BioGPU: A High Computation Performance Tool for Genome-Wide Association Study

The performances of computing tools for genome-wide association study (GWAS) are measured by their computing speed, memory requirements and statistical power, which are determined by their statistical methods and how these methods engineered to make full use of computer hardware resources. We developed a computing tool named BioGPU that implemented a new statistical method, which also be refer to as BioGPU (Batched and Iterative Optimization for Genomic Probability Unification). The method effectively control false positive due to population structure and un-equal relatedness among individuals and improve statistical power even when compared to the mixed linear model methods. The BioGPU method has the much less complex computing time that is close to general linear model with linear computing time complexity. BioGPU was written by C computer language to maximize the capability of direct electronic circuit operations, including binary format of genotype input file and bit operation for matrix manipulations. For further speeding up, BioGPU was developed with consideration of parallel computation, and time consuming decreased linearly in the number of central processing unit threads. Meanwhile, The parallel component is so small that graphic processing units are also capable to perform the parallel computation. For solving the bottleneck of memory footprint, BioGPU allows users to control memory usage by themselves when big data is analyzed on computers with limited memory, it means that users has options to trade computing time for less memory usage. Based on these features above, BioGPU makes the analysis of big data set to be feasible without supercomputer.