As shown in Table VI, after Bonferroni correction, no SNP pair was significantly associated with any of the traits in either AAs or EAs. However, this conservative correction does not account for the correlation between SNP pairs, and likely resulted in an inflation of type II error. In permutation tests to determine an empirical significance threshold, >106 pairs were tested for each randomly permuted phenotype based on the case–control ratios for the three traits: DSM-IV CD, Groups 4 and 5. We parallelized the standard permutation approach based on logistic regression (i.e., using PLINK) and distributed the process to 120 processors to identify empirical thresholds for significance. Table VI lists the SNP pairs with Pgee < 1E–05 (GEE), Plr < 1E–05 (logistic regression using PLINK [Purcell, 2009]), or Pboost< 1E–04 (BOOST) for at least one of the traits, which identifies candidates for possible subsequent validation. Five SNP pairs were identified in AAs, including one that was strongly associated with DSM-IV CD and Group 4 (heavy cocaine use with infrequent intravenous injection), one associated only with Group 4 and three that were
