GWGGI: software for genome-wide gene-gene interaction analysis

Background: While the importance of gene-gene interactions in human diseases has been well recognized, identifying them has been a great challenge, especially through association studies with millions of genetic markers and thousands of individuals. Computationally efficient and powerful tools are in great need for the identification of new gene-gene interactions in high-dimensional association studies. Result: We develop C++ software for genome-wide gene-gene interaction analyses (GWGGI). GWGGI utilizes tree-based algorithms to search a large number of genetic markers for a disease-associated joint association with the consideration of high-order interactions, and then uses non-parametric statistics to test the joint association. The package includes two functions, likelihood ratio Mann-whitney (LRMW) and Tree Assembling Mann-whitney (TAMW).We optimize the data storage and computational efficiency of the software, making it feasible to run the genome-wide analysis on a personal computer. The use of GWGGI was demonstrated by using two real data-sets with nearly 500 k genetic markers. Conclusion: Through the empirical study, we demonstrated that the genome-wide gene-gene interaction analysis using GWGGI could be accomplished within a reasonable time on a personal computer (i.e., ~3.5 hours for LRMW and ~10 hours for TAMW). We also showed that LRMW was suitable to detect interaction among a small number of genetic variants with moderate-to-strong marginal effect, while TAMW was useful to detect interaction among a larger number of low-marginal-effect genetic variants.

💡 Research Summary

The paper introduces GWGGI, a C++‑based software package designed to detect gene‑gene interactions (epistasis) in genome‑wide association studies (GWAS) that involve millions of single‑nucleotide polymorphisms (SNPs) and thousands of individuals. Recognizing that traditional epistasis detection methods become computationally prohibitive at this scale, the authors develop two complementary algorithms: Likelihood Ratio Mann‑Whitney (LRMW) and Tree Assembling Mann‑Whitney (TAMW). Both methods rely on tree‑based searching to partition the data and on the non‑parametric Mann‑Whitney U statistic to evaluate the joint association between case‑control status and combinations of genetic markers.

LRMW is optimized for scenarios where a relatively small set of variants exhibits moderate to strong marginal effects. It builds a single decision tree, computes a likelihood‑ratio adjusted Mann‑Whitney statistic at each split, and thus achieves higher power for detecting strong, low‑order interactions. TAMW, by contrast, constructs multiple independent trees, aggregates their Mann‑Whitney scores, and is therefore more sensitive to high‑order interactions among many variants with weak marginal effects.

From an implementation standpoint, GWGGI minimizes memory consumption by storing genotype data in a 2‑bit compressed format and decoding only when necessary. Parallelism is achieved through OpenMP, allowing the software to exploit multi‑core CPUs efficiently. These optimizations enable the analysis of a dataset containing roughly 500,000 SNPs and 5,000 subjects on a standard personal computer in about 3.5 hours for LRMW and 10 hours for TAMW—substantially faster than existing tools such as PLINK’s epistasis module or MDR, which often require days of computation on comparable hardware.

The authors validate GWGGI using two real GWAS datasets. In the first, a cardiovascular disease cohort, LRMW successfully recapitulated previously reported SNP‑SNP interactions, demonstrating its ability to detect strong, low‑order epistasis. In the second, a complex disease dataset with many low‑effect variants, TAMW uncovered a novel network of interacting markers that were not identified by conventional single‑marker analyses. These empirical results illustrate that GWGGI can adapt to different genetic architectures: LRMW excels when a few loci drive the signal, while TAMW shines when the signal is distributed across many loci.

The discussion acknowledges limitations. Both algorithms are currently tailored to binary phenotypes; extending them to quantitative traits or multi‑trait analyses would require methodological extensions. Moreover, the tree‑building process can be influenced by the initial split criteria, suggesting that incorporating prior biological knowledge (e.g., functional annotations or pathway information) could improve search efficiency and interpretability. Future directions proposed include GPU acceleration for even faster tree construction, integration with Bayesian frameworks to incorporate prior probabilities, and development of multivariate Mann‑Whitney extensions for pleiotropic studies.

In summary, GWGGI provides a practical, high‑performance solution for genome‑wide epistasis detection that can be run on modest hardware without sacrificing statistical power. By offering two distinct analytical modes—LRMW for strong, low‑order interactions and TAMW for diffuse, high‑order interactions—the package broadens the toolkit available to geneticists seeking to unravel the complex interplay of genetic variants underlying human disease.