Program for testing incongruences

Andrew J. Roger aroger at
Thu May 9 13:06:01 EST 1996

Thomas K. Dibenedetto wrote:

> Fine, I go along with all this. It is true that we often have knowlege of
> relationships at higher and lower levels than what we are studying, and I
> am not averse to using that knowledge. It just strikes me that most of
> the problematical situations (lat. transfers) are either at a pretty gross
> level (not really in need of a statistical significance test), or may be
> operative at the (species) level in question (the level of relationships
> which one is attempting to determine), and thus not the ones we can easily
> distinguish.

This could be. I have dealt with datasets, however, which show
complete lack of phylogenetic resolution. In cases like these, you
often see ridiculous optimal topologies. However, if you impose
a sensible topology on the dataset, the number of steps hardly
increases....and if one applies a Templeton test (as an example of a
statistical test) one sees that the ridiculous topology is
not significantly worse than a sensible one. If such a dataset
were being combined with another it would be useful to find out
if the ridiculous topology it yields is strongly or weakly in conflict
with the second dataset. This is where combinability tests would be useful.

> : >
> : >Would you care to point me toward those
> : >sources of evidence which have led you to conclude
> : >that "anomolous" gene trees are anywhere near as common as would be
> : >necessary to upset a total evidence analysis?
> : They are probably not very common in a lot of cases. There is one
> : fundamental case where they might be. Recently Golding and Gupta
> : published a paperin Molecular Biology and Evolution where they
> : described comparisons of eukaryote/archaebacteria/gram positive/
> : gram negative homologs of roughly 2 dozen genes. They found that
> : about 1/3 of the genes gave a significantly different relatioship of
> : these groups to roughly 1/2 of the dataset. The rest of the genes
> : did not resolve relationships very well.
> thanks, I'll check it out. (did they explain this with "chimaerism"?)

Yes. However, his paper, in fact, is not as solid as it may seem because
they drastically undersampled the taxanomic diversity in all of the
gene families. So, for instance, paralogy is confused with lateral transfer
(to mention only one of the shortcomings of the analysis). Nevertheless,
there are some clear cases of radical lateral tranfer (transfer between
unrelated prokaryotic groups and between prokaryotes and eukaryotes)-
genes such as glutamine synthetase (GS) and dozens (if not hundreds) of
genes transferred after the endosymbiotic origin of eukaryotic organelles. 
> : However, the point is
> : that the fundamental relationships between eukaryotes, archaebacteria
> : and eubacterial groups are unknown and the data significantly
> : conflict in their answers.  Total evidence on this dataset
> : would mean nothing.
> Do you mean that total evidence would _indicate_ nothing? If so, i guess
> I would say Good! Total evidence does not guarantee an answer, nor does
> it (most significantly) give you the false security of an answer when in
> fact the evidence is ambiguous. It is a "method" which derives from the
> rather simple principle that there really is no good reason in science to
> ignore evidence relevant to a particular question.

I mean that total evidence may yield a topology that would in no
sense describe the history of any of the genes (nor the organisms). The 
problem is that there is always a maximum parsimony tree (or set of trees).
It is possible that the consensus of these trees would yield no resolution
of the tree- this would maybe allow you to infer that the data are mutually
contradicting each other. But it is also possible that there is one MP tree.
In this case you would have no idea that your dataset is partitioned
into two mutually exclusive alternatives....and it may be the case that
the MP tree accurately reflects neither of the alternatives because of the wierdly
conflicting data that is fed into the analysis.

I guess that I'd rather know in advance that my data was partitioned into
two strongly contradictory sets of genes than risking not finding this out
at all by combining the data and taking the total evidence result at face

	Andrew J. Roger				ph: (902) 494 3569
	Dept. of Biochemistry,		FAX: (902) 494 1355
	Dalhousie University,		email: aroger at
	Halifax, N.S.
	Canada, B3H 4H7

More information about the Bio-soft mailing list