Inclusion of variants discovered from diverse populations improves polygenic risk score transferability

Abstract

Summary The majority of polygenic risk scores (PRSs) have been developed and optimized in individuals of European ancestry and may have limited generalizability across other ancestral populations. Understanding aspects of PRSs that contribute to this issue and determining solutions is complicated by disease-specific genetic architecture and limited knowledge of sharing of causal variants and effect sizes across populations. Motivated by these challenges, we undertook a simulation study to assess the relationship between ancestry and the potential bias in PRSs developed in European ancestry populations. Our simulations show that the magnitude of this bias increases with increasing divergence from European ancestry, and this is attributed to population differences in linkage disequilibrium and allele frequencies of European-discovered variants, likely as a result of genetic drift. Importantly, we find that including into the PRS variants discovered in African ancestry individuals has the potential to achieve unbiased estimates of genetic risk across global populations and admixed individuals. We confirm our simulation findings in an analysis of hemoglobin A1c (HbA1c), asthma, and prostate cancer in the UK Biobank. Given the demonstrated improvement in PRS prediction accuracy, recruiting larger diverse cohorts will be crucial—and potentially even necessary—for enabling accurate and equitable genetic risk prediction across populations.

Admixed Populations Improve Power for Variant Discovery and Portability in Genome-Wide Association Studies

“…Our work suggests another advantage for conducting genetic studies in admixed populations, which comes from elevated allele frequencies when traits are moderately to highly differentiated. Moreover, discoveries from such studies aid improvement in cross-population PRS, which is critical in clinical prediction in personalized medicine yet presently has suboptimal performance for many biomedical traits in non-European populations ( Martin et al, 2019 ; Rosenberg et al, 2019 ; Cavazos and Witte, 2021 ). We therefore highlight that insights gained from admixed populations provide improved and appealing generalizable properties compared to homogeneous populations.…”

“…Despite the fast development and practicality of these methods, they have not often been applied to sample sizes of hundreds of thousands to millions because study design and data collection in mega-scale cohorts routinely prioritize recruitment of participants of single ancestry ( Atkinson et al, 2021 ). This greatly impedes downstream progress, such as polygenic risk score application across populations, where much lower accuracy is observed in non-European populations for many traits ( Duncan et al, 2019 ; Martin et al, 2019 ; Cavazos and Witte, 2021 ).…”

False discovery rate control in genome-wide association studies with population structure

“…By allowing a simultaneous analysis, we can increase power because different LD patterns uncover causal variants more effectively ( 48 ). Our discoveries may also be useful to better explain phenotypic variation in minority populations ( 73 , 74 ). Since the UK Biobank mostly comprises British individuals, the increase in power resulting from the analysis of other samples can only be relatively small.…”

The limits of personalization in precision medicine: Polygenic risk scores and racial categorization in a precision breast cancer screening trial

“…Another fundamental concern about polygenic risk scores is their reliance on repositories of SNPs that do not represent the full range of human genetic variation, raising the question: If clinical validity is established, for whom will it be valid [16]? Allele frequencies (or the incidence of a genetic variant in a population) vary by genetic ancestry, yet GWAS have predominantly included participants of European descent [17].…”