RogueTaxa: Drop rogue taxa to generate a more informative consensus

In Rogue: Identify Rogue Taxa in Sets of Phylogenetic Trees

Drop rogue taxa to generate a more informative consensus

Description

RogueTaxa() finds wildcard leaves whose removal increases the resolution or branch support values of a consensus tree, using the relative bipartition, shared phylogenetic, or mutual clustering concepts of information.

Usage

RogueTaxa(
  trees,
  info = c("spic", "scic", "fspic", "fscic", "rbic"),
  return = c("taxa", "tree"),
  bestTree = NULL,
  computeSupport = TRUE,
  dropsetSize = 1,
  neverDrop = character(0),
  labelPenalty = 0,
  mreOptimization = FALSE,
  threshold = 50,
  verbose = FALSE
)

QuickRogue(
  trees,
  info = "phylogenetic",
  p = 0.5,
  log = TRUE,
  average = "median",
  deviation = "mad",
  neverDrop,
  fullSeq = FALSE,
  parallel = FALSE
)

C_RogueNaRok(
  bootTrees = "",
  runId = "tmp",
  treeFile = "",
  computeSupport = TRUE,
  dropsetSize = 1,
  excludeFile = "",
  workDir = "",
  labelPenalty = 0,
  mreOptimization = FALSE,
  threshold = 50
)

Arguments

trees	List of trees to analyse.
info	Concept of information to employ; see details.
return	If taxa, returns the leaves identified as rogues; if tree, return a consensus tree omitting rogue taxa.
computeSupport	Logical: If FALSE, then instead of trying to maximize the support in the consensus tree, RogueNaRok will try to maximize the number of bipartitions in the final tree by pruning taxa.
dropsetSize	Integer specifying maximum size of dropset per iteration. If dropsetSize == n, then RogueNaRok will test in each iteration which tuple of n taxa increases the optimality criterion the most, pruning taxa accordingly. This improves the result, but run times will increase at least linearly.
neverDrop	Tip labels that should not be dropped from the consensus.
labelPenalty	A weight factor to penalize for dropset size when info = "rbic". The higher the value, the more conservative the algorithm is in pruning taxa. The default value of 0 gives the RBIC; 1 gives Pattengale's criterion.
threshold, mreOptimization	A threshold or mode for the consensus tree that is optimized. Specify a value between 50 (majority rule consensus, the default) and 100 (strict consensus), or set mreOptimization = TRUE for the extended majority rule consensus. Note that rogue taxa identified with respect to different thresholds can vary substantially.
verbose	Logical specifying whether to display output from RogueNaRok. If FALSE, output will be included as an attribute of the return value.
p	Proportion of trees that must contain a split before it is included in the consensus under consideration. 0.5, the default, corresponds to a majority rule tree; 1.0 will maximize the information content of the strict consensus.
log	Logical specifying whether to log-transform distances when calculating leaf stability.
average	Character specifying whether to use "mean" or "median" tip distances to calculate leaf stability.
deviation	Character specifying whether to use "sd" or "mad" to calculate leaf stability.
fullSeq	Logical specifying whether to list all taxa (TRUE), or only those that improve information content when all are dropped (FALSE).
parallel	Logical specifying whether parallel execution should take place in C++.
bootTrees	Path to a file containing a collection of bootstrap trees.
runId	An identifier for this run, appended to output files.
treeFile, bestTree	If a single best-known tree (such as an ML or MP tree) is provided, RogueNaRok optimizes the bootstrap support in this best-known tree (still drawn from the bootstrap trees); the threshold parameter is ignored.
excludeFile	Taxa in this file (one taxon per line) will not be considered for pruning.
workDir	Path to a working directory where output files are created.

Details

"Rogue" or (loosely) "wildcard" taxa \insertCiteNixon1992Rogue are leaves whose position in a tree is poorly constrained, typically because much of the phylogenetic data associated with the taxon is either missing or in conflict with other data \insertCiteKearney2002Rogue.

