Improving Polygenic Prediction in Ancestrally Diverse Populations

Abstract

Polygenic risk scores (PRS) have attenuated cross-population predictive performance. As existing genome-wide association studies (GWAS) were predominantly conducted in individuals of European descent, the limited transferability of PRS reduces its clinical value in non-European populations and may exacerbate healthcare disparities. Recent efforts to level ancestry imbalance in genomic research have expanded the scale of non-European GWAS, although they remain under-powered. Here we present a novel PRS construction method, PRS-CSx, which improves cross-population polygenic prediction by integrating GWAS summary statistics from multiple populations. PRS-CSx couples genetic effects across populations via a shared continuous shrinkage prior, enabling more accurate effect size estimation by sharing information between summary statistics and leveraging linkage disequilibrium (LD) diversity across discovery samples, while inheriting computational efficiency and robustness from PRS-CS. We show that PRS-CSx outperforms alternative methods across traits with a wide range of genetic architectures and cross-population genetic correlations in simulations, and substantially improves the prediction of quantitative traits and schizophrenia risk in non-European populations.

Multi-Ancestry Meta-Analysis yields novel genetic discoveries and ancestry-specific associations

“…Methods that generate biased estimates but less sampling varianceincluding standard, inverse-variance-weighted meta-analysis of cross-ancestry GWAS summary statistics, not taking into account LD and other population differences-can outperform MAMA in prediction accuracy. Exploring optimal ways to use GWAS summary statistics from European-ancestry populations to improve polygenic prediction in non-European-ancestry populations is an active and important area of research 27,28 .…”

Analysis across Taiwan Biobank, Biobank Japan and UK Biobank identifies hundreds of novel loci for 36 quantitative traits

“…For each biomarker (BMI, DBP, SBP, HDL-C, LDL-C, TG, and HbA1c), we collected the largest GWAS in EAS (the meta-analysis of TWB and BBJ) and EUR populations (Neale Lab UKBB GWAS for all biomarkers except BMI, for which GWAS summary statistics from the GIANT study 33 were used). Population-specific PRS for each biomarker was calculated using PRS-CSx 25 , a Bayesian polygenic prediction method that jointly models GWAS summary statistics from multiple populations to improve polygenic prediction. Specifically, for a fixed global shrinkage parameter (phi = 1e-6, 1e-4, 1e-2, and 1.0 in this study) that models the overall sparseness of the genetic architecture, PRS-CSx returned posterior SNP effect size estimates for each discovery population (i.e., EAS and EUR), which were used to calculate both an EAS-specific PRS and an EUR-specific PRS in the left-out TWB sample (N=10,285) that was unrelated to the discovery samples in TWB.…”

“…To assess the clinical utility of biomarker GWAS, we examined whether polygenic risk scores (PRS) of biomarkers can be used to predict the risk of common complex disease. We applied PRS-CSx 24,25 , a Bayesian polygenic prediction method, to integrate the GWAS summary statistics of EAS and EUR ancestry, and calculate both an EAS-specific and an EUR-specific PRS for each biomarker. We then predicted five complex diseases [obesity (Ncase = 824; defined as BMI >= 30), overweight (Ncase = 3,873; defined as BMI >= 25), hypertension (Ncase = 1,149), hyperlipidemia (Ncase = 771), and type 2 diabetes (Ncase = 508)], in a held-out sample of the TWB (N = 10,285; TWBv2 array), using a linear combination of PRS from one or more biomarkers (Figure 5), controlling for age, sex and top 20 principal components (PCs) of genotype data.…”

Quantitative Human Paleogenetics: What can Ancient DNA Tell us About Complex Trait Evolution?

“…This approach aims to improve PRS portability in present-day populations by reducing the fraction of spurious associations due to the cohort specific LD structure of the GWAS reference panel. Another promising approach is to jointly model PRS using GWAS summary statistics from multiple populations ( Márquez-Luna et al, 2017 ; Ruan et al, 2021 ; Turley et al, 2021 ). By including information from genetically distant groups, these methods can account for the variance in effect sizes inferred between GWAS cohorts.…”