6533b86cfe1ef96bd12c8a52

RESEARCH PRODUCT

Faster GPU-Accelerated Smith-Waterman Algorithm with Alignment Backtracking for Short DNA Sequences

subject

description

In this paper, we present a GPU-accelerated Smith-Waterman (SW) algorithm with Alignment Backtracking, called GSWAB, for short DNA sequences. This algorithm performs all-to-all pairwise alignments and retrieves optimal local alignments on CUDA-enabled GPUs. To facilitate fast alignment backtracking, we have investigated a tile-based SW implementation using the CUDA programming model. This tiled computing pattern enables us to more deeply explore the powerful compute capability of GPUs. We have evaluated the performance of GSWAB on a Kepler-based GeForce GTX Titan graphics card. The results show that GSWAB can achieve a performance of up to 56.8 GCUPS on large-scale datasets. Furthermore, our algorithm yields a speedup of up to 53.4 and 10.9 over MSA-CUDA (the first stage) and gpu-pairAlign on the same hardware configurations.