The accelerating implementation of BLAST with stream processor

Sequence alignment is one of the most fundamental and important operation in bioinformatics. Through sequence alignment, we can find the sequence´s information of function, structure and evolution. BLAST is one of the most popular algorithms in the field of sequence alignment. In this paper, we have designed a GPU-based parallel BLAST algorithm and implemented it on the brook⁺ platform. The main task is to parallel the construction of words lists procedure and the match-expansion procedure. As for match-expansion procedure, we designed two ways to parallelize it including internal parallelism and mixed parallelism. According to results of experiments based on AMD´s HD4850, we got more than 3x gains on parallel implementing the procedure of construction of words list and the implementation of internal parallelism respectively compared with CPU-based implementations. Besides, we got more than 4x gains for mixed parallelism as well. Furthermore, the experiments verified that the procedure of reading and writing data through PCI-E was the bottleneck of AMD´s GPU applications.