William R. Pearson
wrp at dayhoff.med.Virginia.EDU
Sun May 1 21:33:50 EST 1994
In article <Cp4r05.H2o at gpu.utcc.utoronto.ca>,
L.A. Moran <lamoran at gpu.utcc.utoronto.ca> wrote:
>William R. Pearson <wrp at dayhoff.med.Virginia.EDU> wrote:
>>Distance methods have the great strength that they do not require
>>multiple alignments; parsimony and maximum likelihood do. The
>>difficulties encounted in the multiple alignments are often ignored
>>when comparing the different methods.
>Distance methods SHOULD require multiple alignments and parsimony and
>maximum likelihood methods SHOULD accommodate user-defined multiple
>alignments. I do not think that it is a "great strength" of the distance
>methods that they can be used with pairwise alignments - why do you
>believe that this is a strength?
The multiple alignment problem for evolutionary trees is a
very difficult one that is unlikely to be solved soon. User-defined
multiple alignments certainly do not help - an unsuspecting user may
define a sub-optimal multiple alignment that prefers a certain tree
topology. The nice thing about pairwise alignments is that you can
know that they are optimal for the scoring parameters. For problems
of interesting size, this is virtually never the case with multiple
So, although one can make strong philosophical arguments about
why "distance" does not necessarily imply an accurate phylogeny, I
feel uncomfortable with methods that require a multiple alignment,
since I know that that foundation is almost certainly flawed for
interesting problems. One must choose between the frying pan and the
fire - would you rather use a method that doesn't reflect exactly what
you want but relies on pretty robust premises (distance), or use a
method that would tell you exactly what you wanted IFF your multiple
alignment is correct, which is practically never the case.
Neither method is without fundamental flaws.
More information about the Bioforum