Multiethnic polygenic risk scores improve risk prediction in diverse populations

Abstract

Methods for genetic risk prediction have been widely investigated in recent years. However, most available training data involves European samples, and it is currently unclear how to accurately predict disease risk in other populations. Previous studies have used either training data from European samples in large sample size or training data from the target population in small sample size, but not both. Here, we introduce a multi-ethnic polygenic risk score that combines training data from European samples and training data from the target population. We applied this approach to predict type 2 diabetes (T2D) in a Latino cohort using both publicly available European summary statistics in large sample size (Neff=40k) and Latino training data in small sample size (Neff=8k). Here, we attained a >70% relative improvement in prediction accuracy (from R2=0.027 to R2=0.047) compared to methods that use only one source of training data, consistent with large relative improvements in simulations. We observed a systematically lower load of T2D risk alleles in Latino individuals with more European ancestry, which could be explained by polygenic selection in ancestral European and/or Native American populations. We predict T2D in a South Asian UK Biobank cohort using European (Neff=40k) and South Asian (Neff=16k) training data and attained a >70% relative improvement in prediction accuracy, and application to predict height in an African UK Biobank cohort using European (N=113k) and African (N=2k) training data attained a 30% relative improvement. Our work reduces the gap in polygenic risk prediction accuracy between European and non-European target populations.

Genome‐wide association study identifies loci associated with liability to alcohol and drug dependence that is associated with variability in reward‐related ventral striatum activity in African‐ and European‐Americans

“…Studies suggest that even when discovery samples of non-European individuals are small, including them in individual discovery analyses and trans-ancestral analyses can result in novel insights into the genetic architecture of the disorder and in polygenic prediction. 12,19,20 Differences in prevalence and patterns of substance dependence across ancestrally diverse groups in the United States 21 underscore the importance of conducting GWAS on these phenotypes in these groups. In particular, the study of African-Americans, one the largest minorities represented in GWAS data in the United States, provides an opportunity to address this notable disparity in genomic research.…”

On prs for complex polygenic trait prediction

“…The purpose of PRS is to aggregate effects from a large number of causal SNPs, each of which has small contribution to the phenotype. However, recent years have witnessed widespread not so satisfying performances of PRS in real data applications (Bogdan, Baranger and Agrawal, 2018;Zheutlin and Ross, 2018;Márquez-Luna, Loh and Price, 2017;Torkamani, Wineinger and Topol, 2018;Clarke et al, 2016;Mistry et al, 2018a,b;Socrates et al, 2017). Though previous studies (Daetwyler, Villanueva and Woolliams, 2008;Dudbridge, 2013;Chatterjee et al, 2013) have found that the prediction accuracy of PRS is related to (n, p), the asymptotic properties of PRS is largely unknown.…”

“…There were over 3,000 PRS-related publications in 2018. However, the prediction power of PRS remains disappointingly low with little clinical utility, even for traits with known high heritability (Zheutlin and Ross, 2018;Márquez-Luna, Loh and Price, 2017;Torkamani, Wineinger and Topol, 2018). Two legitimate reasons for the poor performance of PRS include 1) low quality SNP arrays with low coverage of causal SNPs; and 2) low quality top-ranked SNPs in tagging causal SNPs (Chatterjee, Shi and García-Closas, 2016;Wray et al, 2013).…”

Localizing components of shared transethnic genetic architecture of complex traits from GWAS summary data

“…Indeed, for certain post- GWAS 28 analyses such as disease mapping 23,31,36 and statistical fine-mapping 28,[37][38][39][40] , under the assumption that two 29 populations share one or more causal variants, population-specific LD patterns can be leveraged to improve 30 performance over approaches that model a single population. On the other hand, several studies have shown 31 that heterogeneity in genetic architectures limits transferability of polygenic risk scores (PRS) across 32 populations 5, [41][42][43][44][45][46][47][48] ; critically, if applied in a clinical setting, existing PRS may exacerbate health disparities 33 among ethnic groups 49 . The population-specificity of existing PRS as well as estimates of transethnic 34 genetic correlations less than one reported in the literature 30,[50][51][52][53] indicate that (1) LD tagging and allele 35 frequencies of shared causal variants vary across populations, (2) that a sizeable number of causal variants 36 are population-specific, and/or (3) that causal effect sizes vary across populations due to, for example, 37 different gene-environment interactions.…”

“…transferability of PRS across populations 19,41,42,45,46 ) and uncover novel disease etiologies . 43 In this work, we introduce PESCA (Population-spEcific/Shared Causal vAriants), an approach that 44 requires only GWAS summary association statistics and ancestry-matched estimates of LD to infer genome-45 wide proportions of population-specific and shared causal variants for a single trait in two populations.…”