Golan et al. (1) report that restricted maximum likelihood (REML) seriously underestimates SNP heritability when using a case-control design. Their conclusions are based on results from simplified linkage equilibrium SNP simulation (SLES), which the authors acknowledge may be unrealistic. We simulated case-control data using the liability threshold model (1, 2), based on a real genome-wide association study (GWAS) of 800,000 SNPs from 64,000 samples, i.e., a genome-wide linkage disequilibrium SNP simulation (GLDS). Our simulation used a population disease risk of K = 0.01 and proportion of cases in the sample of P = 0.5 (therefore, there were 640 cases and 640 controls in the estimation analyses). |
|