PERFORMANCE OF PHYLOGENETIC METHODS IN SIMULATION

Computer simulations are useful because they can characterize the expe cted performance of phylogenetic methods under idealized conditions. H owever, simulation studies are also subject to several sources of bias that make the results of different simulation studies difficult to in terpret and often contradictory. In this study, I examined the perform ance of 26 commonly used methods of phylogenetic inference for three s tatistical criteria: consistency, efficiency, and robustness. Methods examined included parsimony (general, weighted, and transversion), max imum likelihood (assuming Jukes-Cantor and Kimura models of DNA substi tution), and UPGMA, minimum evolution, and weighted and unweighted lea st squares (with uncorrected, Jukes-Cantor, Kimura, modified Kimura, a nd gamma distances). The performance of methods was examined under thr ee models of DNA substitution for four taxa. The branch lengths of the four-taxon trees were varied extensively in this simulation. The resu lts indicate that most methods perform well (i.e., estimate the correc t tree greater than or equal to 95% of Be time) over a large portion o f the four-taxon parameter space. In general, maximum likelihood perfo rmed best, followed by the additive distance methods and the parsimony methods. Lake's method of invariants and UPGMA are, respectively, ine fficient and extremely sensitive to branch-length inequalities. In gen eral, differential weighting of character-state transformations increa ses the performance of methods when the weighting can be applied appro priately. Although methods differ in their consistency, efficiency, an d robustness, additional criteria-mainly falsifiability-are extremely important considerations when choosing a method of phylogenetic infere nce.