To correct for stratification, we modified the script that implements association tests (see Web resources) to employ a permutation scheme in which case/control status was permuted within each population (HIV and SCZ), assuming known population labels. This permutation scheme does not change the computational cost of the study. The results show that the permutation procedure adequately controlled for population stratification, removing the excess of significant signals. For example, for T5, the most significant p-value after correction was 0.0001 and the proportions of p-values were 0.0340 at level 0.05, and 0.0060 at level 0.01. As mentioned previously, the deficiency of statistically significant signals is due to low counts and is consistent with the null distribution (Figure 2). Results were similar for the other statistical tests, and for other proportions of HIV samples assigned as cases. These results show that the permutation-based correction was effective at controlling Type-I errors.
