Reusable dynamic programming: Updating sequence alignment

Sequence alignment in genomics and proteomics is mostly done via dynamic programming (DP) based approaches. In this work, we show how computational results from DP can be reused to update alignments when analyzing new versions of a sequence. We derive relative tolerance bounds on node distances from a root node that guarantee that partial shortest path distances remain optimal. We then propose an algorithm that uses these bounds to skip all unperturbed parts of a sequence when recomputing an alignment. We also discuss techniques to reduce the memory requirements of the algorithm by focusing on the highly conserved segments of the sequence. Experimental results establish that our proposed alignment procedure can update alignment decisions of modified sequence with 4.6% to 18% of the number of computations required by the normal Needleman-Wunsch algorithm, depending on sequence length. Higher computational savings are achieved with longer sequences.