IUBio Biosequences .. Software .. Molbio soft .. Network News .. FTP


jamesmc at acer.gen.tcd.ie jamesmc at acer.gen.tcd.ie
Mon Aug 28 07:01:15 EST 1995

In response to Mark Siddall's posting

>>Briefly, the argument against parsimony is that the first thing you do is tak
>>the majority of the characters in your data set and throw them out.  Also the
>Misrepresenation #1.  We do not throw any characters out.  I presume

yes you do!

>that what you are referring to is that characters that are invariant
>or that are different for only one taxon are phylogenetically 
>uninformative.  This does not mean we throw them out.  Go ahead and leave them
>in for all I care.  It will not change the answer.  The point is that

The point is that many characters can go unused in cladistic analysis.

>if a charater has not changed... there has been NO evolution!  If you'd
>care to re-define evolution then go ahead.  Until then, evolution is
>about change (no?) and phylogeny reconstruction, is as well.

O.K. take a hypothetical gene from four taxa. 

species1  GATC
species2  GTTC
species3  GTTA
species4  GTCC

So change has occured in the EVOLUTION of this gene. Using a distance method,
this gene sequence can be used for the purposes of tree construction, using
parsimony all of these characters are uninformative, but hold it a second...
hasn't there been change here?? (no?).

I am merely pointing out that a problem some people have with parsimony is 
that often many characters are ignored.

>>is no statistical framework built into the parsimony procedure, so should you
>>accept the most parsimonious tree as the one and only, with no idea of how go
>>or bad this tree really is (what about multiple equally parsimonious trees?)?
>I'm not sure that this was a complete sentence but I think I get your
>drift. No there is not a statistical framework built in.  But then
>we hold that the phylogenetic history of a set of taxa is a historical 
>singularity, not a population that can be sampled.  That is, it occurred once.
>I have said much more on this elsewhere.

Yes obviously the phylogenetic history of a set of taxa is unique.  
Reconstructing that history is not an easy task.  As sequence divergence 
increases and multiple hits begin to accumulate, there are mathematical 
methods of attempting to correct this (Nei has published his findings on when
a particular correction is suitable). According to the proponents of distance
matrix methods, this is the real strength of the method, and the weakness that
is sited for parsimony.  On the other hand parsimony methods have statistics
such as retention- and homoplasy-indices.

>>The argument against distance based methods (in addition to the ones mentione
>>above by M. Siddall) is that you use some characters in a negative way (an
>>analogy: mushrooms, E. coli and the oak tree are joined together by virtue of
>>not possessing vertebrae), with cladistic methods, organisms are put together
>>only if they share characters that have arisen relatively recently
>>(shared-derived characters).
>Misrepresentation #2.  We do not only concern ourselves with changes that are
>relatively recent.  We do concern ourselves with changes however.

Notice my use of the word *relatively*.  Is cladistics not based on the 
principle that taxa are united by the presence of shared-derived characters?
And these character states are more recent than the ancestral state?

>That is, we concern ourselves with evolution.  That jellyfish, ants
>and Parmecium spp. lack a backbone has nothing to do with evolution.
>That is, nothing has happened to give them each this characteristic...
>more to the point... the only reference point for the question is
>that vertebrates have  had some thing happen in their 
>history.  Again... this is about evolution... about change.

Despite your best efforts, I think you are agreeing with my summation of 
cladistics  here.

>>Now for my humble opinion...A distance method can be very good (as we see fro
>>simulation studies) at retrieving the correct tree, especially if one chooses
>Oh?  And how do you know when you have the correct tree?
>>the proper correction for multiple hits.  When a good tree building method li
>>the neighbor joining method is used then the chances of retrieving the correc
>Neighbor Joining is NOT a good tree building method.  If you change the order 
>input of taxa, you change the tree.  Ummm... sound like a problem?
>If you replicate a taxon you change the tree elsewhere.  
>There are manmy other problems.

There have been some other postings on this point.  But people in glasshouses
should not throw stones.  With more than 8 or 9 taxa it is necessary to use
a heuristic search procedure to find the most parsimonious tree.  Change
the input order and the tree changes (didn't I just hear that this was a 
problem with NJ?).

>>tree are pretty o.k.
>Again I ask how you know when you got the correct tree?????

In the context of the original question (which is better etc...) I was giving
perhaps an oversimplified answer and *correct* may have been too strong a word.
I am simply reiterating the results of many simulation studies and NJ is said
to be robust.

>Does the hand of Providence come down and put a gold star on your 

Sarcasm may be the tool you use to conduct a discussion but lets try not to let
this one degenerate to this level.  This question is beneath comment.

>>Parsimony is really neat for examining how evolution has occurred, which

What???   ;)

>>characters are leading to confusion and instability, what is the general mode
>>of evolution (transformation from one character state to another).
>>The philosophy of the investigator is personal, but as a bench scientist who
>>cares more about the result and the hypotheses I can make from my data than
>>any philosophical do's and don'ts, my advice is to use both.  Agreement is go
>Agrement between two or more bogus methods means diddly.

Well Mark, if I calculate distance matrices from my data, draw NJ trees, 
bootstrap the data and find the taxa of interest are joined 95% of the time
and I get the same result using parsimony (i.e. 95%) I am happy and I go for
a few beers.  I don't think you are going to change my mind.  

Look at the literature...how many scientists publish trees by one method and 
indicate (in the methods section)that they also examined the data by the other 
method and if it agreed then they seem to be happy.  The journal editors also 
seem to be happy to publish this.

As you said at the outset, your responce was philosophically biased, but alas
I fear it was also a misrepresentation of some of the realities.


More information about the Mol-evol mailing list

Send comments to us at biosci-help [At] net.bio.net