Transdiagnostic connectome-based predictive modeling of many behavioral phenotypes reveals brain network mediators of clinical-cognitive relationships

Abstract A key assumption of the NIMH’s RDoC framework is that disordered circuits in the brain should manifest in observable behaviors, including psychiatric symptomatology and cognitive deficits. However, how disordered circuitry impacts multiple behaviors remains poorly understood. Connectome-based predictive modeling (CPM) applied to functional MRI connectivity data can identify networks associated with specific behavioral measures across individuals. Prediction strength reflects how closely a measure relates to network connectivity, while derived networks provide evidence of where an individual’s disordered circuits are located. Using CPM, we predicted a broad range of self-reported clinical and objective cognitive measures in a large, transdiagnostic sample with extensive fMRI data (n = 317). Prediction performance varied substantially across instruments, with objective cognitive tests yielding stronger models than self-reported clinical measures (p < 0.001). To test whether circuits underlying cognitive deficits related to symptomatology reside in regions where networks overlap, we examined the prediction strength of these sparsely shared circuits. Their connectivity strongly predicted cognitive performance and were primarily localized within the frontoparietal network and between the frontoparietal and default mode networks. These findings reveal how much various behavioral measures reflect brain networks and how circuits within the shared network space contribute to cognitive deficits associated with symptomatology. Competing Interest Statement G.S. has served as a consultant or scientific advisory board member to Axsome Therapeutics, Biogen, Biohaven Pharmaceuticals, Boehringer Ingelheim International, Bristol-Myers Squibb, Clexio, Cowen, Denovo Biopharma, ECR1, EMA Wellness, Engrail Therapeutics, Gilgamesh, Janssen, Levo, Lundbeck, Merck, Navitor Pharmaceuticals, Neurocrine, Novartis, Noven Pharmaceuticals, Perception Neuroscience, Praxis Therapeutics, Sage Pharmaceuticals, Seelos Pharmaceuticals, Vistagen Therapeutics and XW Labs; and received research contracts from Johnson & Johnson (Janssen), Merck and Usona. G.S. holds equity in Biohaven Pharmaceuticals and is a co-inventor on a US patent (8,778,979) held by Yale University and a co-inventor on US provisional patent application no. 047162-7177P1 (00754), filed on 20 August 2018 by Yale University Office of Cooperative Research. Yale University has a financial relationship with Janssen Pharmaceuticals and may receive financial benefits from this relationship. The University has put multiple measures in place to mitigate this institutional conflict of interest. Questions about the details of these measures should be directed to Yale University's Conflict of Interest office. V.H.S. has served as a scientific advisory board member to Takeda and Janssen. The remaining authors declare no competing interests. Footnotes Figure 3 was revised to include additional comparisons of distribution measures to prediction strength. Figures 6 and 7 were replaced entirely. Previously, those figures contained results from a series of mediation analyses. We replaced those figures to include data that more compellingly showed that the shared network subspace explains the variability in cognitive deficits that co-occur with psychiatric symptoms. Now, these figures and their corresponding analyses utilize predictive modeling on the shared features between clinical and cognitive networks in the training data to predict cognitive performance. The networks were localized such that circuits within the frontoparietal network predicted cognitive functioning, while circuits between DMN and frontoparietal networks predicted cognitive deficits. The language in the introduction and discussion has been substantially modified to highlight that this approach enhances the specificity of CPM to identify edges that predict multiple measures, and points to their role in explaining the variance in relationships between measures. https://github.com/YaleMRRC/YaleNeuroConnect_FullPredictions.git