Power Comparison under Simulation Model 1
Four different testing methods are examined: the proposed JLS-Fisher (red) and JLS-minP (purple) methods, the distribution test (blue) of Aschard et al.,16 and the LRT (black) of Cao et al.17 Phenotype values for 2,000 independent subjects were simulated under Model 1, E[Y] = βGG + βE1E1 + βGE1G∙E1, where the MAF of G was 0.3 and the exposure variable E1 was simulated as a Bernoulli variable with frequency = 0.3. The effect of the exposure E1, βE1, was fixed at 0.3 while the other effects varied.
(A–C) Results when the main genetic effect βG and the interaction effect βGE1 are in the same direction.
(D–F) Results when βG and βGE1 are in opposing direction.
Power was calculated at the 5 × 10−8 level based on 500 replicates. For other simulation details, see Appendix A. Additional power results are in Figures S3–S5.