Interesting point. I'm not sure that there is an accurate way to analyse
relationships among organisms using RAPD bands. The most popular way
judging by the literature is to calculate some distances using the inverse
of Jaccard's similarity co-efficient (Jaccards *distance* co-efficient :-) )
and constructing a dendrogram using the UPGMA method.
I think this method is popular because of (a) historical reasons and (b)
the distance calculations have a small variance.
Obviously constraining the tree by the UPGMA method may be inefficient if there
are unequal rates of evolution, but this is not the only problem.
A modification of the method of Nei and Li (1979) was published recently in
PCR methods and applications (about 6 months - 1 year ago) that *works* in
simulation studies. The unfortunate thing about these simulations is that the
assumptions implicit in the method are incorporated into the simulation
so the simulation is BOUND to work. (sorry the paper I've just quoted is:
Clark, A.G. and Lanigan, C.M.S. (1993) Mol. Biol. Evol. 10 (5): 1096-1111
Prospects for estimating nucleotide divergence with RAPDs.)
The distance measure that they use appears to be a *good* method with
appropriate underlying assumptions. Personally I think that using
a distance metric such as this and perhaps representing the relationships
in a split decomposition graph is as close as you might come to being
accurate concerning these phylogenies.
This is not a gospel, the best way for representing phylogenetic relationships
using RAPDs has not yet been universally agreed upon.
My $0.02 worth. Would appreciate a discussion.