In the Repository cohorts, before constructing PCs, we removed markers with imputation accuracy less than 70% or minor allele frequency less than 1%, as well as markers in long-range LD blocks (chr5:44mb-51.5mb, chr6:25mb-33.5mb, chr8:8mb-12mb, chr11:45mb-57mb). Next, we restricted the sample to individuals with European ancestries, as described immediately above. We further pruned the markers to obtain a set of approximately independent markers, using a 1Mb rolling window (incremented in steps of 5 variants) and an R2 threshold of 0.1. We used this set of markers to estimate a genetic relatedness matrix. We identified all pairs of individuals with a relatedness coefficient greater than 0.05 as calculated by Plink1.979. We excluded one individual from each pair, calculated the first 20 PCs for the resulting sample of unrelated individuals using Plink 1.9, and projected the PCs onto the sample of unrelated individuals. In HRS, we re-labeled the PCs in sets of five in order to address identifiability concerns.
