When was calculated for all 966 nuclear-encoded mitochondrial genes using the T2D DIAGRAM+ GWA meta-analysis, we found that their scores were on average less significant than random (Figure S1). Observing that the mitochondrial genes are smaller on average than all other genes in the genome (Table S1), we next examined the effect of confounders on the most significant SNP p-value per gene, . Towards this goal, we generated 1,000 null distributions of gene scores, through phenotype permutations of the Diabetes Genetics Initiative (DGI) GWA study, for which we have access to genotype data (see Materials and Methods). In these randomized data sets no genome-wide significant associations are expected. We observed significant correlations of the scores for each gene across permutations (mean Pearson's correlation coefficient across all genes for pairs of permutations: r = 0.36). This suggested that inherent properties of genes (not specific to the phenotype studied) confound gene scores.
