parsimony/biology

[Don Colless]

Zdenik Skala wrote:
 The
>nearest-neighbor cluster analysis is parsimonious as well, yet in another
>sense and, of course, applied to phylogenetic data gives quite different
>solutions.

Curiously, the nearest neighbour algorithm is, **in the absence of
homoplasy**, quite likely to give the same result as any effective
phylogenetic method (see my analysis in Sys. Zool. 19:352). It is not, of
course, guaranteed to do so; and with more realistic assumptions about the
quality of data, a pair-group method is likely to do better (and usually
does so in practice). And by the way, one doesn't have to assume uniform
rates of evolution on all lines, only that certain rates do not get TOO
unequal!

In any case, it seems that the actual historical accuracy of "phylogenetic"
methods is becoming increasingly doubtful! Personally, I prefer to apply a
battery of algorithms: as a minimum, principal component (or coordinate)
ordination, a series of phenograms of varying clustering "strength", a
minimum spanning tree, and a series of the shortest length trees. Where
congruence is clear, I'm prepared to accept that someone is trying to tell
me something; where clashes occur, I might hope to see just why
(plesiomorphy, long branches, etc.). And where matters can't be clearly
resolved, why, I'm happy to leave an unresolved cluster. From the aspect of
straight-out classification, this is OK by me; but for the Sherlock
Holmes's who demand that all T's be crossed and I's dotted, I can only
recommend optimism.

Dopn Colless



                 *****************************************
                 * Don Colless, Div of Entomology, CSIRO,*
                 *    GPO Box 1700, CANBERRA, ACT 2601   *
                 *      donc at spider.ento.csiro.au        *
                 *          tel: (02)62464270            *
                 *    Tuz li munz est miens envirun      *
                 *****************************************