When it comes to a command; tree.shared, I understand that Mothur generates a tree by clustering groups using the UPGMA algorithm using the distance between communities as calculated using any of the calculators describing the similarity in community membership or structure.
Does this mean that the UPGMA algorithm is the best option for this purpose?
There are several tree building algorithms, such as maximum likelyhood and parsimony.
I would like to know whether we can use other tree building algorithms for this purpose or the UPGMA algorithm is the best for tree building in this case.
I appreciate your help.
I had this question a while back, and here is what I have settled on so far (but it doesn’t mean I’m right)…first, we are building trees from a simple distance matrix. Granted the distances can be made by comparing different “community” comparison estimators (Jest, Jclass, ThetaYC, etc.) instead of a nucleotide sequence. Without the additional information provided in the nucleotide sequences, you can’t justify using likelyhood models and and parsimony (which assumes an ancestral state of a sequence).
If you are familiar with PAUP, you know that it won’t allow you to build any type of likelyhood or parsimony tree without the nucleotide sequences. If you provide it with a distance matrix, you are allowed to build a UPGMA tree, or a NJ tree.
Now, would a NJ tree be appropriate in building community trees? Not sure. While answering this post, I found out that UPGMA assumes a molecular clock, where NJ does not. If you want to try it out, you can import a community distance matrix MOTHUR will spit out by using dist.shared() into PAUP or NEIGHBOR from the Phylip package
Hope this helps.