[Taxacom] Why character-tracking doesn't happen?
mesibov at southcom.com.au
Sat Sep 13 19:04:44 CDT 2008
Unfortunately I won't be able to follow this discussion for the next few
days, but it's a great read so far.
Neil Bell wrote:
"In fact something like what you are suggesting happens informally all
the time. Any tree immediately reveals homoplasies of particular
morphological features to anyone familiar with the organisms. Some of
these will be more credible than others, and this will act as a spur
further research (e.g. further sampling) or critical examination of the
data or methodology. Some people seem to think that such an "iterative"
approach is unscientific, but like any search strategy I think it isn't
how you get there that matters, more how you evaluate your postion once
you are there."
Hmmm. 'Informally'? Can you cite any papers in which trees were
re-evaluated based on the credibility of their inferred homoplasies?
Curtis Clark wrote:
"I'm not convinced after all these years that I was on the right track,
but the issues seemed obvious even then."
Nice presentation, and slide no. 1 seems to be what Neil means by
'iterative' (above), so the tautology angle is spot on.
I notice, though, that your ITS+morphology analysis used combined data.
This raises an issue I've been afraid to ask about molecular data,
but... Here goes, the worst that can happen is that someone
knowledgeable will call me an idiot.
When you load a morphology matrix, you try not to include characters
that are non-independent. A few shouldn't hurt the analysis (they'll
keep appearing together as 'simultaneous' changes between nodes), but a
lot might swamp other signals with their noise. Independent characters
When you infer phylogenies based on nucleic acid sequences, you assume
that each position is independent, unless you're dealing with a
protein-coding sequence, in which case you know that the third, 'wobble'
position in a triad has some correlation with the first two positions,
and you can correct for this. So the independence assumption should be
OK for non-coding sequences?
Well, no. The secondary structure of things like ribosomes, for example,
is strongly conserved. This means that some substitutions in rDNA will
require 'compensatory' substitutions elsewhere in the sequence, which
means there's non-independence. There are ways to test for
non-independence. For example, I've seen published work on
autocorrelation and Fourier decomposition analyses of whole genomes
which show periodicities in sequences forced by things like how
chromosomes are put together (in eukaryotes) and how strands twist (in
prokaryotes). But how do you do such tests on shorter sequences,
especially when there are indels - interruptions scattered here and
I used to think that each position in a nucleic acid sequence was a
character. In fact, this is not how molecular phylogeny programs work. A
multiple sequence alignment program first does its best to find matching
patterns around indels. The five character states then analysed (A,T,G,C
and missing) are states of an artificially created character - a
vertical column in a multiple sequence alignment. This 'column'
character varies from analysis to analysis, depending on which sequences
are used. How can you test the independence of such things? Aren't all
the columns covering an indel really a single entity, so that the A,T,G
and C's in the 'non-indel' sequences at this point are non-independent?
Dr Robert Mesibov
Honorary Research Associate
Queen Victoria Museum and Art Gallery and
School of Zoology, University of Tasmania
Home contact: PO Box 101, Penguin, Tasmania, Australia 7316
(03) 64371195; 61 3 64371195
More information about the Taxacom