Suffix trees and suffix arrays have been used widely in bioinformatics applications, especially for DNA sequence alignments in the initial exact match phase of heuristic algorithms. In this paper, a new GPU implementation and optimization of the suffix tree and suffix array on both multi-core and many-core platforms to accelerate multiple genome sequence searching is presented. The comparative performance evaluation between the suffix tree and suffix array is then carried out. The results showed that the suffix array needed only 20–30% of memory space compared with the suffix tree, and that the mean search time of the suffix array was significantly shorter than the mean search time of the suffix tree because of the use of a binary search with coalesced memory access and tile optimization under the GPU architecture. Moreover, the GPU implementation of the suffix array gained a speedup of approximate 99 times compared with the corresponding CPU serial implementation. This study showed that the massively parallel sequence matching algorithm based on suffix array was an efficient approach with the high-performance in the process of multiple DNA sequence matching.
