New version of DAMBE

Xia Xuhua xxia at hkusub.hku.hk
Sun Aug 1 20:56:48 EST 1999


Dear All,

I have just uploaded a new version of DAMBE to my URL site:

http://web.hku.hk/~xxia/software/software.htm

One particular enhancement that I wish to highlight is the statistical
tests of alternative phylogenetic hypotheses by using distance
(Fitch-Margoliash method), maximum parsimony, and maximum likelihood
methods. In short, you provide alternative phylogenetic hypotheses
represented by different tree topologies in a file in PHYLIP format, and
DAMBE will evaluate relative statistical support for each of the topologies
relative to the best. It takes just a few mouse clicks to perform these
tests.

Here is a brief rational of the tests:

1. For Fitch-Margoliash method: the test is based on the fit to the
distance matrix. DAMBE calculates the distance matrix based on the distance
you specify, evaluates tree branch lengths for each of the UNROOTED
topologies in the user tree file (or all possible topologies generated in
DAMBE), and obtains the pairwise distances based on the estimated tree
branch lengths. The differences between these pair-wise distances and those
in the original distance matrix are them calculated, so are the variance of
the differences. If the differences are all zero (perfect fit), then the
variance is zero. If the fit is poor, then you have a large variance. If
you have 10 user trees, then you have 10 variances, and the best tree is
the one with the smallest variance. A Bartlett's test of homogeneity of
variance is then carried out as an overall test. If the test is not
significant, then you may stop there and conclude that the best tree is not
significantly better than others. If the test is significant, then you can
use multiple comparisons to test whether each alternative topology is worse
than the best tree. To adjust for the experimentwise error rate associated
with multiple comparisons, a test analogous to Newman-Keuls test (which is
based on the same statistic as Tukey's HSD test) is applied with q
calculated as 

q = (ln(VarA)-ln(VarB))/SE

The main weakness of the test is that the distances in the distance matrix
are not statistically independent of each other. The main advantage of the
test is that it should be very powerful in rejecting poor topologies. The
qualify of the test, of course, depends entirely on how appropriate the
genetic distances are.

2. For the maximum parsimony method, a ROOTED tree is required to represent
alternative topologies. This limitation arises from DAMBE's inheritance of
the code from DNAPARS in Joe's PHYLIP. DNAPARS has already provided a
significance test if you provide user trees. In short, it computes the
number of steps (changes in character states) for each topology, the
difference in the number of steps between the best and each alternative
topology, and the associated (large-sample) variance of the differences.
The z-score is computed and declared as significant if it is larger than
1.96. The main problem with this test is that the result can be interpreted
probabilistically only when you have just two topologies and is not
appropriate with multiple comparisons. DAMBE uses Newman-Keuls test which
is valid for multiple comparisons.

3. For the maximum likelihod method, the Kishino-Hasegawa test (RELL) is
implemented as in PAML whose code I have used, again with the adjustment
for multiple comparisons. The Kishino-Hasegawa test, as is practiced in
literature, is analogous to the test in DNAPARS, except that the test is
based on likelihood instead on the number of steps. In short, you calculate
the maximum likelihood for each topology, the difference in likelihood
between the best tree and each alternative topologies, and the variance of
the differences estimated by resampling. The z-score is then calculated and
declared as significant if it is larger than 1.96. Again, such
interpretation is heuristic and is not appropriate probabilistically if
there are more than two topologies being compared. DAMBE uses Newman-Keuls
test which is valid for multiple comparisons.

To perform one of these tests in DAMBE, you open a sequence file with
aligned sequences. Click PHYLOGENETICS menu and then one of the three
submenus: (DISTANCE, MAXIMUM PARSIMONY or MAXIMUM LIKELIHOOD). When the
dialog box appears, click the USERTREE option. A standard file OPEN|SAVE
dialog box appears for you to choose the file containing user trees in
PHYLIP format (alternatively you may click the option for all possible
trees if the number of OTUs are small, e.g., 5 or 6). Click whatever other
options according to your intuition, and then the DONE button. DAMBE will
carry out the test (It may take a long time with the maximum likelihood
method).

Best.
Xuhua
==================================================================
Dr. Xuhua Xia                       | Tel: (852) 2857 8239 (lab)
Dept. Ecol. & Biod.                 | Tel: (852) 2975 5629 (office)
Rm 603, T.T. Tsui Building          | Fax: (852) 2517 6082
The University of Hong Kong         | Email: xxia at hkusua.hku.hk
Pukfulam Road                       | WWW: http://web.hku.hk/~xxia
Hong Kong                           |
==================================================================




More information about the Bio-soft mailing list