SALSA: Sequence ALignment via Steiner Ancestors

We describe SALSA (Sequence ALignment via Steiner Ancestors), a public{domain suite of programs for generating multiple alignments of a set of genomic sequences. We allow the use of either of the two popular objectives, Tree Alignment or Sum-of-Pairs. The main distinguishing feature of our method is that the alignment is obtained via a tree in which the internal nodes (ancestors) are labeled by Steiner sequences for triples of the input sequences. Given lists of candidate labels for the ancestral sequences, we use dynamic programming to choose an optimal labeling under either objective functions. Finally, the fully labeled tree of sequences is turned into into a multiple alignment. Enhancements in our implementation include the traditional space-saving ideas of Hirschberg as well as new data-packing techniques. The running-time bottleneck of computing exact Steiner sequences is handled by a highly eeective but much faster heuristic alternative. Finally, other modules in the suite allow automatic generation of linear-program input les that can be used to compute novel lower bounds on the optimal values. We also report on some preliminary computational experiments with SALSA.