Abstract
Complex traits differ in their genetic architectures, and these differences can affect polygenic score performance. Examining 177 complex traits from the UK Biobank, we first identified pairs of traits that have trait-associated SNPs in shared genomic regions. We then compared and contrasted three aspects of genetic architecture (SNP heritability, trait-specific recombination rates, and a novel metric of polygenicity) with three aspects of polygenic score performance (correlations between predicted and actual trait values, portability of genetic predictions, and divergence across populations). Although highly heritable traits tended to be easier to predict, heritability was largely uninformative with respect to the portability of genetic predictions. By contrast, there was a positive relationship between trait-specific recombination rates and the portability of genetic predictions. Analyzing 100kb bins, we used Gini coefficients to quantify the extent that SNP heritability is unequally distributed across the genome. Polygenic score performance was largely independent of Gini – traits with more Mendelian architectures need not be easier to predict. By contrast, Gini coefficients were negatively correlated with the prevalence of binary traits. We also found that binary traits were more difficult to predict than quantitative traits. Interestingly, lifestyle and psychological traits tend to have low heritability, low Gini coefficients, as well as poor predictability and portability across populations. Because of this, our results caution against the application of polygenic scores to traits like general happiness, alcohol frequency, and average income, especially when polygenic scores are applied to individuals who have an ancestry that differs from the original source population.
Competing Interest Statement
The authors have declared no competing interest.