Genome-wide association studies (GWASs) are conducted to identify replicable genomic risk loci associated with diseases and traits (Uffelmann et al., 2021). Polygenic indexes (PGIs; also called polygenic scores or PRS) are genetic predictors of phenotypes computed using GWAS-generated summary statistics for the corresponding phenotype. The recent increase in the sample size of GWASs has helped improve the predictive performance of PGIs (Duncan et al., 2019; Wang et al., 2020). However, this improvement is unbalanced across ancestral groups due to the over-representation of European ancestry individuals in most GWASs (80 percent of all GWAS participants are of European descent despite making up only 16 percent of the global population) (Martin et al., 2019; Priv et al., 2022; Wang et al., 2020). Relative to individuals of European ancestry, the predictive accuracy of PGIs falls by an average of 37, 50, and 78 percent in individuals of South-Asian, East-Asian, and African ancestries, respectively (Martin et al., 2019). However, most studies that examine the relative predictive accuracy of PGIs focus on a few traits primarily biological phenotypes such as body mass index, blood pressure, and hemoglobin levels. This paper examines the reduction in predictive accuracy for behavioral traits such as educational attainment, substance use, and neuropsychiatric conditions. For such traits, we may expect the relative predictive accuracy to be lower, given the complex environmental pathways through which these PGIs operate. Furthermore, we estimate what share of the PGI accuracy loss is explained by differences in minor allele frequencies (MAF), linkage disequilibrium, and gene-environment interaction effects.
Robel Alemu, University of California Los Angeles
Patrick Turley, University of Southern California
Aysu Okbay, Vrije Universiteit
Daniel Benjamin, University of California Los Angeles