DNA ML Questions

Joe Felsenstein joe at evolution.genetics.washington.edu
Mon May 1 23:07:42 EST 1995


In article <v01510100abcb00148c70@[140.254.12.85]>,
Robert Rumpf <rumpf.1 at osu.edu> wrote:
>We've been using Fast DNAml to do some phylogenetic trees and have some
>general questions about ML algorithms in general:  Is it necessary to
>bootstrap ML trees, since the ML algorithm is in itself a statistical
>procedure?

No, if you really believe the model.  Yes, if you think there may be genuine
heterogeneity among sites in rates of evolution.  Also, yes, even if the model
is believed, actually, since there is no formal test of the likelihood
difference between tree topologies.  The Hidden Markov Model rate variation
method we have in DNAML makes the model more believable -- I think it is not
yet incorporated in fastDMAml.

> Also, should we expect bootstrap values to be inherently lower
>with the ML algorithm (as opposed to parsimony or distance methods)?

Offhand, no.

>We've run trees on a various data sets of 37 taxa with over 2000
>characters and found that the ML algorithm consistently gives us MUCH
>lower bootstrap values...

Possible explanations would be biases due to long-branch-attraction
(the you-know-who Zone) in parsimony, or due to incorrect correction for
multiple hits in distance methods.  Either might make there appear to
be evidence for some groups in these methods that did not appear in
ML (thought the latter might appear there too).

-----
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



More information about the Mol-evol mailing list