We used the Diabetes Genetics Initiative (DGI) GWA study [17] as a test case for developing MAGENTA, as we had access to genotypes of all individuals in this study (as opposed to the GWA meta-analyses analyzed in this paper where we do not have access to genotype data). The analysis was done only on the population-based samples of the DGI study - 1,022 cases and 1,075 controls that were matched for age, gender, body mass index and region of origin. Specifically, the T2D case/control labels were randomly permuted 1,000 times between individuals from the same collection center and the same gender. A genome-wide association test (logistic regression) that assumes an additive allelic model (1 degree of freedom) followed by a genomic control (adjustment for lambda larger than 1) was then applied to each of the 381,099 genotyped SNPs across the 1,000 permutations, resulting in an association p-value, for each SNP i and each permutation. was calculated for all genes in the genome, across the 1,000 DGI permutations. A gene p-value adjusted for confounding effects with permutation analysis, was then calculated
