Biology Reference
In-Depth Information
of fit. The best tree will have the smallest percentage standard deviation. The
Fitch-Margoliash and UPGMA methods should result in very similar trees if a
molecular clock is operating.
Maximum-parsimony methods
focus on the character values observed for
each species, rather than working with the distances between sequences that
summarize differences between character values. These methods
minimize
the numbers of changes in sequences between species over the tree, usu-
ally making the assumption that there have been approximately constant
rates of change. Branch lengths usually are not obtained. Maximum parsi-
mony is widely used and works well when change is rare or branches are short
(
Pagel 1999
). Parsimony methods can work poorly when rates of character
evolution are high and the phylogeny includes some long branches because it
tends to underestimate the amount of change in long branches. In some circum-
stances, maximum-likelihood and parsimony methods can provide equivalent
results.
For each possible parsimony tree, the sequences at each node are inferred
to be those that require the least number of changes to give each of the two
sequences for the immediate descendants. The total number of changes
required over the whole tree is found, and the tree with the minimum number
of changes is the most parsimonious. Parsimony methods assume that genetic
changes are improbable. However, if there are large amounts of change, par-
simony methods can yield estimated trees that are inaccurate (
Swofford and
Olson 1990
).
Stewart (1993)
pointed out that parsimony analysis can be prob-
lematic for two general reasons: 1) failure to find the shortest tree and 2) the
shortest tree is not the correct phylogeny. Failure to find the shortest tree can
occur if too many taxa or too few informative data are used.
Likelihood methods
of analyzing DNA data rely on genetic models, and can
provide a basis for statistical inference. Likelihood is an amount proportional
to the probability of observing the data, given a model. Likelihood methods
are more difficult to compute than the methods described above (
Weir 1990
).
Maximum-likelihood methods of tree construction assume the form of the tree
and then choose the branch length to maximize the likelihood of the data given
that tree. These likelihoods then are compared over different possible trees and
the tree with the greatest likelihood is considered to be the best estimate.
Unfortunately, the number of possible trees increases very rapidly as the num-
ber of taxa under consideration increases in likelihood methods. Thus, if three
species are being compared, the number of possible unrooted trees is one,
with four species it is three trees, with six species it is 105 trees, and with eight
