Phylogenetic Inference using Genetic Algorithm-based Least Squares Methods

The only illustration in Darwin’s Origin of the Species shows a phylogenetic tree. This type of tree is used to represent the evolutionary relationship between different species. Phylogenetic inference is constructing such a tree based on the measured pairwise distance of all species. There exists many methods to evaluate how well a particular tree fits the measurements. One of these is called the least squares method, in which we wish to minimize the square of the error of all species between the measured distance and the distance in the tree. Because the search space of trees is huge, it is not possible to examine all possibilities; it is necessary to utilize a search heuristic. We have chosen to use genetic algorithms (GA) which handles and changes a population of trees. GAs have been used in several articles on similar problems, but we find that it is often applied superficially. For instance there is seldom spent time on choosing the GA’s operators, the initial population or how to choose the parameters of the GA. In this thesis we have therefore implemented a genetic algorithm to solve the phylogeny problem with respect to least squares, and we have examined how the initial population and in particular the parameters influences the efficiency of the GA. Our experiments have shown that a self-adaptive approach gives good results without the need for time consuming tuning of parameters, and that the GA can be quickened by basing the initial population on a faster heuristic. In addition, our GA has been able to find better results than the neighbor joining algorithm.