Major Methods for Estimating Phylogenetic Trees

Today, I worked through Chapter 5 in my book, which gives a good overview of the major methods for estimating phylogenetic trees.

There are two primary approaches to tree estimation: algorithmic and tree-searching. The algorithmic approach uses an algorithm to estimate a tree from the data. The tree-searching method estimates many trees, then uses some criterion to decide which is the best tree.

The algorithmic approach has two advantages. It is fast, and it yields only a single tree from any given data set. The Neighbor Joining method is the most common algorithmic method, and I'll be learning how to use it next week.

All the other currently used approaches are tree-searching methods. They are generally slower, and some will produce several equally good trees. Methods such as Parsimony, Maximum Likelihood, and Bayesian analysis search for the tree that best meets the criteria by evaluating individual trees. Maximum Likelihood looks for the tree that, under some model of evolution, maximizes the likelihood of observing the data. Bayesian Inference is a recent variant of Maximum Likelihood. Instead of seeking the tree that maximizes the likelihood of observing the data, it seeks those trees with the greatest likelihoods given the data, and produces a set of trees with roughly equal likelihoods. Parsimony is the simplest method, and it looks for the tree or trees with the minimum number of changes.

 It's almost impossible to evaluate each possible tree because of the sheer number of possibilities (even with only 10 taxa, there are more than 34 million rooted trees). Therefore, something called a "branch-addition algorithm" is used to find each of the possible trees, which I won't go into detail here.

It's important to realize that since we don't know what happened in the past, we can never be entirely sure how accurate the tree is. In addition, there is no "right" tree-- we can only hope to find the tree that most closely approximates what happened in the past.