The simulations of Dickson et al. (Box 1) are parameterised in terms of the GRR (genotype relative risk, Box 3) of the causal variants. They consider a range of effect sizes (GRR 2–6) for each of the causal variants and GWAS case-control samples of 2,000 to 6,000. Although such simulations allow evaluation of the ability to detect synthetic associations and allow extrapolation to smaller effect sizes and larger sample sizes, the results generated from the selected parameters do have an impact on the perception of the importance of synthetic associations. To interpret the genetic architecture of their model we calculated the variance explained under the liability threshold model (see Box 3), assuming that each causal variant has frequency 0.01 (the average of the range considered, which will make our results conservative since the distribution of variance explained for causal variants of frequency 0.005–0.02 is not symmetric). When baseline disease probability was 0.01, a single GRR of 2 corresponds to a variant explaining 0.15% of the variance in liability and a GRR of 6 corresponds to 1.2% of the variance in
