Investigating the lack of transferability of polygenic risk scores in cohorts with admixed ancestry

Abstract |

Polygenic risk scores (PRS) can be used to summarize the results of genome-wide association studies (GWAS) into a single number representing the risk of disease. For some traits (for example, cardiovascular disease, breast cancer) PRS allows us to identify individuals with clinically actionable levels of risk in the tails of the PRS distribution. One barrier to the use of PRS in clinical practice is that the majority of GWAS come from cohorts of European ancestry, and predictive power is lower in non-European ancestry cohorts. There are many possible reasons for this decrease; here we investigate the performance of PRS in admixed cohorts to identify some of these reasons. We focus on the performance of PRS for height (a model polygenic trait) in cohorts with admixed African and European ancestry. Having multiple ancestry components in the same genome allows us to test for ancestry-related differences in PRS prediction while controlling for environmental differences. We first show that that the predictive power of height PRS increases linearly with European ancestry (partial R2ranges from 0.015-0.15 for 0-100% European ancestry). This effect is unaltered when we re-estimate effects-sizes using sibling pairs, ruling out residual population structure as an explanation. Second, we show that this pattern persists when PRS is computed using subsets of SNPs in regions of both high and low linkage disequilibrium (LD), indicating that differences in LD are not the only cause. Third, we show that frequency differences of associated variants between African and European ancestry backgrounds explain only up to 25% of the observed reduction in predictive power. Finally, we find that there is no association between ancestry and phenotypic variance, indicating that there is no relationship between ancestry and genetic variance, and that the reduction in PRS predictive power cannot be explained by causal variants that are specific to the African ancestry background. In conclusion, no single factor we investigated can explain the difference in predictive power across ancestries, hinting that other factors – for example heterogeneity in effect size – or a combination of multiple factors is responsible for this pattern. This study further highlights the need for more diversity in GWAS, as well as a better understanding of the complexities of variant discovery and portability across cohorts and ancestries."