Gene–gene interactions of SNPs in the CHRNA5/A3/B4 cluster were analyzed using a newly developed pedigree-based generalized multifactor dimensionality reduction (PGMDR) method (Lou and others 2008). As with the previous association analyses, age and sex were included as covariates for the EA and AA samples and age, sex, and ethnicity as covariates in the pooled sample. Gene–gene interactions were examined for all two- to seven-locus models. The top-ranked interaction model was chosen for a given order interaction, and its P value of prediction accuracy (PA) was evaluated by a permutation test based on 10,000 shuffles of the adjusted phenotypic values. Because all P values reported here were based on permutation tests of each given interaction model, no correction was needed. On the other hand, we considered a P value of 0.0083 (i.e., 0.05/6) as an experiment-wide corrected one at the 0.05 significance level given that our gene–gene interaction analyses considered six interactive models for each sample. For detailed information on the PGMDR approach, please refer to the paper by Lou et al. (2008).
