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

consensus trees with branch lengths?

Doug Eernisse DEernisse at fullerton.edu
Sun Nov 21 12:03:06 EST 1999

In article <813u89$2e9 at net.bio.net>, Dianosaur <dmf_52 at hotmail.com> wrote:

> Hi all,
> We are having an ongoing discussion/disagreement in our lab about
> displaying branch lengths on consensus trees derived from a parsimony
> search, in other words displaying a consensus tree (ie strict consensus)
> as a "phylogram".  Folks in my lab (not me) sent in a manuscript with a
> "consensus phylogram" and the reviewers said not to display a consensus
> tree as a phylogram with no explanation as to what was wrong.
> I agree with the reviewers but am having a hard time articulating why.
> It's two reasons, really, that I can think of.  First, polytomies in
> consensus trees do not imply branches with zero length.  They are
> collapsed areas of multiple trees, probably with branch lengths>1.

This objection is at least partly true but I would think that
it could be pointed out in a figure legend, i.e., that branch
lengths are artifactually long immediately below polytomies
due to forced optimization of changes below the node that would
otherwise be resolved on particular minimum length trees.

> Information is lost in consensus, and it doesn't seem right to
> reintroduce that information. 

Not sure I follow this logic. One of the advantages of displaying
the phylogram version is that it nicely depicts the autapomorphies
for terminal lineages, and I don't think these should differ between
particular resolved trees and the consensus phylogram. Somebody
prove me wrong in this claim, i.e., that terminal branch lengths
are largely unaffected by the collapse.

 Second, branch lengths are character
> state optimizations across a single phylogeny. Calculating steps on a
> consensus tree just doesn't work for me.  It would mean that the
> polytomies in the consensus were true or "hard" polytomies.  

Again, it would seem that a disclaimer that depicting branch lengths
does not imply a hard polytomy would suffice.

>This is not
> what is conveyed in a consensus diagram.
> I'd love to hear any other thoughts or clarifications on this matter.
> Thanks,
> Diane Ferguson
> Michigan State University

While clarifications would be necessary in the figure legend, I 
remain to be convinced that the options would be a better solution.

They are:

1) Display a cladogram, that does not include branch length information.

2) Display an arbitrarily chosen tree among all minimum length trees,
probably with some sort of arrows showing those nodes collapsed in
a particular consensus method.

3) Display both of the above, so double the number of figures required.

4) Use one of a couple available (but unsupported by most popular
tree drawing programs) approaches that average branch lengths over
all individual optimizations.

I don't like 1 because a phylogram contains useful information about
long-branch tips or clades. Plus, cladograms are very difficult
to look at in my current trees of about >400 tips and only relatively
few polytomies.

I don't like 2 because it is arbitrary, it adds clutter, and can be 
misleading to those not used to thinking about collapsing nodes,
so some might be misled to think certain groupings are supported when
they are not.

3: Journals are already choking on big tree depictions. 

The approach of 4 might be OK but there are some who would object
to the lack of meaning of an average in this context.

I can see where this might be a problem when a consensus result is largely
unresolved but, in my experience, if there are relatively few polytomies,
one can hardly see this difference in branch length with the consensus
is compared against individual resolved trees. 

One solution that occurred to me is to standardize a new symbol in
the affected branch just below a polytomy, perhaps something like a 
small zigzag


(sorry) as opposed to a broken


line that implies that length has been deleted for convenience,
and the zigzag could then be explained in the legend.

Post additional thoughts please because I am dealing with similar
reviewers and I am not sure I am right about the terminal branch
lengths for all possible optimizations (which are not collectively
shown in phylogram depictions anyway).


Doug Eernisse
Department of Biological Science
California State University
Fullerton, CA 92834-6850 USA

More information about the Mol-evol mailing list

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