Parallel Biological Sequence Comparison in Linear Space with Multiple Adjustable Bands.

In this paper, we propose and evaluate Fickett-MM, a parallel strategy that combines the algorithms Fickett and Myers-Miller, splitting a pairwise sequence comparison into multiple comparisons of subsequences and calculating an appropriate Fickett band to each subsequence comparison (block). With this approach, we potentially reduce the number of cells calculated in the dynamic programming matrix when compared to Fickett, which uses a unique band to the whole comparison. Our adjustable multi-block strategy was integrated to the stage 4 of CUDAlign, a state-of-the-art parallel tool for optimal biological sequence comparison. Fickett-MM was used to compare real DNA sequences whose sizes ranged from 10KBP (Thousands of Base Pairs) to 47MBP (Millions of Base Pairs), reaching a speedup of 59.60\(\times \) in the 10MBP \(\times \) 10MBP comparison when compared to CUDAlign stage 4.