Relative Efficiencies of the Fitch-Margoliash, Maximum-Parsimony, Maximum-Likelihood, Minimum-Evolution, and Neighbor-joining Methods of Phylogenetic Tree Construction in Obtaining the Correct Tree

Abstract

The relative efficiencies of several tree-making methods for obtaining the correct phylogenetic tree were studied by using computer simulation. The methods examined were the Fitch-Margoliash (FM), maximum-parsimony (MP), maximumlikelihood (ML), minimum-evolution (ME), and neighbor-joining (NJ) methods. We simulated the evolutionary changes of six DNA sequences each with a length of either 300 or 600 nucleotides. Both constant and varying rates of nucleotide substitution were considered. The DNA sequences generated were used to reconstruct phylogenetic trees by applying the five tree-making methods, and the trees obtained were compared with the model (correct) tree. This process was repeated 50 times for each case, and the following results were obtained: ( 1) The efficiency of obtaining the correct tree for the FM method was considerably lower than those for the other methods. (2) The NJ and ME methods showed a high performance in obtaining the correct tree, and their relative efficiencies were similar to each other. ( 3) For distance methods (NJ, FM, and ME), the results obtained by using corrected nucleotide substitutions were much better than those obtained by using nucleotide differences when the rate of substitution varied greatly among different branches. (4) The ML method was slightly inferior to the NJ and ME methods when a constant rate of nucleotide substitution was assumed, but it was slightly better than the latter two methods when the evolutionary rate varied drastically among branches. If one considers the computational time involved, the NJ method seems to be a method of choice.