Aligning Alignments

While the area of sequence comparison has a rich collection of results on the alignment of two sequences, and even the alignment of multiple sequences, there is little known about the alignment of two alignments. The problem becomes interesting when the alignment objective function counts gaps, as is common when aligning biological sequences, and has the form of the sum-of-pairs objective. We begin a thorough investigation of aligning two alignments under the sum-of-pairs objective with general linear gap costs when either of the two alignments are given in the form of a sequence (a degenerate alignment containing a single sequence), a multiple alignment (containing two or more sequences), or a proole (a representation of a multiple alignment often used in computational biology). This leads to ve problem variations, some of which arise in widely-used heuristics for multiple sequence alignment, and in assessing the relatedness of a sequence to a sequence family. For variations in which exact gap counts are computationally diicult to determine, we offer a framework in terms of optimistic and pessimistic gap counts. For optimistic and pessimistic gap counts we give eecient algorithms for the sequence vs. alignment, sequence vs. proole, alignment vs. alignment, and proole vs. proole variations, all of which run in essentially O(mn) time for two input alignments of lengths m and n. For exact gap counts, we give the rst provably eecient algorithm for the sequence vs. alignment variation, which runs in essentially O(mn log n) time using the candidate-list technique developed for convex gap-costs, and we conjecture that the alignment vs. alignment variation is NP-complete.