74-OR: Improving Performance of Type 2 Diabetes Polygenic Scores in African and Hispanic Ancestries

Abstract

Most polygenic scores (PS) are developed based on data from persons of European (EUR) ancestry and perform poorly in non-EUR populations. To improve the performance of type 2 diabetes (T2D) PS in the under-studied African (AFR) and Hispanic (HIS) ancestries, we created more comprehensive reference panels using the “Tagging Iterative of single nucleotide variant in multiple populations” (TagIt) method and aggregated genome-wide association study (GWAS) meta-analyses from 5 major continental ancestries (360K cases, 1.8M controls, 32% of non-EUR ancestry). We used PRS-CS and PRS-CSx to develop ancestry-specific and multi-ancestry PSs with HapMap3 (HM3) and TagIt panels and evaluated their performance based on the variance explained by the liability R2 (lR2). Trained models were then tested in independent datasets. Compared to HM3 panels, TagIt panels for AFR and HIS ancestries had 4X improved coverage for variants with minor allele frequency 1-5% and showed significant improvement of ancestry-specific PSs (p of difference<0.01). Multi-ancestry PSs based on linkage disequilibrium (LD) and GWAS data from 5 ancestries outperformed the ancestry-specific PSs (lR2 AFR: 0.024 AFR-based PS vs. 0.038 multi-ancestry PS, p of difference = 1.3×10-3; lR2 HIS: 0.017 HIS-based PS vs. 0.061 multi-ancestry PS, p of difference = 4×10-8). It also outperformed the PS based on the EUR GWAS and LD data (lR2 AFR: 0.024 EUR-based PS, p of difference = 0.003; lR2 HIS: 0.045 EUR-based PS, p of difference = 0.003). TagIt reference panels did not show significant improved performance when applying multi-ancestry PS. By leveraging existing GWAS data from multiple ancestries and building more comprehensive reference panels we can improve the performance of the PS for diverse ancestries. However, increasing available GWAS data from diverse ancestries is still needed to improve performance in non-EUR populations and reduce health disparities. Disclosure A.Huerta: None. M.Vujkovic: None. B.F.Voight: None. X.Sim: None. R.J.Loos: None. J.E.Below: None. T.Ge: None. Va million veteran program: n/a. J.C.Florez: Consultant; AstraZeneca, Novo Nordisk, Other Relationship; AstraZeneca, Merck & Co., Inc. A.Manning: None. M.Ng: None. J.Kim: None. J.M.Mercader: None. I.Diabetes polygenic risk scores in multiple ancestr: n/a. R.Mandla: None. Y.Lu: None. L.E.Petty: None. K.Suzuki: None. S.S.Lee: None. H.Ng: None. M.Loh: None. Funding American Diabetes Association (11-22-ICTSPM-16 to J.M.M.); National Human Genome Research Institute (U01HG011723)