PA-Star: A disk-assisted parallel A-Star strategy with locality-sensitive hash for multiple sequence alignment

Abstract

Multiple Sequence Alignment (MSA) is a basic operation in Bioinformatics, and is used to highlight the similarities among a set of sequences. The MSA problem was proven NP-Hard, thus requiring a high amount of memory and computing power. This problem can be modeled as a search for the path with minimum cost in a graph, and the A-Star algorithm has been adapted to solve it sequentially and in parallel. The design of a parallel version for MSA with A-Star is subject to challenges such as irregular dependency pattern and substantial memory requirements. In this paper, we propose PA-Star, a locality-sensitive multithreaded strategy based on A-Star, which computes optimal MSAs using both RAM and disk to store nodes. The experimental results obtained in 3 different machines show that the optimizations used in PA-Star can achieve an acceleration of 1.88× in the serial execution, and the parallel execution can attain an acceleration of 5.52× with 8 cores. We also show that PA-Star outperforms a state-of-the-art MSA tool based on A-Star, executing up to 4.77× faster. Finally, we show that our disk-assisted strategy is able to retrieve the optimal alignment when other tools fail.