knox at spruce.gsfc.nasa.gov (Robert Knox) writes:
>As a follow-up to the extensive discussions last autumn,
>on deriving phylogenetic trees from mitochondrial
>sequences, take a look at the February 7 issue of
>_Science_, vol. 255, pp. 686 and 737.
>The short version: the authors of the original study
>relied on a small, biased sample of randomly generated
>trees to infer that a tree with an exclusively African
>root was significantly more parsimonious (i.e. required
>fewer character state changes) than those with a less
>clear geographic interpretation for the origins of modern
>humans.
>A take home message: If you're going to use the PAUP
>program on a moderately large data set, use a fast
>computer and let it run for a long time!
Whoa! Be careful here. The real message is that if you use PAUP,
make MANY different runs. According to the Science articles, the
problem arose because the 100 trees reported in the Vigilant et al
paper came from a single run of PAUP. Unfortunately, all trees
from a single run of PAUP are inherently related. No matter how
long you let a SINGLE run last, this will be true. The only way
around it is to make multiple runs.
(I'm basing all this on the articles -- I've never used PAUP).
>A more fundamental take home message: Never take _p_
>values at face value. Be sure you understand what the
>probability means, relative to what universe of
>alternative outcomes.
>Well, back to the numbers.
>- Bob Knox
>- Biospheric Sciences Branch, Code 923.0
>- NASA/Goddard Space Flight Center, Greenbelt, MD 20771
>- knox at spruce.gsfc.nasa.gov | BITNET: knox at ricevm1
