relative rate test program

Joe Felsenstein joe at evolution.genetics.washington.edu
Fri Jul 12 13:17:10 EST 1996


In article <v01530500ae0c2bcf25ca@[192.107.185.84]>,
Michael Nedbal <nedbal at FMPPR.FMNH.ORG> wrote:
>Does anyone know of a program that will perform relative rate tests among
>pairwise comparisons of taxa (nucleotide sequence data)?

Once one has many species there is the sticky issue of the tests among
different triples of species being correlated.  (You have to use triples,
not pairs).

The general way to do the RRT is to do a likelihood ratio test (LRT) of the
proposition that the tree is clocklike.  This is done by computing the
likelihood of an n-species tree without a clock assumed, and also with the
clock assumed (note that the tree topologies have to be the same, considered
as unrooted topologies).

The former estimates 2n-3 branch lengths, the latter n-1 ancestral node
times.  The difference in parameters being n-2, that is the degrees of
freedom for the likelihood ratio test.  Double the difference in log
likelihoods between the two trees, and look it up in a chi-square table
with n-2 degrees of freedom.  You will find this explained in my 1988
Annual Review of Genetics paper.

These likelihoods can be computed using the DNAML and DNAMLK programs in
PHYLIP.  You will also want to explore the options that deal with unequal
rates of change at different sites when you do that.

-- 
Joe Felsenstein         joe at genetics.washington.edu     (IP No. 128.95.12.41)
 Dept. of Genetics, Univ. of Washington, Box 357360, Seattle, WA 98195-7360 USA



More information about the Mol-evol mailing list