Our simulations identified scenarios under which analysis of GWA SNP data at the level of gene sets adds power in detecting associations of small effect (e.g. odds ratio of 1.07 for an allele frequency of 0.2–0.3 and sample size of 10,000 individuals [41]) compared to single SNP analysis. For example, consider a total of 100 causal genes in the genome each with an effect size, sufficient to provide 1% power of detecting an association at the individual SNP level at genome-wide significance. In this setting, MAGENTA has 50% power of detecting enrichment if a given set of 1,000 genes (e.g. nuclear-encoded mitochondrial genes) contains ∼3% or 30 genes with a modest effect, when 100 genes (e.g. OXPHOS genes) contain ∼10% or 10 genes with a modest effect, or when 25 genes (e.g. on the order of the number of nuclear regulators of mitochondrial genes) contain ∼25% or 6 genes with a modest effect.
