Discrete Branch Length Representations for Genetic Algorithms in Phylogenetic Search

Likelihood analysis for phylogenetics is a well suited domain for optimization by genetic algorithm. A major factor in the performance of genetic algorithms is representation and operators. In this paper we present promising preliminary results on using a discrete set of edge lengths in the phylogenetic trees. We find that discretizing the edge lengths changes the fundamental character of the search and can produce higher quality trees. Our results suggest a search that is more robust to traps in local optima and an opportunity to better control the balance between topology search and edge length search.