Biology Reference
In-Depth Information
sequence evolution.
b
Such models of evolution are not used to
classify tree building methods.
The ones used are rather related to the underlying tree struc-
ture. Some character and distance methods, for example, assume a
molecular clock. Under a molecular clock, the underlying phylo-
genetic tree has equal root-to-leaf path lengths for all lineages.
Such a tree is called an ultrametric tree. Another example concerns
distance methods. As mentioned above, evolutionary distances
have to be estimated and are therefore uncertain. Nevertheless,
there are phylogenetic methods that assume exact distances called
additive tree construction methods. Distance methods that take
the inference uncertainty into account, on the other hand, use a
model to characterize the error. Least squares tree construction is
an example of such a method; Sec. 4 is devoted to the correspon-
ding method developed in our research group. A property
assumed implicitly by essentially all distance methods is additivity.
In the evolution from an object
A
to
B
and then from
B
to
C
,
additive distances have the property
d
AB
+
d
AC
. To be able to
derive branch lengths from leaf-to-leaf distances (sums of many
branches), the distance function has to be additive. Distance esti-
mates of nonadditive distances are not suitable for tree recon-
struction.
c
d
BC
=
•
Algorithmic methods
. Algorithmic methods can be divided into
two classes. In the first class, the tree inference and the definition
of a preferred tree are combined into a single statement. Such
methods usually perform some sort of clustering and have the
advantage of being fast. They exist for character and distance input
data. They are widely used for the latter and are in this case typi-
cally designed to infer the correct tree, given that the distances are
b
The first-order Markov property implies that each character mutates with a proba-
bility that depends only on itself and not on the history of previous mutations.
Additionally to the Markov property, the sites in a sequence are in almost all models
assumed to mutate independently of each other.
c
The distance metric related to the percent identity (essentially one minus the fraction
of identical residues in an alignment) is an example of a nonadditive distance.
