Transdiagnostic connectome-based predictive modeling of many behavioral phenotypes reveals brain network mediators of clinical-cognitive relationships
Connectome-based predictive modeling (CPM) applied to functional MRI connectivity data can identify brain networks that vary with behavioral measures across subjects. The prediction strength also provides an index of how closely an external instrument relates to specific brain networks, potentially impacting their clinical interpretation. Here we use a deeply phenotyped transdiagnostic population (n = 317) to evaluate CPM performance across a variety of clinical and cognitive measures. The findings revealed a wide range of predictive performance for external instruments, with cognitive tests generally predicting better than self-reported clinical measures (unpaired t-test, p < 0.001). Testing the hypothesis that networks supporting cognition should be apparent in networks related to symptomatology, we examined the networks' overlap. The overlap was sparse, but primarily identified the thalamus, cerebellum, somatomotor networks, and the dorsolateral prefrontal cortex as key hubs in mediating relationships between clinical and cognitive measures. The findings reveal the extent to which external measures reflect underlying brain networks and highlight that examining network overlap can identify networks specific to clinically relevant cognitive dysfunction.