A GPU-based approach for detecting genome-wide SNP-SNP interactions of quantitative trait in ADNI cohorts

Genome-wide association studies (GWAS) have proven that thousands of single nucleotide polymorphisms (SNP) are closely related to complex disease. As a neurodegenerative complex disease, Alzheimer’s disease (AD) still is a topic under active research in recent GWAS. Studies have shown that SNPSNP interactions (a.k.a. epistasis) can be used to explain genetic variation of AD. However, most existing epistasis detection methods are difficult to perform an exhaustive search, because of the huge computational challenges. Moreover, the possible SNPSNP interactions of low main effects are largely ignored in common GWAS of AD. In this context, potential risk variants at AD loci identified by GWAS only explain part of the heritability. In particular, the developed epistasis detection tools are mostly designed for case-control studies, whereas the available tools and approaches for quantitative trait (Qt) studies are quite limited. To overcome this problem, we propose an approach named GEEpiQt (GPU-based exhaustive epistasis detection of Qt). With GEEpiQt, 159 billion SNPs pairs were detected based on GPUs, and a total of 318 billion regression calculations were performed. All these calculations took about 26 hours to implement an exhaustive search across the whole genome. In terms of the results we have got, a total of 14,016 SNPs pairs showed statistical significance. Among the 14,016 pairs of SNPs, 467 pairs meeting the cell-size criterion. By analyzing with Enrichr disease database, 188 genes were found associated with AD, and, in particular, 91 novel genes were identified worth in-depth study. In summary, the proposed approach is an exhaustive epistasis research without missing any signals across the whole genome, which has a better detection power outperform other competitive approaches. The new suspected interactions discovered by GEEpiQt are helpful for the AD diagnosis and prevention. Therefore, there is much room for exploration in exhaustive SNP-SNP interactions detection based on GPUs in a genome-wide scale in the future.