Multifactor dimensionality reduction (MDR) has been successfully applied to identification of gene-gene interactions for the complex traits. Generalized MDR (GMDR) was its extension that allows adjustment for covariates. The current GMDR software mainly focuses on candidate gene association studies with a relatively small number of genetic markers and has some limitations to be extended to genome-wide association studies (GWAS) with a large number of genetic markers. GWAS-GMDR is an effective parallel computing program package with special features for GWAS with a large number of genetic markers by using distributed job scheduling method and/or CUDA-enabled high-performance graphic processing units (GPU).

1. First, it implemented an effective memory handling algorithm and efficient procedures for GMDR to make joint analysis of multiple genes feasible for GWAS.
   
   
2. Second, a weighted version of cross-validation consistency based on 'top-K selection' (WCVC^K) was proposed to report multiple candidates for causal gene-gene interactions.
   
   
3. Third, various performance measures were implemented to evaluate MDR classifiers, including balanced accuracy, tau-b, likelihood ratio and normalized mutual information.
   
   
4. Fouth, three popular methods for handling missing genotypes were implemented: complete, available and missing category.
   
   
5. Finally, our applications support both CPU-based and GPU-based parallel computing system.