These functions use heuristic methods to identify rogue taxa whose removal improves the information content of a consensus tree, by the definitions of information discussed below.

Value

RogueTaxa() returns a data.frame. Each row after the first, which describes the starting tree, describes a dropset operation. Columns describe:

• num: Sequential index of the drop operation

• taxNum: Numeric identifier of the dropped leaves

• taxon: Text identifier of dropped leaves

• rawImprovement: Improvement in score obtained by this operation

• IC: Information content of tree after dropping all leaves so far, by the measure indicated by info.

C_RogueNaRok() returns 0 if successful; -1 on error.

Functions

• QuickRogue(): Shortcut to "fast" heuristic, with option to return evaluation of all taxa using fullSeq = TRUE.

Information criteria

The splitwise phylogenetic information content measure produces the best results \insertCiteSmithConsRogue. It uses the splitwise information content as a shortcut, which involves double counting of some information (which may or may not be desirable). The same holds for the mutual clustering information measure; this measure is less obviously suited to the detection of rogues. This measure interprets split frequency as a proxy for the probability that a split is true, which is a valid interpretation of a Bayesian posterior sample \insertCiteHolder2008Rogue, a reasonable but imperfect interpretation of a bootstrap sample \insertCiteBerry1996Rogue, and a bad interpretation of a sample of most parsimonious trees.

The "relative bipartition information criterion" (RBIC) is the sum of all support values divided by the maximum possible support in a fully bifurcating tree with the initial set of taxa. The relative bipartition information content approach employs the 'RogueNaRok' implementation \insertCiteAberer2013Rogue, which can handle large trees relatively quickly. The RBIC is is not strictly a measure of information and can produce undesirable results \insertCiteWilkinson2017Rogue.

C_RogueNaRok() directly interfaces the 'RogueNaRok' C implementation, with no input checking; be aware that invalid input will cause undefined behaviour and is likely to crash R.

Author(s)

Martin R. Smith ([e-mail]), linking to RogueNaRok C library by Andre Aberer (<andre.aberer at googlemail.com>)

References

\insertAllCited

Examples

library("TreeTools", warn.conflicts = FALSE)

trees <- list(read.tree(text = ("(a, (b, (c, (d, (e, (X1, X2))))));")),
              read.tree(text = ("((a, (X1, X2)), (b, (c, (d, e))));")))
RogueTaxa(trees, dropsetSize = 2)

trees <- list(
     read.tree(text = "((a, y), (b, (c, (z, ((d, e), (f, (g, x)))))));"),
     read.tree(text = "(a, (b, (c, (z, (((d, y), e), (f, (g, x)))))));"),
     read.tree(text = "(a, (b, ((c, z), ((d, (e, y)), ((f, x), g)))));"),
     read.tree(text = "(a, (b, ((c, z), ((d, (e, x)), (f, (g, y))))));"),
     read.tree(text = "(a, ((b, x), ((c, z), ((d, e), (f, (g, y))))));")
     )
cons <- consensus(trees, p = 0.5)
plot(cons)
LabelSplits(cons, SplitFrequency(cons, trees) / length(trees))
reduced <- RogueTaxa(trees, info = "phylogenetic", ret = "tree")
plot(reduced)
LabelSplits(reduced, SplitFrequency(reduced, trees) / length(trees))

QuickRogue(trees, fullSeq = TRUE)

bootTrees <- system.file("example/150.bs", package = "Rogue")
tmpDir <- tempdir()
XX <- capture.output( # Don't print verbose run details to console
  C_RogueNaRok(bootTrees, workDir = tmpDir)
)

# Results have been written to our temporary directory
oldwd <- setwd(tmpDir)
head(read.table("RogueNaRok_droppedRogues.tmp", header = TRUE))

# Delete temporary files
file.remove("RogueNaRok_droppedRogues.tmp")
file.remove("RogueNaRok_info.tmp")

setwd(oldwd)

Rogue documentation built on May 29, 2024, 5:05 a.m.