Kimura alignment

Mark Siddall mes at zoo.toronto.edu
Fri Feb 3 05:39:08 EST 1995


In article <3gok0i$j3n at news.duke.edu> histone at acpub.duke.edu (Ronald DeBry) writes:
>You've hit on the real problem with this last sentance.  Any multiple
>alignment should really be conditional on the evolutionary relationships
>(whether you are using a Kimura substitution model or not).  Two kinds

I do not know thatI agree.  If you are using a Kimura model it must be for
the reasons outlined in my first post.  If you are using a Jukes-Cantor model
you need not since one is only minimizing the cost of any kind of change.
Inasmuch as CLUSTAL builds a F-M phenogram to align, MALIGN does not.

>of solutions have been proposed.  One can use an iterative reciprocal
>process, where a trial alignment is used to give a trial phylogeny,

Ah.  But will this not bias your final answer towards some tree(s)
or islands of trees.

>which is used to generate a new alignment then a new phylogeny and so on
>until it converges on a stable alignment/phylogeny.  This approach has
>been implemented as a computer program by Jotun Hein.  The on;y
>alternative that I know of is a series of papers by Rich Thorne.  In his
>method (essentially) all possible alignments are generated.  Each

ALL????  I'm afraid I like to get alignment answers within the week!

>alignment is given a likelihood based on a model incorporating both
>insertion/deletion and substitution, and a distance matrix is calculated
>by weighting each alignment based on its likelihood.  The phylogeny is

But to infer a likelihood estimation you need a model.  I know this
is another thread but I have problems with this too.  For example, 
although a Kimura 2-param model may be appropriate in one region of the
18S rDNA gene (for example) I doubt it is all the way across.  So far
I don't know of a practical way to a) decide where a priori or b) 
implement such a procedure.

>then inferred from that weighted distance matrix.
>
>As someone who subscribes to the idea that phylogenetic inference is a
>statistical problem and that maximum likelihood in some form is the best

I don't but that too is another issue.

>approach, I am very interested in Thorne's papers.  However, it is
>obviously very computer-intensive, and I am not aware of any practical
>applications of his methods yet.

I will look at them.  In the meantime, I think it is best to just align 
in a 1-parameter way and stand as unconvinced that I should do
otherwise.

Mark

-- 
Mark E. Siddall                "I don't mind a parasite...
mes at vims.edu                    I object to a cut-rate one" 
Virginia Inst. Marine Sci.                     - Rick
Gloucester Point, VA, 23062



More information about the Mol-evol mailing list