2. PREVIOUS AND RELATEDWORK
Several consensus tree methods have already been proposed,
see [3] for a good overview and comparison. Unfortunately,
most methods have the drawback of being relatively
strict, e.g. restricting the consensus tree to common
substructures, and that the used tree metric is often coarsegrained,
nally producing a quite poorly resolved or less
intuitive solution tree. Prominent examples are the strict
and majority consensus methods operating on clusters. A
cluster is a subset of the set of taxa which contains all the
descendants of its most recent common ancestor. The strict
consensus method only retains clusters common to all input
trees and the majority consensus method those appearing in
more than half of them. The latter method can be regarded
as a median method minimizing the number of non-common
clusters, i.e. minimizing with respect to the symmetric dis-
tance metric. Further to mention is that the classical methods
do not make use of any sophisticated search procedures
and rely, if at all, on simple greedy approaches (e.g. the
greedy consensus tree method available in PHYLIP [9]).
A recently proposed tree similarity measure, the Tree-
Rank measure [23], originally introduced to handle database
queries for similar trees1, allows for more sophisticated procedures
due to its ne granularity. This measure utilizes the
quadratic Up matrix U which states for each pair of taxa
(a; b) the number U[a; b] of necessary up-traversals to reach
from taxon a the least common ancestor of both taxa; see
Figure 2 for an example. It can be derived in O(jLj2) [23].
The authors also dened the Down matrix D in an analogous
way, but since U = DT it is redundant and the Up
matrix is also called UpDown matrix. Having this matrix
for two trees T1 and T2 and assuming equal taxa sets, one
can calculate the UpDown distance between them by
2. PREVIOUS AND RELATEDWORKSeveral consensus tree methods have already been proposed,see [3] for a good overview and comparison. Unfortunately,most methods have the drawback of being relativelystrict, e.g. restricting the consensus tree to commonsubstructures, and that the used tree metric is often coarsegrained, nally producing a quite poorly resolved or lessintuitive solution tree. Prominent examples are the strictand majority consensus methods operating on clusters. Acluster is a subset of the set of taxa which contains all thedescendants of its most recent common ancestor. The strictconsensus method only retains clusters common to all inputtrees and the majority consensus method those appearing inmore than half of them. The latter method can be regardedas a median method minimizing the number of non-commonclusters, i.e. minimizing with respect to the symmetric dis-tance metric. Further to mention is that the classical methodsdo not make use of any sophisticated search proceduresand rely, if at all, on simple greedy approaches (e.g. thegreedy consensus tree method available in PHYLIP [9]).A recently proposed tree similarity measure, the Tree-Rank measure [23], originally introduced to handle databasequeries for similar trees1, allows for more sophisticated proceduresdue to its ne granularity. This measure utilizes thequadratic Up matrix U which states for each pair of taxa(a; b) the number U[a; b] of necessary up-traversals to reachfrom taxon a the least common ancestor of both taxa; seeFigure 2 for an example. It can be derived in O(jLj2) [23].The authors also de ned the Down matrix D in an analogousway, but since U = DT it is redundant and the Upmatrix is also called UpDown matrix. Having this matrixfor two trees T1 and T2 and assuming equal taxa sets, onecan calculate the UpDown distance between them by
การแปล กรุณารอสักครู่..
