[Taxacom] Barcoding and reliability
Richard Zander
Richard.Zander at mobot.org
Wed May 30 12:05:09 CDT 2007
I've gotten a few offline responses to my suggestion that barcoding may
be less than impressive as support for standard taxonomic methods. The
"tar and feathers" message was in fun, I'm sure.
First, barcoding is fine if done right, as a general search for
distinctive genotypes that may be further investigated with standard
methods. If you have the time and money, barcoding could be a great
thing.
My cautions had to do with the tendency we all have to use a new method
or technology as a shortcut. Barcoding alone cannot suffice to
distinguish new taxa. Commonly, in my opinion, it cannot even be support
for a taxonomic decision that is somewhat dubious itself.
The Bayes' Formula basically indicates that if you have two studies that
are independent and the results of both are greater than 50% probability
of being correct, then the posterior probability goes up. The two
studies support each other. The only way that barcoding can demonstrate
a 50% or greater probability of distinguishing a new taxon (versus a
false positive) is a study that, for a particular group, demonstrates
that more than 50% of the time, the barcoding distinguished a new taxon.
Not just a fully cryptic taxon (distinguishable by barcoding alone) but
one with distinctive evolutionary ecology, at least, if not morphology.
Then, for additional specimens collected in that group, barcoding is
supportive of standard taxonomy.
But what about extending that particular method of barcoding to another
group? If it can be determined that extension to another group is valid
at merely more than 50% of the new groups, that is NOT sufficient to
cover all groups with an expectation that all will have greater than 50%
of the taxa distinguishable by barcoding. This is a multiple tests
problem, and the probabilities must be multiplied because both must be
true at the same time. You need about 75% chance of distinguishing a
taxon in the first group, plus a 75% chance of extending the method to
other groups before you can assert that barcoding with a particular
method provides support (greater than 50% ability to distinguish new
taxa) for subsequent taxonomic study in many groups. This is because the
joint probability is the product, or a little more than 50%. I don't
think this has been demonstrated. Otherwise, you need to demonstrate
greater than 50% probability of barcoding identifying new taxa in each
and every new group, or the forgoing applies.
What about barcoding as a means of distinguishing new taxa without
subsequent standard analysis? Such that it provides sufficient support
itself to distinguish new taxa. Suppose one demonstrates that barcoding
avoids false positives 95% of the time for a particular group. (Again,
please note that fully cryptic new taxa can involve circular reasoning.)
If we agree with other fields (psychology, ecology) that promote
statistical analysis in encouraging a minimum 95% level of confidence,
then that's great, 95% does it. Extension of the method to other groups,
however, requires demonstration that MORE than 95% of other groups
respond well to this method. You need 97.5% confidence in the first
group AND demonstration that the method works at 97.5% in more than
97.5% of other groups. This is because 97.5 times 97.5 = a little more
than 95%, which is the joint probability. THEN you can extend
stand-alone barcoding to other groups without determining confidence in
ALL groups. OTHERWISE, you must demonstrate greater than 95% probability
of avoiding false positives in each and every group. This has not been
demonstrated, either.
This isn't rocket science, and, heck, I might have it wrong. This kind
of simple statistics is central, however, to a science that asserts a
particular method has some kind of special reliability.
******************************
Richard H. Zander
Voice: 314-577-0276
Missouri Botanical Garden
PO Box 299
St. Louis, MO 63166-0299 USA
richard.zander at mobot.org
Web sites: http://www.mobot.org/plantscience/resbot/
and http://www.mobot.org/plantscience/bfna/bfnamenu.htm
For FedEx and UPS use:
Missouri Botanical Garden
4344 Shaw Blvd.
St. Louis, MO 63110
******************************
More information about the Taxacom
mailing list