Cognition-centric brain activity across diverse imaging tasks constrains the representation of mental health in brain function data

Robust brain-based mental health biomarkers remain largely elusive. One line of thought attributes this to suboptimal quality and modeling of brain and behavioral data. However, an alternative explanation is that neural activity evoked by common brain imaging paradigms simply reflects little about mental health. Here, we find evidence for this latter explanation by examining multivariate mental health and brain function profiles in hundreds of individuals using high-reliability, latent mental health factors derived from 87 neurocognitive/psychiatric assessments, and tensor independent component analysis of five common functional magnetic resonance imaging (fMRI) task and movie-watching paradigms with putative mental health relevance. Across all tasks and brain networks, individual differences in evoked brain activity poorly reflected variability in negative affect, positive affect, or substance use - despite robustly reflecting variability in cognition. Moreover, across-subject diversity in task-evoked brain function profiles significantly lagged across-subject diversity in mental health profiles. Finally, clustering subjects by their brain function profile versus by their mental health profile produced discordant subtypes, and significantly modulated the findings of group-difference analyses. Findings empirically bound the recoverable information about non-cognitive mental health in normative samples using standard task batteries and motivate use of alternative paradigms with higher sensitivity to individual-specific affective and motivational signals.