Last corrected 4/18/13
The following code performs the two-step GxG test described in Lewinger, Morrison et al article “Efficient Two-Step Testing of Gene-Gene Interactions in Genome-Wide Association Studies” Genetic Epidemiology, In Press 2013.
The code needs to be compiled and run on a high performance computer (HPC). The programs are written to use the MPI which must be installed on the HPCC. A run of 1500 subjects with 500,000 SNPs required 13 days to run using 128 processors.
The first program to run is mpiSNP. This does the stratified Mantel-Haenszel test for an association for all pairs of SNPs described in the paper.
The second program, SNPLogReg3, does logistic regression on disease status for every SNP individually. It also does logistic regression on disease status for SNP x SNP interactions on a specified subset of SNP pairs. This subset would tipically consist of the all possible SNP pairs among those showing evidence of a direct SNP effect, or the set of SNP pairs that pass the Mantel-Haenszel screening step.
The program is currently tailored for the specific dataset described in the paper using hard-coded number of subjects and SNPs. In order to run it with your data you need to modify it appropriately. We are in the process of doing the same analysis on another dataset and are rewriting the code to allow the user to more easily specify the data to use and to use the plink format for the data. We plan to post the updated program in the near future.