I am quoting a book called "Phylogenetic Analysis of DNA sequences". Chapter 6 by
Masatoshi Nei is called "Relative efficiencies of different tree making methods for
Molecular Data." I don't have the editor offhand but I can check it out if you are
interested.
Maximum Likelihood (ML) method:
In this method, the nucleotides of all DNA sequences at each nucleotide site are
considered separately, and the log-likelihood of having these nucleotides are
computed for a given topology by using a particular probability model (Felsenstein,
1981a). This log-likelihood is added for all nucleotide sites, and the sum of the
log-likelihood is maximized to estimate the branch length of the tree. This
procedure is repeated for all possible topologies, and topology that shows the
highest likelihood is chosen as the final one."
Maximum Parsimony (MP) method
In this method, the DNA (or amino acid) sequences of ancestral species are inferred
from those of extant species, considering a particular tree topology, and the
minimum number of evolutionary changes that are required to explain all the observed
differences among the sequences is computed. This number is obtained for all
possible topologies, and the topology which shows the smallest number of
evolutionary changes is chosen as the final tree. This method is used mainly for
finding the topology of a tree, but branch lengths can be estimated under certain
assumptions. When the MP method is applied to morphological characters, it is
customary to assume that the primitive and derived character states are known. In
the case of molecular data, this assumption generally does not hold, and different
character states are often reversible. It is, therefore, important to use the MP
method, which permits reversible mutations. In numerical taxonomy , this type of MP
method is sometimes called the Wagner parsimony method.
The book also lists about 7 different types of distance methods. Distance methods
require some sort of conversion of the data into a matrix form. By contrast, MP and
ML are both examples of Discrete Character Data. The distance methods given in this
book areUPGMA. Transformed Distance method, Fitch and Margoliash (FM), Minimum
Evolution method, Distance Wagner method, Neighborliness method, Neighbor Joining
method.
I had to give a seminar to a fourth year molecular biology course on methods of
statsitical analysis! Ug.
Fiona McArthur
Mike Syvanen wrote:
>fatherdes at hotmail.com wrote:
>> > I detect a cooling off these days and a coming together as ML methods become
> > more practical and influential. Maybe everyone sees these as a way of going
> > forward without giving in to the other side. Personally, I imagine Ml methods
> > will become more and more widely used and with good reason but parsimony and
> > distance methods still have their uses.
> >
> > Des Higgins
> > Department of Biochemistry
> > University College
> > Cork Ireland
>> Please correct an impression that I have carried around. I do not use ML
> methods, but after once hearing it described, I thought it worked on the same
> assumptions as did the traditional distance methods. Hence, there are just two
> categories of methods -- namely distance and parsimony. If this view is not
> correct, could someone provide a reference that describes ML (in grammatical
> sentences) and contrasts it to distance and parsimony methods.
>> Thanks
>> Mike Syvanen