Linking Functional Brain Networks To Psychopathology And Beyond

Neurobiological abnormalities associated with neuropsychiatric disorders do not map well to existing diagnostic categories. High co-morbidity suggests dimensional circuit-level abnormalities that cross diagnoses. As neuropsychiatric disorders are increasingly reconceptualized as disorders of brain development, deviations from normative brain network reconfiguration during development are hypothesized to underlie many illness that arise in young adulthood. In this dissertation, we first applied recent advances in machine learning to a large imaging dataset of youth (n=999) to delineate brain-guided dimensions of psychopathology across clinical diagnostic boundaries. Specifically, using sparse Canonical Correlation Analysis, an unsupervised learning method that seeks to capture sources of variation common to two high-dimensional datasets, we discovered four linked dimensions of psychopathology and connectivity in functional brain networks, namely, mood, psychosis, fear, and externalizing behavior. While each dimension exhibited an unique pattern of functional brain connectivity, loss of network segregation between the default mode and executive networks emerged as a shared connectopathy common across four dimensions of psychopathology. Building upon this work, in the second part of the dissertation, we designed, implemented, and deployed a new penalized statistical learning approach, Multi-Scale Network Regression (MSNR), to study brain network connectivity and a wide variety of phenotypes, beyond psychopathology. MSNR explicitly respects both edge- and community-level information by assuming a low rank and sparse structure, both encouraging less complex and more interpretably modeling. Capitalizing on a large neuroimaging cohort (n=1,051), we demonstrated that MSNR recapitulated interpretably and statistically significant associations between functional connectivity patterns with brain development, sex differences, and motion-related artifacts. Compared to common single-scale approaches, MSNR achieved a balance between prediction performance and model complexity, with improved interpretability. Together, integrating recent advances in multiple disciplines across machine learning, network science, developmental neuroscience, and psychiatry, this body of work fits into the broader context of computational psychiatry, where there is intense interest in the quest of delineating brain network patterns associated with psychopathology, among a diverse range of phenotypes