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.