or 0.5. We calibrated the prediction error by calculating the regression of Gjk on Ajk. We established empirically that the regression coefficient β=1−(c+1/N)var(Ajk) (Fig. 1), where N is the number of SNPs used to calculate Ajk and the term in c depends on the MAF of the causal variants (Online Methods). If the causal loci that have the same spectrum of allele frequency as the genotyped SNPs (θ = 0.5), then c = 0, and 1/N can be interpreted as the sampling error for estimating the relationship over the whole genome from N random SNPs. The parameter c is > 0 if θ < 0.5 because the relationship at causal variants with low MAF is typically less than the average relationship over the whole genome.
