An R package for fast phylogenetic consensus tree algorithms. • ConsTree

‘ConsTree’ is an R package providing a comprehensive, efficient suite of methods for summarizing a collection of phylogenetic trees — for example a bootstrap or Bayesian posterior sample — as a single consensus tree.

It builds on the tree and split infrastructure of ‘TreeTools’, and repackages a family of asymptotically-efficient consensus algorithms developed by Jesper Jansson and colleagues, several of which have not previously been available in R.

Consensus methods

Split-selection methods

Each method takes a list of trees (or a multiPhylo) sharing the same leaves and returns a phylo. They differ in which groupings (splits or clusters) they retain:

Function	Retains a grouping when…
`Strict()`	it occurs in every tree
`Majority()` / `MajorityRule()`	it occurs in > half the trees (tunable via `p`)
`Loose()`	no tree contradicts it (semi-strict / combinable-component)
`MajorityPlus()`	more trees display it than contradict it
`Frequency()`	it is more frequent than every grouping that conflicts with it (frequency-difference)
`Greedy()`	added greedily, most frequent first, if compatible with those already kept (extended majority-rule)
`Adams()`	constructed from the finest root-level partition shared by every tree (may introduce novel groupings; rooted)
`Local()`	based on rooted triplets shared by every tree (minimum rooted/induced local consensus; ≤ 20 leaves)
`RStar()`	each rooted triplet grouping that wins a plurality against each alternative separately

Distance and branch-length summaries

A second family summarizes the trees through a distance or treespace criterion rather than by selecting groupings:

Function	Summary
`Average()`	the tree best fitting the mean path-length (patristic) distances of the inputs
`Quartet()`	an approximate median minimizing the total quartet distance to the inputs; often more resolved than majority-rule
`BHVMean()`	the Fréchet mean tree in Billera–Holmes–Vogtmann treespace, with branch lengths; `BHVDistance()`, `BHVPairwiseDistances()` and `BHVVariance()` provide the supporting geodesic distances and dispersion

Usage

library("ConsTree")

trees <- ape::as.phylo(1:100, 8)   # 100 eight-leaf trees

Strict(trees)        # most conservative
Majority(trees)      # the familiar 50% majority-rule tree
Loose(trees)         # everything not actively contradicted
Frequency(trees)     # frequency-difference: often more resolved than majority
Greedy(trees)        # most resolved of the split-based summaries

Installation

Install the development version from GitHub:

if (!require("remotes")) install.packages("remotes")
remotes::install_github("ms609/ConsTree")

‘ConsTree’ is not yet on CRAN.

Relationship to other packages

‘ConsTree’ is the front-end consensus toolkit for the TreeTools ecosystem, which also includes ‘TreeDist’ (tree distances and information-theoretic consensus) and ‘TreeSearch’ (phylogenetic search). TreeTools itself remains the fast engine for the strict and majority-rule consensus, which ‘ConsTree’ exposes through consistently-named wrappers.

‘TreeDist’ also offers a complementary summary that ‘ConsTree’ does not duplicate: the tree from a sample that has the lowest median clustering-information distance (CID) to the others — a single representative of the sample rather than a constructed consensus.

The quartet machinery underlying Quartet() builds on the ‘Quartet’ package, which counts the resolved- and shared-quartet statistics between trees; and the BHV summaries relate to ‘distory’, which computes geodesic distances in the same treespace.

The ‘Rogue’ package identifies unstable (‘rogue’) leaves whose removal can improve the resolution and support of a consensus tree; dropping rogues before summarizing with ‘ConsTree’ often yields a better-resolved result.

Citation and attribution

The algorithms repackaged here originate in the FACT, FACT2 and FDCT prototypes of Jesper Jansson and colleagues, incorporated with permission; see inst/REFERENCES.bib for the source papers, and DESCRIPTION for copyright attribution. ‘ConsTree’ is released under GPL (≥ 3).

Please note that this project is released with a Contributor Code of Conduct. By contributing, you agree to abide by its terms.