rooting as a test of rooting

[James Francis Lyons-Weiler]

On Mon, 15 Sep 1997, Curtis Clark wrote:

> At 04:31 PM 9/15/97 -0500, Byron Adams wrote:
> >>At 02:15 PM 9/15/97 -0700, Richard Zander wrote:
> >>> A series of cladograms,
> >>>each depending on the one previous to establish a correct outgroup, is a
> >>>house of cards.
> >
> >Curtis Clark wrote:
> >
> >>An outgroup is not necessary to form a cladogram, only to root it.
> >
> >        I think what you meant to say was that an outgroup is not necessary
> >to form, but only to root a _phenogram_ or _dendrogram_.  Assuming one uses
> >cladistic treebuilding methodology, by definition an outgroup is required
> >(in order to establish character polarity) and the resulting tree is a
> >cladogram.
>
> See Clark, Curtis and Daniel J. Curran. 1986. Outgroup analysis, homoplasy,
> and global parsimony: A response to Maddison, Donoghue, and Maddison. Syst.
> Zool. 35:150-154.
>
> >I don't think many would argue with me that these (unrooted
> >phenograms/dendrograms) are not the same thing as cladograms.  Nor would
> >they argue that somebody who can arbitrarily polarize characters _sans_
> >outgroup comparison is actually doing cladistics.
>
> Of course not, but the point of our 1986 paper was that rooting the tree is
> algorithmically independent of finding the shortest tree, and that there
> are good reasons to keep them separate. The "series of cladograms"
> postulated by Richard is unnecessary and counterproductive.
>
>
        This issue has not yet been completely explored.  There are
        equally good reasons for exploring the effects of
        character assumptions that MIGHT be used at the same
        time as rooting the matrix, if not the tree; so keeping
        them algorithmically separate may be needed if one is
        performing rooting of trees, but is not at all desirable
        if one wants to test the consequences of making those
        assumptions, such as whether or not the outgroup taxon
        one has might mislead the tree-based analyses.

        We are using rooted rasa to choose among competing outgroup
        candidates, and have learned to distinguish between very poor
        outgroup taxa and those that are better suited to the task.
        As a result, we've learned a great deal about the utility
        of making assumptions for the purpose of testing them, and
        can choose among what appear otherwise to be equally good
        outgroup taxa.

        Hopefully the issue of constraining tree building algorithms
        will be more fully explored in this context in the future.
        For instance, maximum likelihood is conducted in the same
        way whether the tree has an outgroup or not.  This feature is
        advertised a "time-reversibility".  However, models that
        are time-reversible may not be realistic.  We know that
        trends in codon or base usage exist over time, for example.
        Whenwe root with rasa, we constrain the putative outgroup states
        of the purpose of testing the assumption of plesiomorphy.
        When the assumption of plesiomorphy is least violated,
        signal increases.  It's really quite nice; it's statistical,
        but doesn't make as many unrealistic statements about
        evolutionary processes as a model approach requires.

        There are many subtle points here; I hope no one responds
        by saying "but that's not how it's done!".  If they do,
        my response would simply be "quelle dommage".  I have even
        gone so far as to say that a cladistic maximum likelihood
        approach would be really useful in this respect.  I hear
        all sorts cringing out there, but certainly one can
        determine the log likelihood of an observed distribution
        of plesiomorphy?  the idea is to make the assumptions
        explicit, and therefore testable.  Why can't we all just
        get along :)?

        James Lyons-Weiler