Iterative Usage of Fixed and Random Effect Models for Powerful and Efficient Genome-Wide Association Studies
Three statistical models were performed by the five packages: 1) GLM by PLINK; 2) MLMs by EMMAX, GenABEL, and MLMM; and 3) FarmCPU by FarmCPU. Computing time (a) and memory usage (b) in response to sample size are displayed. The analyses were performed on a laptop (Asus A53S) running a Linux system (Ubuntu 12.10, 64 bit) with a 4.0 Gb of Random-Access Memory (RAM) and an Inter duo Core i3-2310M processor at 2.1 GHz. One core was used for this test. All datasets had 60,000 markers, but response was measured as a function of sample size. The last data point indicates the maximum sample size each software package could process without freezing the computer, except for PLINK and FarmCPU. The limitations for these two software packages were not reached with the maximum sample size examined.