I am currently doing some analysis on rRNA sequences from some archaea. Anyone
who has worked with archaeal rRNAs knows that there are some compositional
biases in the data. In an attempt to overcome this problem I've recoded
the nucleotides as G's and C's (i.e. only transversions are analysed A=G and
I want to do some maximum likelihood analysis on the data.
I ran the data through Gary Olsen's fastDNAml program and the resulting
tree had topology, but the branch lengths were at infinity. Running the same
data through Joe Felsenstein's DNAml prog. gave branch lengths O.K.
I don't have Gary Olsens Email address to ask him this question directly, so
maybe somebody out there might have the answer to why this is so. Also, is
there a way around this problem? Joe Felsenstein's advice, although I haven't
tried it yet, is to code the data as purine (U) and Pyrimidine (Y) and define
the base frequencies.
Any suggestions greatfully appreciated.