Joe Felsenstein wrote:
>The best way is to fit the tree using maximum likelihood, at a given
>transition/transversion ratio (which you feed in as a parameter of the
>Then record the likelihood. Do it at other Ts/Tv ratios, and record.
>The maximum likelihood estimate of the Ts/Tv ratiion is the one that
>achieves the highest likelihood.
>This can be done by a number of runs of my program DNAML or by a single
>run of David Swofford's forthcoming PAUP*.
>This approach solves the problems of deciding among alternative
>numbers of transitions and transversions that would be needed to
>explain a given set of nucleotides at one site, without any arbitrary
The only thing I have to add to what Joe said is that Ts/Tv ratio can be
underestimated if site-to-site rate hetergeneity is ignored. This effect
can be trivial if neither parameter (Ts/Tv and some measure of among-site
rate variation, say gamma shape parameter) is extreme, but can be rather
strong if either or both are. Also, when both are strong, the effect of
topology on estimates of Ts/Tv is exacerbated. John Wakeley wrote an
excellent review of these factors in a recent TREE review on estmation of
transition bias (v.11, #4).
Jack Sullivan e-mail: sullivan at onyx.si.edu
Laboratory of Molecular Systematics Phone: (301) 238-3444
MSC, MRC-534 Fax: 301-238-3059
Washington, D.C. 20560