Our second ranking of signals, based on increased explained variance under an RPM analysis, was aimed at identifying pairs with joint effects much larger than the sum of the individual effects. We note that this is a filtered ranking because of our requirement that any reported pair-wise signal explain at least 1% of the phenotypic variance. The RPM and logistic regression results for this set of pairs are summarized in Table 4. Once again, all of the pair-wise RPM models had empirical p values of no more than 10–6 (based on results from 1,000,000 permutations each). Because this ranking was designed to find pairs with large synergistic effects, it is perhaps not surprising that the logistic regression analysis found the interaction terms for each of these signals to be nominally significant, with uncorrected p values ranging from 0.04 to 2.2 × 10–5. None of these pairs were testable in the independent ACS data because in each case, at least one of the SNPs was not genotyped in the ACS data and no good surrogate could be identified.
