Motivation: The Robinson–Foulds (RF) metric is widely used by biologists, linguists and chemists to quantify similarity between pairs of phylogenetic trees. The measure tallies the number of bipartition splits that occur in both trees—but this conservative approach ignores potential similarities between almost-identical splits, with undesirable consequences. ‘Generalized’ RF metrics address this shortcoming by pairing splits in one tree with similar splits in the other. Each pair is assigned a similarity score, the sum of which enumerates the similarity between two trees. The challenge lies in quantifying split similarity: existing definitions lack a principled statistical underpinning, resulting in misleading tree distances that are difficult to interpret. Here, I propose probabilistic measures of split similarity, which allow tree similarity to be measured in natural units (bits). Results: My new information-theoretic metrics outperform alternative measures of tree similarity when evaluated against a broad suite of criteria, even though they do not account for the non-independence of splits within a single tree. Mutual clustering information exhibits none of the undesirable properties that characterize other tree comparison metrics, and should be preferred to the RF metric. Availability and implementation: The methods discussed in this article are implemented in the R package ‘TreeDist’, archived at https://dx.doi.org/10.5281/zenodo.3528123. Supplementary information: Supplementary data are available at Bioinformatics online.
Phylogenetic analyses often produce large numbers of trees. Mapping trees’ distribution in “tree space” can illuminate the behavior and performance of search strategies, reveal distinct clusters of optimal trees, and expose differences between different data sources or phylogenetic methods—but the high-dimensional spaces defined by metric distances are necessarily distorted when represented in fewer dimensions. Here, I explore the consequences of this transformation in phylogenetic search results from 128 morphological data sets, using stratigraphic congruence—a complementary aspect of tree similarity—to evaluate the utility of low-dimensional mappings. I find that phylogenetic similarities between cladograms are most accurately depicted in tree spaces derived from information-theoretic tree distances or the quartet distance. Robinson–Foulds tree spaces exhibit prominent distortions and often fail to group trees according to phylogenetic similarity, whereas the strong influence of tree shape on the Kendall–Colijn distance makes its tree space unsuitable for many purposes. Distances mapped into two or even three dimensions often display little correspondence with true distances, which can lead to profound misrepresentation of clustering structure. Without explicit testing, one cannot be confident that a tree space mapping faithfully represents the true distribution of trees, nor that visually evident structure is valid. My recommendations for tree space validation and visualization are implemented in a new graphical user interface in the “TreeDist” R package. [Multidimensional scaling; phylogenetic software; tree distance metrics; treespace projections.]
Phylogenetic analysis aims to establish the true relationships between taxa. Different analytical methods, however, can reach different conclusions. In order to establish which approach best reconstructs true relationships, previous studies have simulated datasets from known tree topologies, and identified the method that reconstructs the generative tree most accurately. On this basis, researchers have argued that morphological datasets should be analysed by Bayesian approaches, which employ an explicit probabilistic model of evolution, rather than parsimony methods—with implied weights parsimony sometimes identified as particularly inaccurate. Accuracy alone, however, is an inadequate measure of a tree's utility: a fully unresolved tree is perfectly accurate, yet contains no phylogenetic information. The highly resolved trees recovered by implied weights parsimony in fact contain as much useful information as the more accurate, but less resolved, trees recovered by Bayesian methods. By collapsing poorly supported groups, this superior resolution can be traded for accuracy, resulting in trees as accurate as those obtained by a Bayesian approach. By contrast, equally weighted parsimony analysis produces trees that are less resolved and less accurate, leading to less reliable evolutionary conclusions.
Morphological data play a key role in the inference of biological relationships and evolutionary history and are essential for the interpretation of the fossil record. The hierarchical interdependence of many morphological characters, however, complicates phylogenetic analysis. In particular, many characters only apply to a subset of terminal taxa. The widely used "reductive coding" approach treats taxa in which a character is inapplicable as though the character's state is simply missing (unknown). This approach has long been known to create spurious tree length estimates on certain topologies, potentially leading to erroneous results in phylogenetic searches-but pratical solutions have yet to be proposed and implemented. Here, we present a single-character algorithm for reconstructing ancestral states in reductively coded data sets, following the theoretical guideline of minimizing homoplasy over all characters. Our algorithm uses up to three traversals to score a tree, and a fourth to fully resolve final states at each node within the tree. We use explicit criteria to resolve ambiguity in applicable/inapplicable dichotomies, and to optimize missing data. So that it can be applied to single characters, the algorithm employs local optimization; as such, the method provides a fast but approximate inference of ancestral states and tree score. The application of our method to published morphological data sets indicates that, compared to traditional methods, it identifies different trees as "optimal." As such, the use of our algorithm to handle inapplicable data may significantly alter the outcome of tree searches, modifying the inferred placement of living and fossil taxa and potentially leading to major differences in reconstructions of evolutionary history.