A Bayesian Regularized and Annotation-Informed Integrative Analysis of Cognition (BRAINIAC)

We present the novel Bayesian Regularized and Annotation-Informed Integrative Analysis of Cognition (BRAINIAC) model. BRAINIAC allows for estimation of total variance explained by all features for a given cognitive phenotype, as well as a principled assessment of the impact of annotations on relative enrichment of predictive features compared to others in terms of variance explained, without relying on a potentially unrealistic assumption of sparsity of brain-behavior associations. We validate BRAINIAC in Monte Carlo simulation studies. In real data analyses, we train the BRAINIAC model on resting state functional magnetic resonance imaging (rsMRI) and neuropsychiatric data from the Adolescent Brain Cognitive Development (ABCD) Study and use the trained model in an out-of-study application to harmonized resting-state data from the Human Connectome Project Development (HCP-D), demonstrating a substantial improvement in out-of-study predictive power by incorporating relevant annotations into the BRAINIAC model.