Comparative Evaluation of Cross-Ancestry Polygenic Risk Scoring of Type 1 Diabetes in the All of Us Cohort

Type 1 diabetes is a highly heritable autoimmune condition characterized by the destruction of pancreatic beta cells, resulting in insulin deficiency. Here, we developed a novel polygenic score we call the HLA-Augmented SBayesRC Framework (HLA-ARC). HLA-ARC integrates direct modeling of HLA haplotypes, with a Bayesian regression approach for the non-HLA component. SBayesRC leverages extensive functional genomic annotations and linkage disequilibrium patterns across approximately 7.4 million variants, substantially enhancing predictive accuracy. We systematically compared HLA-ARC to three existing T1D polygenic scores (Polygenic Risk Score extension for Diabetes Mellitus [PRSedm], Trans-Ancestry Polygenic Score for Diabetes [TA-PS], and Type 1 Diabetes Multi-Ancestry Polygenic Score [T1D-MAPS]) using data from the ancestrally-diverse All of Us cohort. Among the three existing methods, T1D-MAPS showed superior performance in all ancestry groups. However, HLA-ARC consistently outperformed the existing methods, achieving AUROC values exceeding 0.91 in European individuals and 0.89 in non-European groups. Our results demonstrate that integrating HLA haplotype modeling with genomic annotation and ancestry-informed linkage disequilibrium methods significantly improves polygenic risk prediction for autoimmune diseases characterized by major genetic risk loci.