ABSTRACT
The trees generated from a particular number of taxa are distributed in three-dimensional tree space, in which trees can be grouped into islands with optimality relationships. A computationally rapid search such as neighbor joining can be used to select a starting tree within a fairly optimal location in tree space. Subsequent likelihood or parsimony analysis on this NJ starting tree may produce higher optimality. A locally optimal location in tree space can be reached by use of branch swapping methods such as nearest-neighbor interchange, subtree pruning and regrafting, and tree bisection and reconnection. The most critical step in exploring tree space is to determine whether there is another peak that reflects a more parsimonious or more likely solution, rather than remaining in a local maximum peak. The ratchet technique is an efficient tool to accelerate this search. Rate heterogeneity can be incorporated into likelihood models by use of a γ distribution. Simple metrics can be used as measures of consistency in a parsimony analysis. These metrics include tree length, consistency index, retention index, rescaled consistency index, and Bremer or decay index. The robustness of nodes on a phylogenetic tree can be analyzed by resampling techniques such as bootstrapping and jackknifing. A distribution model can be added in a technique known as parametric bootstrapping.
