Machine learning-derived multimodal Neurobiological profiles of behavioral activation traits in adolescents.

Behavioral activation (BA) traits mediate responses to positive reinforcement, and then to promote reward-seeking actions. However, few studies have investigated the neurobiological profiles of BA traits in adolescents based on multimodal neuroimaging and machine learning techniques. In this study, a total of 6626 adolescents with both valid multimodal magnetic resonance imaging (MRI) and questionnaire data were included in the Adolescent Brain Cognitive Development Study. Machine learning-based elastic net regression with 5-fold cross-validation (CV) was used to characterize the neurobiological profiles of BA traits using multimodal MRI data as predictors. Using 5-fold CV, the multi-region neurobiological profiles substantively predicted BA traits, and this finding was robust in an out-of-sample. Regarding specific regions, neurobiological profiles were enriched in the bilateral pallidum. Regarding functional networks, functional connectivity of the cingulo-opercular and the fronto-parietal networks with both the pallidum and nucleus accumbens, showed high beta weights. The relationships of the neurobiological profiles with BA traits were further supported by traditional univariate linear mixed effects models, in which many of the profiles identified as part of the neurobiological pattern showed significant univariate associations with BA traits, including the hub region pallidum. In summary, these findings revealed robust machine learning-derived neurobiological profiles of BA traits, those that comprised a key node the pallidum, which is involved in the motivational brain network. These findings suggested that the pallidum might play a vital role in developing BA traits in adolescents.