The Robinson-Foulds (RF) metric is arguably the most widely used measure of
phylogenetic tree similarity, despite its well-known shortcomings: For example,
moving a single taxon in a tree can result in a tree that has maximum distance
to the original one; but the two trees are identical if we remove the single
taxon. To this end, we propose a natural extension of the RF metric that does
not simply count identical clades but instead, also takes similar clades into
consideration. In contrast to previous approaches, our model requires the
matching between clades to respect the structure of the two trees, a property
that the classical RF metric exhibits, too. We show that computing this
generalized RF metric is, unfortunately, NP-hard. We then present a simple
Integer Linear Program for its computation, and evaluate it by an
all-against-all comparison of 100 trees from a benchmark data set. We find that
matchings that respect the tree structure differ significantly from those that
do not, underlining the importance of this natural condition.

该研究提出了 RF 指标一种自然的扩展方式，该指标不仅考虑相同的分类单元，还考虑相似的分类单元，并且需要保持树的结构，用一个简单的整数线性规划来计算广义 RF 指标，发现与不保持树结构的匹配方式有显着不同。