Why Aren'T Guide Trees (Necessarily) Phylogenetic Trees?

Question

[Click downvote]

As the title says...

Answer 1

[Hamish]

Assuming that you mean the guide tree produced by multiple sequence alignment programs, such as ClustalW, which use a tree guided progressive alignment method. In these methods an initial step performs a quick set of sequence comparisons in order to create a hierarchical clustering of the sequences (the guide tree). The guide tree is then used to determine in which order sequences are added to the alignment. Since the guide tree is often a bit rough and ready, and does not necessarily reflect the multiple sequence alignment produced by the method, it isn't really suitable to use as a phylogentic tree. Deriving a phylogentic tree from the multiple sequence alignment (assuming that that alignment makes sense) is more robust, and allows you to choose a appropriate phylogeny method for the data.

That said, if you only want to get a rough idea of the possible relationships between the sequences, then the guide tree may be sufficient.

[Click downvote]

So progressive vs. optimal alignment is one key factor then?

[lh3]

Guide tree is only used for progressive alignment to guide how sequences are added. Optimal multiple alignment does not need a guide tree at all, but it is NP-hard, so no programs use that. Earlier versions of clustalw computes distance from all-vs-all pairwise alignment and then uses neighbor-joining to build guide tree. Because pairwise alignment is usually less accurate, the tree is not reliable.

[Hamish]

There are a number of groups working on optimal multiple sequence alignment methods (see http://www.google.com/search?q=optimal+multiple+sequence+alignment). The are implementations available, but these are commonly limited for small alignments, for example see http://europepmc.org/abstract/MED/2734293.

Answer 2

[Larry_Parnell]

Generally, a guide tree won't (necessarily) use parsimony in the construction of that tree. Parsimony is key to phylogeny.

[David W]

Parsimony isn't the key to phylogeny -- it's just one way to estimate a phylogeny. The clustering methods usually used to make a guide tree aren't parismony methods, but they are attempts to estimate a tree. We shouldn't use them to make biological inferences - but they _are_ phylogenetic trees.

[Josh Herr]

+1, and in agreement with what Larry_Parnell said. Guide trees are often constructed (sometimes manually or even randomly) to represent a testable hypothesis: create a tree to guide an analysis and then see how a phylogenetically produced tree fits the guide tree.