Multiple UPGMA and Neighbor-joining Trees and the Performance of Some Computer Packages

Although the unweighted pair-group method using arithmetic averages (UPGMA) and neighbor-joining (NJ) algorithms are designed to produce single trees, they may derive more than one topology from a single matrix, depending on the order of data entry. This "chaotic" behavior is due to ties, the effects of which are rarely considered. Therefore we present examples of ties and show that multiple UPGMA and NJ trees cannot be neglected with molecular data based on allozyme distances or "binary" distances derived from random amplified polymorphic DNA, restriction fragments, DNA fingerprints, or general protein patterns. We also compare the performance of 15 computer packages with respect to ties. Five programs recognize the problem (PHYLIP, MVSP, SAS, SYN-TAX, and NTSYS) but deal with it in different ways. We further point out that if ties are not properly taken into account, they might affect bootstrap and jackknife confidence estimates. Finally, we observed that NTSYS, PHYLIP, MVSP, and MVSP87 have different efficiencies in finding ties, and that some programs, including MEGA, may produce single alternative UPGMA topologies, probably due to their different rounding precisions or tie tolerances.