Robust Analysis of Phylogenetic Tree Space

Abstract

Phylogenetic analyses often produce large numbers of trees. Mapping trees’ distribution in ‘tree space’ can illuminate the behaviour 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 datasets, 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.

Using Information Theory to Detect Rogue Taxa and Improve Consensus Trees

“…The existence of rogues militates for a detailed evaluation of optimal trees. The only way to fully evaluate the signal within a set of trees is to scrutinize each individual topology, perhaps assisted by an exploration of the underlying tree space ( St. John 2017 ; Smith 2022 ) and the construction of a “profile” of complementary consensus trees ( Wilkinson 1996 ; Wilkinson 2003 ).…”

The Cambrian cirratuliform Iotuba denotes an early annelid radiation

“…A subsample of the individual bifurcating trees reconstructed by each approach was compared by mapping the clustering information and quartet distances [109,110] between each pair of trees into two dimensions using classical multidimensional scaling [111], using the R packages 'QUARTET' and 'TREEDIST' [112,113]. Adequacy of projection was indicated by trustworthiness and continuity metrics above 0.90 and minimal deformation of the minimum spanning tree [114][115][116][117]. This analysis indicated that all methods except uncorrected Fitch parsimony converged onto a similar region of tree space.…”

Geodesics to Characterize the Phylogenetic Landscape

“…treespace (Jombart et al 2017) is an R package that allows a wide variety of tree metrics and methods for clustering trees to be used. Smith (2022) evaluated the performance of multiple aspects of low-dimensional representations of sets of trees, and provides an R package TreeDist for users to do the same. Finally, R We There Yet (RWTY) (Warren et al 2017), a package for analyzing Bayesian analyses convergence, can produce nonlinear MDS visualization of landscapes using the RF or path difference distance (Steel and Penny 1993) and colored by the likelihood.…”

“…Finally, R We There Yet (RWTY) (Warren et al 2017), a package for analyzing Bayesian analyses convergence, can produce nonlinear MDS visualization of landscapes using the RF or path difference distance (Steel and Penny 1993) and colored by the likelihood. Some authors (Amenta et al 2015; Wilgenbusch et al 2017; Smith 2022) have analyzed how well MDS visualizes treespace, and suggested validation measures. There are other ways to visualize sets of related trees beyond dimensionality reduction, such as super-imposing the trees on each other, as in DensiTree (Bouckaert 2010), or sophisticated tree comparison visualizers, like ADView (Liu et al 2019).…”

“…Two more recent packages for visualizing sets of trees are T reescaper (Huang et al 2016), which offers different cost functions for dimensionality reduction and community detection, and treespace (Jombart et al 2017), which offers a wide variety of tree metrics and methods for clustering trees. Smith (2022) evaluated the performance of multiple aspects of low-dimensional representations of sets of trees. Note that there are other ways to visualize sets of related trees beyond dimensionality reduction, such as super-imposing the trees on each other, as in DensiTree (Bouckaert 2010), or sophisticated tree comparison visualizers, like ADView (Liu et al 2019).…”