There were a total of 8,001 SNP-pairs analyzed in the hypothesis generating dataset (COGEND). We present our top results for each of two rankings: the first based on total explained variance, the second based on increase in explained variance by the joint analysis of the pair compared to the sum of the univariate effects. Table 2 reports the top six results in terms of total variance explained. For each result, we list the chromosomes, gene clusters, and rs numbers for the two SNPs representing the signal. This information is followed by the percentage of the phenotypic variance accounted for by each of the SNPs under univariate RPM analyses and the variance explained under RPM analyses combining both SNPs in the model. All of the pair-wise RPM models in this table had empirical p values of ≤10–6 (based on results from 1,000,000 permutations each; 10 times as many as were used for the univariate estimates).
