Computational time for running gene-ε and other gene-level association approaches, as a function of the total number genes analyzed and the number of SNPs within each gene.

Methods compared include: gene-ε, PEGASUS [12], VEGAS [7], RSS [14], MAGMA [10], and SKAT [20]. Here, we simulated 10 datasets for each pair of parameter values (number of genes analyzed, and number of SNPs within each gene). Each table entry represents the average computation time (in seconds) it takes each approach to analyze a dataset of the size indicated. Run times were measured on a MacBook Pro (Processor: 3.1-gigahertz (GHz) Intel Core i5, Memory: 8GB 2133-megahertz (MHz) LPDDR3). Only a single core on the machine was used. PEGASUS, SKAT, and MAGMA are score-based methods and, thus, are expected to take the least amount of time to run. Both gene-ε and RSS are regression-based methods, but gene-ε is scalable in both the number of genes and the number of SNPs per gene. The increased computational burden of RSS results from its need to do Bayesian posterior inference; however, gene-ε is able to scale because it leverages regularization and point estimation for hypothesis testing.