In this paper, we propose and evaluate MSA-GPU, a solution to implement the exact Multiple Sequence Alignment algorithm in Graphics Processing Units (GPUs). In our solution, we use the Carrillo-Lipman upper and lower bounds to reduce the amount of computation. We propose a fine-grained strategy to explore the search space by using 2D projections. The results were obtained with a GTX 580 NVidia GPU comparing sets of 3 sequences (real and synthetic). We show that, for sequences with medium/low similarity, our GPU approach is able to outperform the MSA 2.0 CPU program, achieving a speedup of 8.6x.
展开▼