Branch Support Metrics for Phylogenetic Trees

Abstract There are various approaches to assessing the reliability of phylogenetic tree branches. In this study, several branch support metrics were tested. Some of the tested metrics are based on various replication generation methods, including bootstrap, jackknife, methods recently implemented in the Muscle5 package, and our proposed procedure based on iterative shuffling of the order of objects in a distance matrix. Faster versions of the bootstrap and jackknife methods, as well as several probabilistic branch reliability estimators, were also tested. For each metric, the area under the ROC, the optimal branch support value based on the F1 metric, and the sensitivity and specificity for the optimal value were calculated. Of the two methods for calculating branch support based on replications, the classical Felsenstein approach demonstrated the best predictive power. All replication generation methods yield support metrics with similar predictive power. Probability estimates were significantly inferior not only to classical bootstrap and jackknife, but also to their accelerated versions.