Testing the Effects of Speciation and Mutation Rates on Distance-Based Phylogenetic Tree Construction Accuracy Using EcoSim

Abstract

Biologists regularly construct phylogenetic trees during their research in order to understand the evolutionary history of their subjects. Typically the actual phylogenetic tree is not known, and thus the phylogenetic tree produced is only an estimate. There are two main categories of phylogenetic tree construction, distance-based methods and character-based methods. In all cases, the effects of mutation rate and genetic compactness of species clusters on phylogenetic tree construction accuracy are not well understood. In order to test the correctness of a particular method, it is imperative to perform the study in a system for which the actual phylogenetic tree is known. Thus, realistic and complex simulations in which evolution and speciation occur provide a perfect platform for such a study. EcoSim is such a simulation, and is a simulation in which predator and prey agents interact, evolve, and speciate. Agents in EcoSim possess a complex, evolving, heritable behavioral model which provides meaningful evolution. Here, EcoSim was used as a platform on which to test the effects of mutation rate and speciation threshold on the accuracy of several distance-based phylogenetic tree construction methods. EcoSim has the ability to record all speciation events during a run, therefore we were able to construct the actual phylogenetic trees allowing us to properly compare estimation methods. We created four EcoSim run types: mutation rate increased (MRI), mutation rate decreased (MRD), speciation threshold increased (STI), and speciation threshold decreased (STD). We created five runs of each type, as well as five runs using the standard EcoSim configuration that all lasted 10000 time-steps. At various time-steps throughout each run, we performed Neighbor-Joining, UPGMA, and Fitch-Margoliash tree construction (with and without bootstrapping) on random subsets of 10 species that existed during that time-step, and compared the results to the actual phylogenetic tree using the symmetric distance metric. We found that Neighbor-Joining and Fitch-Margoliash performed nearly equally well, whereas UPGMA performed relatively poorly overall. Further, we found that an increase in speciation rate leads to performance losses in phylogenetic tree construction whereas modifying the mutation rate typically leads to performance gains.