pls3b

Comparing and Contrasting Methodologies

G.F. Guala

Distance/Phenetic Methods

The assumption is that the more alike organisms are, the more closely related they are.

If every "state" is recorded and those states are statistically compared by any method that groups soley on the numbers of states shared between organisms, then there is an inherent assumption of equal rates of evolution because every state is treated as every other and thus must be evolving equally to be compared equally in any equation. This is both the strength and the inherent weakness of simple molecular clock algorithms as well.

Physical pairwise distance comparison techniques, the most famous of which was DNA hybridization are historical now.

PCA

Principal Components Analyses

Standardize (usually -mean then /SD)
Draw a line through the centroid that minimizes the square of the distance of each point to that line.
Draw a second orthogonal line to that one through the centroid with the same constraint
and so on.....
All eigenvalues (the expression that the computer generates to describe the vector of each axis) will add up to the number of variables so you can just take their percentage of the whole to give you the amount of variation described by each.

This method is not really appropriate for phylogenetic analyses for the reasons described above but it is very useful as a way to look at very similar taxa and to parse out groups within and among them for further study and to find new charateristics to look at. It is very useful when trying to define OTUs for a phylogenetic analysis.

UPGMA/WPGMA
(un)weighted pair group method using arithmetic averages

Generate a matrix of pairwise distances.
Think of every OTU as a cluster.
Merge the two most similar clusters and recalculate the matrix with them as a single cluster.
Repeat until there are no more OTUs (thus all are now in a series of hierarchical clusters that you have been keeping track of as you merge them). This involves avergeing the distance from the individual OTUs in each cluster to all others. If you use the simple averages of the distances to each component OTU dived by 2, then you are doing WPGMA and if you detrend that by inserting the number of taxa in each cluster in the equation, you are doing UPGMA.

These trees are fast to compute and provide a useful representation of the data (i.e. they might be useful in telling you if you need to sequence more in a big molecular study) but they are not useful for generating phylogenies for the reasons discussed above.

Here is a link to a very brief example by Steven Carr.

Comparing and Contrasting Methodologies

Distance/Phenetic Methods

NEXT