Making the Shortest-paths Approach to Sum-of-pairs Multiple Sequence Alignment More Space Eecient in Practice

The MSA program is one of the few existing programs that attempts to nd optimal alignments of multiple protein or DNA sequences. The MSA program was written and distributed in 1989. It implements a branch-and-bound technique together with a variant of Dijkstra's shortest paths algorithm to prune the basic dynamic programming graph. We have made substantial improvements in the time and space usage of MSA. The improvements make feasible a variety of problem instances that were not feasible previously. On some runs we achieve an order of magnitude reduction in space usage and a signi cant multiplicative factor improvement in running time. To explain how the improvements work, it is necessary to give a much more detailed description of MSA than has been previously published. In practice, MSA rarely produces a provably optimal alignment; the reasons for lack of optimality are discussed at length in the body of the paper. 2