In simulation 4 which has causal gene-gene interaction without main effects, IGENT_exhaust performed better than or equivalent to BOOST in most simulation models.
In the Lesnick et al. [ 2007] study, the nominal axon guidance pathway association was computed from a set of SNPs thresholded for main effects with P values better than 0.05, but the significance of the association was tested by sampling from a pool of SNPs that was not thresholded for main effects.
When we used in the MLM model the structure based on the STS detected on chromosome 2 (all or a subset excluding the positions with the main effects), the correction was much better, and many associations were no more significant (Table 2), confirming that many associations could be due to the structure.
By adopting a liberal bias in the detection and treatment of sex differences, the noise introduced by sex effects is reduced, allowing better detection of main effects.
The interaction model (Model 2d) did not fit the data better than the main effects model (Model 1d) for BMI (Likelihood ratio test, p = 0.27).
These designs allow the efficient estimation of main effects and first-order interactions and perform better than alternative designs available in the literature.
Compared with the main effects model, the reduced model has a better fit (Pseudo R = 0.28), with the main effects of a similar sign and significant.
Part of the motivation for developing AI was our intuition that, in a scenario where the 2 df main-effects test is more powerful than than the single-SNP test, then model 1 with two main effects should, by definition, fit these data better than model 2 with one main effect (at the partner SNP) only.
PLSR in general has better estimation performance, especially for main effects, because all candidate predictors are considered in the model, while in SPCA individual exposures with non-zero main effects could be eliminated at the pre-screen because their univariate associations with the outcome are weak in the presence of non-zero interaction effects.
We conclude that Rosenfield and colleagues' [ 37] theory is better suited to explain the main effects of gender on psychopathology, rather than the child gender-moderated associations with parenting.
