ARID1B damaging variants from more than one million genomes, cause human diseases by impairing protein-protein interactions, stability, and regulation

ARID1B is the most frequently de novo altered gene across a spectrum of human neurodevelopmental disorders and cancers. We found 1,456 missense variants in ARID1B's C-terminal domains, 94% of which are clinically uninterpreted and 59% of which are observed in human disorders and cancers. Using integrative modeling of ARID1B cBAF and DNA bound structures, we calculate how these variants impact key functional sites, structural stability, and topological features, finding that 617 variants clearly impair ARID1B's interactions within BAF, interactions with DNA, fold stability, or post-translational regulation. Our study enables scalable and precise variant interpretation by illuminating the molecular mechanisms by which missense variants damage ARID1B function, with commonalities across malignancies and congenital anomalies mutually informing therapeutic developments.