BACKGROUND AND AIMS: Polygenic risk scores (PRS) may soon be used to predict inflammatory bowel disease (IBD) risk in prevention efforts. We leveraged exome-sequence and SNP array data from 29,358 individuals in the multi-ethnic, randomly-ascertained health system-based BioMe biobank to define effects of common and rare IBD variants on disease prediction and pathophysiology. METHODS: PRS were calculated from European, African-American, and Ashkenazi Jewish (AJ) reference case-control studies, and a meta-GWAS run using all three association datasets. PRS were then combined using regression to assess which combination of scores best predicted IBD status in European, AJ, Hispanic, and African American cohorts in BioMe. Additionally, rare variants were assessed in genes associated with very early onset IBD (VEO-IBD), by estimating genetic penetrance in each BioMe population. RESULTS: Combining risk scores based on association data from distinct ancestral populations improved IBD prediction for every population in BioMe and significantly improved prediction among European ancestry UK Biobank individuals. Lower predictive power for non-Europeans was observed, reflecting in part substantially lower African IBD case-control reference sizes. We replicated associations for two VEO-IBD genes, ADAM17 and LRBA, with high dominant model penetrance in BioMe. Autosomal recessive LRBA risk alleles are associated with severe, early-onset autoimmunity; we show that heterozygous carriage of an African-predominant LRBA protein-altering allele is associated with significantly decreased LRBA and CTLA-4 expression with T cell activation. CONCLUSIONS: Greater genetic diversity in African populations improves prediction across populations, and generalizes some VEO-IBD genes. Increasing African-American IBD case-collections should be prioritized to reduce health disparities and enhance pathophysiologic insight.