Large-scale genome-wide association studies on a GPU cluster using a CUDA-accelerated PGAS programming model

Detecting epistasis, such as 2-SNP interactions, in genome-wide association studies (GWAS) is an important but time consuming operation. Consequently, GPUs have already been used to accelerate these studies, reducing the runtime for moderately-sized datasets to less than 1 hour. However, single-GPU approaches cannot perform large-scale GWAS in reasonable time. In this work we present multiEpistSearch, a tool to detect epistasis that works on GPU clusters. While CUDA is used for parallelization within each GPU, the workload distribution among GPUs is performed with Unified Parallel C+ + (UPC+ +), a novel extension of C+ + that follows the Partitioned Global Address Space (PGAS) model. multiEpistSearch is able to analyze large-scale datasets with 5 million SNPs from 10,000 individuals in less than 3 hours using 24 NVIDIA GTX Titans.