Abstract
All summary statistics-based methods to estimate the heritability of SNPs (hg2) rely on accurate linkage disequilibrium (LD) calculations. In admixed populations, such as African Americans and Latinos, LD estimates are influenced by admixture and can result in biased hg2 estimates. Here, we introduce covariate-adjusted LD score regression (cov-LDSC), a method to provide robust hg2 estimates from GWAS summary statistics and in-sample LD estimates in admixed populations. In simulations, we observed that unadjusted LDSC underestimates hg2 by 10%-60%; in contrast, cov-LDSC is robust to all simulation parameters. We applied cov-LDSC to approximately 170,000 Latino, 47,000 African American 135,000 European individuals in three quantitative and five dichotomous phenotypes. Our results show that most traits have high concordance of hg2 between ethnic groups; for example in the 23andMe cohort, estimates of hg2 for BMI are 0.22 ± 0.01, 0.23 ± 0.03 and 0.22 ± 0.01 in Latino, African American and European populations respectively. However, for age at menarche, we observe population specific heritability differences with estimates of hg2 of 0.10 ± 0.03, 0.33 ± 0.13 and 0.19 ± 0.01 in Latino, African American and European populations respectively.