Reconstructing Optimal Phylogenetic Trees: A Challenge in Experimental Algorithmics

The benefits of experimental algorithmics and algorithm engineering need to be extended to applications in the computational sciences. In this paper, we present on one such application: the reconstruction of evolutionary histories (phylogenies) from molecular data such as DNA sequences. Our presentation is not a survey of past and current work in the area, but rather a discussion of what we see as some of the important challenges in experimental algorithmics that arise from computational phylogenetics. As motivational examples or examples of possible approaches, we briefly discuss two specific uses of algorithm engineering and of experimental algorithmics from our recent research. The first such use focused on speed: we reimplemented Sankoff and Blanchette’s breakpoint analysis and obtained a 200, 000-fold speedup for serial code and 10-fold speedup on a 512-processor supercluster. We report here on the techniques used in obtaining such a speedup. The second use focused on experimentation: we conducted an extensive study of quartet-based reconstruction algorithms within a parameter-rich simulation space, using several hundred CPU-years of computation. We report here on the challenges involved in designing, conducting, and assessing such a study.