding the optimal structure of affective space. Using graph theoretic tools, we will characterize the neural network architecture of the discrete emotion representations to identify provincial and connector hubs that can be used as novel targets for future symptom-specific or co-morbid neuromodulation interventions, respectively. We will apply the emotion-specific maps to resting-state data from the same participants to create neurophysiological indices of spontaneous emotions and to relate their frequencies to measures of trait and state affect as a validation step. Using stochastic modeling of the resting-state data, we will derive temporal dynamics metrics to test the hypothesis that rumination and worry promote emotional inertia during mind wandering. Finally, we will use existing data repositories to demonstrate that our novel indices of affect dynamics transdiagnostically differentiate resting-state fMRI activity patterns in mental health disorders from healthy controls. The proposed research will improve upon current RDoC formulations of Negative Affect, Positive Affect, and Arousal domains by informing how discrete emotions are organized within and across these domains, by integrating emotion representations across multiple RDoC units of analysis, by informing how rumination and worry impact neurophysiological signatures of spontaneous emotions, and by establishing the clinical utility of computationally-derived metrics of emotion dynamics.
The overall goal of this project is to characterize how emotions are represented in the brain and autonomic nervous system. Computational modeling approaches will be applied to integrate the data across multiple sources and tasks, to test competing theories about how emotions are organized, and to derive new metrics of emotion dynamics. The outcome of the work will inform how emotions should be conceptualized within a broader research framework of mental health domains.
