To detect association, we used the highest measure of signal across the entire genomic region as the test statistic for each association- mapping method. Test statistics were the maximum -\(\log_{10}\)(p-value) from the Fisher’s exact test, VT test, and C-alpha test; the largest PIP for CAVIARBF; the largest VIP for elastic net; the largest probability score for BLOSSOC and the largest Mantel statistic value for the true trees. To obtain permutation pvalues summarizing the evidence for association, we compared the observed values of these test statistics against a null distribution obtained by randomly permuting the case-control labels on the subjects 1000 times.

To address the localization, we computed the average distance of the peak association signals from the risk region. ??In the simulation study, we examined the proportion of times this localization occurred. We also recorded the minimum distance, in kb, from the peak association to the risk region in the simulated datasets.