Phylogenetic trees are fundamental to biology and are bene-fitting several other research areas. Various methods have been developed for inferring such trees, and comparing them is an important problem in computational phylogenetics. Addressing this problem requires tree measures, but all of them suffer from problems that can severely limit their applicability in practice. This also holds true for one of the oldest and most widely used tree measures, the Robinson-Foulds distance. While this measure is satisfying the properties of a metric and is efficiently computable, it has a negatively skewed distribution, a poor range of discrimination and diameter, and may not be robust when comparing erroneous trees. The cluster distance is a measure for comparing rooted trees that can be interpreted as a weighted version of the Robinson-Foulds distance. We show that when compared with the Robinson-Foulds distance, the cluster distance is much more robust towards small errors in the compared trees, and has a significantly improved distribution and range.
展开▼
