Biology Reference
In-Depth Information
•
use of lazy subtree rearrangement (LSR), a technique which avoids
having to recalculate the complete likelihood function after tree
rearrangement;
•
dynamic adaptation of rearrangement distance. The LSR technique
allows larger and smaller degrees of change during tree rearrange-
ment. This parameter is optimized at the beginning of the search;
•
low-level optimizations of the likelihood functions for several
models of evolution; and
•
efficient implementation of a CMP method to create a starting
tree for the ML search.
4.5.5. Creating consensus trees
Often, when searching for the “best tree”, a number of equally good
trees are found. This is especially true for CMP and CC analyses executed
on large molecular datasets, where several MP trees may suggest com-
pletely contradictory evolutionary scenarios. Obviously, one could consider
all of these MP trees individually as alternative solutions; but when
several become several thousands, there is a practicality problem.
So, when there are too many MP trees, it becomes necessary to
create a consensus tree from all or part of the MP trees. This can be done
by drawing a tree containing only the branches common to a certain
percentage of MP trees. When the percentage is 50%, 66%, or 100%, the
consensus is referred to as majority consensus, semistrict consensus, and
strict consensus, respectively.
A somewhat more complicated consensus definition is the Adams
consensus. Each MP tree is traced from the root to the leaves, and at each
bifurcation the two subsets of terminal taxa are determined. If there is an
overlap of these subsets in all MP trees, then it is retained in the Adams
consensus.
r
The method is impractical if the number of trees is large.
r
Consider trees I and II in Fig. 13. The strict consensus would have an unresolved
tetrachotomy, forming the rake-like tree III. Let us see what the Adams consensus
looks like. Starting from the root and arriving at the first intersection of tree I,
we obtain subsets {
a
} and {
b, c, d
}. Likewise, for tree II, we obtain the subsets {
a, b,
c
} and {
d
}. An overlap or nonempty intersection between {
b, c, d
} and {
a, b, c
} exists.
This subset {
b, c
} is retained in the Adams consensus, as shown in tree IV.
