Using computer simulations, we show that MAGENTA has considerable power (i.e. sensitivity) in detecting multiple modest effects relative to traditional single SNP analysis for a range of parameters. For example, for a gene set size of 100 genes, our method has 50% power of detecting enrichment when ∼10 genes have weak effects (that are equivalent to 1% detection power at single SNP level) versus 10% power of detecting only one of the 10 genes in single SNP analysis. By applying MAGENTA to GWA scan meta-analyses for LDL cholesterol, HDL cholesterol and triglyceride levels, we confirmed the method's ability to pick out relevant biological processes. We note that the nominal MAGENTA p-values for these positive controls were not exceedingly low (on the order of 10−2 to 10−6), emphasizing the limited power of the gene set approach. Our simulations allowed us to provide quantitative estimates of these limitations, and indications of possible limiting factors. For example, we found that power levels increase considerably with gene set size, fraction of causal genes in a gene set, and effect size of associated SNPs, and
