Application of hyperalignment to resting state data in individuals with psychosis reveals systematic changes in functional networks and identifies distinct clinical subgroups

Abstract

Psychosis related disorders are severe and difficult to define with brain-based biomarkers due, in part, to heterogeneous psychosis symptoms and individual differences in the brain. Recent innovations in computational neuroscience may address these difficulties. Hyperalignment aligns voxel-wise patterns of neural activity across individuals to improve signal in brain data. Transformation metrics may also serve as biomarkers that reflect clinically relevant differences in pattern connectivity (scale), baseline connectivity (translation), and network topography (rotation). In the present study, we apply hyperalignment to resting state functional connectivity between the frontal cortex and regions throughout the brain in a sample of individuals diagnosed with psychosis and healthy controls. We used binary class support vector machines (SVM) to classify psychosis using unaligned (accuracy=66.50%, p=0.0009) and hyperaligned data (accuracy=65.85%, p=0.0011). Follow-up analyses then used voxelwise rotation estimates to characterize those who were accurately versus inaccurately classified. This revealed two distinct biological subgroups of psychosis characterized by distinct topography of frontal connectivity. Additional analyses relate psychosis to composites of hyperalignment transformations. We report reduced pattern connectivity (t=-2.69, p=0.008) and heightened baseline connectivity (t=2.90, p=0.004) in the psychosis group. These findings may highlight imbalanced frontal connectivity, as those in the psychosis group appear to show general patterns of heightened frontal connectivity while connectivity in more specific regions appear blunted. Results highlight differences in frontal cortex connectivity related to psychosis. Novel methods in the present work may provide a path for future work to apply hyperalignment to brain data from clinical populations to accurately characterize clinical subpopulations within diagnostic categories.