IMPROVING THE ACCURACY OF PHYLOGENETIC ESTIMATION BY COMBINING DIFFERENT METHODS

Three methods of phylogenetic tree estimation, UPGMA clustering (UP), maximum parsimony (MP), and neighbor joining (NJ), were used to estimate trees from a large simulation study of 5,400 eight-taxon data sets. The data sets represented nine evolutionary models and 20 tree topologies. The agreement of the trees estimated by the three methods was highly correlated (0.71) with the average agreement with the true tree from which the data sets had been generated. A simple index is proposed for the agreement of different estimation methods that can serve as a measure of the reliability of the joint estimate. This index was coupled with an unbiased character weighting procedure to increase the accuracy of estimation. Accuracy was increased by an average of 24% by use of this procedure in 47 of 61 data sets examined.