Tree Bipartition and Bootstrapping Phylogenies
These functions analyse bipartitions found in a series of trees.
prop.part counts the number of bipartitions found in a series
of trees given as
.... If a single tree is passed, the
returned object is a list of vectors with the tips descending from
each node (i.e., clade compositions indexed by node number).
prop.clades counts the number of times the bipartitions present
phy are present in a series of trees given as
in the list previously computed and given with
boot.phylo performs a bootstrap analysis.
boot.phylo(phy, x, FUN, B = 100, block = 1, trees = FALSE, quiet = FALSE, rooted = FALSE) prop.part(..., check.labels = TRUE) prop.clades(phy, ..., part = NULL, rooted = FALSE) ## S3 method for class 'prop.part': print(x, ...) ## S3 method for class 'prop.part': summary(object, ...) ## S3 method for class 'prop.part': plot(x, barcol = "blue", leftmar = 4, ...)
boot.phylo must be the function used
to estimate the tree from the original data matrix. Thus, if the tree
was estimated with neighbor-joining (see
nj), one maybe wants
FUN = function(xx) nj(dist.dna(xx)).
boot.phylo specifies the number of columns to
be resampled altogether. For instance, if one wants to resample at the
block = 3 must be used.
check.labels = FALSE in
computing times. This requires that (i) all trees have the same tip
labels, and (ii) these labels are ordered similarly in all
trees (in other words, the element
tip.label are identical in
The plot function represents a contingency table of the different
partitions (on the x-axis) in the lower panel, and their observed
numbers in the upper panel. Any further arguments (...) are used to
change the aspects of the points in the lower panel: these may be
cex, etc. This function
works only if there is an attribute
labels in the object.
The print method displays the partitions and their numbers. The summary method extracts the numbers only.
prop.part returns an object of class
is a list with an attribute
"number". The elements of this list
are the observed clades, and the attribute their respective
numbers. If the default
check.labels = FALSE is used, an
"labels" is added, and the vectors of the returned
object contains the indices of these labels instead of the labels
boot.phylo return a numeric vector
which ith element is the number associated to the ith
trees = TRUE,
a list whose first element (named
"BP") is like before, and the
second element (
"trees") is a list with the bootstraped
summary returns a numeric vector.
prop.clades calls internally
prop.part with the option
check.labels = TRUE, which may be very slow. If the trees
... fulfills conditions (i) and (ii) above, then it
might be faster to first call, e.g.,
pp <- prop.part(...), then
use the option
prop.clades(phy, part = pp).
You have to be careful that by default
the trees as unrooted and this may result in spurious results if the
trees are rooted (see examples).
Efron, B., Halloran, E. and Holmes, S. (1996) Bootstrap confidence levels for phylogenetic trees. Proceedings of the National Academy of Sciences USA, 93, 13429--13434.
Felsenstein, J. (1985) Confidence limits on phylogenies: an approach using the bootstrap. Evolution, 39, 783--791.
data(woodmouse) f <- function(x) nj(dist.dna(x)) tr <- f(woodmouse) ### Are bootstrap values stable? for (i in 1:5) print(boot.phylo(tr, woodmouse, f, quiet = TRUE)) ### How many partitions in 100 random trees of 10 labels?... TR <- replicate(100, rtree(10), FALSE) pp10 <- prop.part(TR) length(pp10) ### ... and in 100 random trees of 20 labels? TR <- replicate(100, rtree(20), FALSE) pp20 <- prop.part(TR) length(pp20) plot(pp10, pch = "x", col = 2) plot(pp20, pch = "x", col = 2) set.seed(1) tr <- rtree(10) # rooted by default prop.clades(tr, tr) # clearly wrong prop.clades(tr, tr, rooted = TRUE) tr <- rtree(10, rooted = FALSE) prop.clades(tr, tr) # correct