For all phenotypes in Supplementary Table 5, we ran three separate GWASs, one for each partition. Briefly, each GWAS in UKB was conducted using mixed-linear models implemented by the software BOLT-LMM41. The dependent variable in each analysis is a phenotype that has been residualized on sex, a third-degree polynomial in birth year (defined as (birthyear − 1900)/100), their interactions, 106 genotyping batch dummies, and the first 40 of the PCs released by the UK Biobank. Details on how each phenotype is coded are provided in Supplementary Table 5. For the variance-component estimation in BOLT-LMM (but not the association analyses), we restricted the set of markers to the set of 622,788 hard-called SNP genotypes that remained after filtering for 1% minor allele frequency and 60% imputation accuracy and pruning with an r2 threshold of 0.3. Our subsequent association analyses were performed on imputed SNP dosages provided by UKB.
