EEG-based brain connectivity and sentiment analysis from smartphone social communication: insights into remitted major depressive disorder among adolescents.

Although prior research has identified abnormal brain connectivity in remitted depressed adolescents, there is limited work associating these alterations with real-time affective dynamics, which may shed insight about specific biological markers that contribute to depression recurrence. Effective brain connectivity was estimated using renormalized partial directed coherence (rPDC) from resting-state EEG data collected in adolescents (N = 144; ages 13-18), including youth with remitted depression (n = 85) and healthy controls (n = 59). Additionally, over the course of 180 days, ~2.3 million messages from adolescents' smartphones were passively obtained, and sentiment (i.e., words connoting positive and negative emotion) was extracted using the Python tweetNLP package. At the baseline and 6-month follow-up assessments, clinical interviews were administered to assess depressive symptom severity. Compared to healthy adolescents, youth with remitted depression exhibited hyperconnectivity in theta and delta frequency bands as well as hypoconnectivity in alpha, beta, and gamma in occipito-temporal regions (ps < 0.05). Across all participants, lower connectivity between the precuneus and middle temporal gyrus within the beta and gamma frequency bands was associated with greater negative sentiment in smartphone-based language (beta: B = -0.019, p = 0.006; gamma: B = -0.180, p = 0.007) but not depressive symptoms (beta: B = 0.073, p = 0.415; gamma: B = 0.093, p = 0.289). Conversely, lower alpha band connectivity from the mid cingulate cortex to the precuneus is associated with greater depressive symptoms at follow-up relative to baseline (B = -0.239, p = 0.013). These findings suggest that resting-state effective connectivity may serve as a neural marker of vulnerability for elevated depressive symptoms and negative affective expression during adolescence, highlighting potentially separable neurophysiological targets that, if replicated, could inform future preventive interventions.