Friday, March 27, 2009

A method that fails . . . or why can’t we all just get along?

Nested Clade Phylogeographic Analysis (NCPA) ranks among the mostly widely used methods in the gene-genealogy based analysis of intraspecific phylogeography. The method was originally described more than 13 years ago (Templeton et al. 1995), and has by some accounts (e.g., Knowles 2008) been used over 1700 times. Certainly, the original description of the method has been cited over 600 times, a number that continues to grow (with 8 citations already reported for 2009 at the time of writing). The method has also been described elsewhere, including in Alan Templeton’s recent book. The primary innovation of the method when introduced (as purported in Templeton et al. 1995 and Templeton 2006) is that by incorporating the temporal information of gene trees, the procedure extends our phylogeographic inference capabilities beyond the traditional FSTs and Nms of classical population genetics.

Nielsen and Beaumont (2009; pp. 1037-1038) provide a wonderfully succinct description of NCPA. Given a sample of haplotypes at a given genetic locus, the user first identifies an estimate of the haplotype network for the sample. He or she then hierarchically agglomerates sets of haplotypes separated by 1, 2, 3, . . ., etc. mutational steps. Next, the user calculates statistics describing the geographic dispersion within clades and between nested clades, and compares these statistics to those obtained by permuting geographical information among samples - retaining the distances from the clades with significant statistics for comparison to an inference key. Inferences liable to result from the procedure include such things as range expansion, long distance migration, and geographic fragmentation.

However, the method has received some pointed criticism as well, most notably from Lacey Knowles at the University of Michigan. Recently, Knowles and Templeton exchanged vitriolic attacks in the pages of the journal Evolution (Knowles 2008; Templeton 2009). Earlier, Knowles and Maddison (2002; maybe someone can tell me why the article, pp. 2623-2635, is missing from the online journal issue here) published a simulation study sharply critical of the method in Molecular Ecology, and Templeton (2004) responded in kind. Criticized shortcomings of the method include an exceedingly high type I error rate in simulation studies (estimated to be ~20% by Templeton 2008, but as high as 75% in other analyses), and subjectivity in the agglomeration of nested clades and in the application of the method's inference key. Other detailed critiques of NCPA can be found in Beaumont and Panchal (2008) and in Nielsen and Beaumont (2009) (the latter even includes a fascinating discussion of the 'Forer effect' as it might pertain to this method).

How do the readers of this blog perceive the current standing of NCPA? Furthermore, but on a much more philosophical note, how do readers view the kind of heated exchange partaken of by Knowles (2008) and Templeton (2009)? Is it true (to paraphrase A. N. Whitehead) that the clash of ideas is more of an opportunity than a calamity?

Note: For other fun exchanges in the recent literature among people I know, please check out: Downhower et al. (2009) and the response by Langerhans and Gifford (2009); or Bokma (2008), and the reply by Rabosky and Lovette (2008).


Dan Rabosky said...

You've posed the question of whether these heated exchanges are good or bad. This reminds me of what might be the gold standard of understated vitriol - the Graur and Martin Trends in Genetics article on molecular clocks (2004). I thought when that article came out that perhaps it would be perceived as an ad hominem attack and thus be less effective in making its point. However, it has been cited nearly 200 times and is probably one of the most recognized articles on the subject published in the past 5 years. Something to think about....

Glor said...

In my view the Graur and Martin article was over the top, to the point of being unprofessional. They certainly got their point across, and good for them if they racked up some citations, but I'm not a fan.

I'm not going to argue in favor of NCA because I've never been comfortable with the method (in spite of learning it from Templeton and later helping him teach it as a TA for his course in population genetics). That said, I think some of the criticisms are unfair. As Dan's subsequent post points out, there are also problems a-plenty with mainstream population genetic methods when it comes to inferring history.

The whole situation was best summarized for me by Monty Slatkin at the end of a seminar I saw him give at UC Davis a number of years ago. He recalled how theorists a decade ago had promised new dimensions of historical population genetic inference, if only the data were available. Now that the data are available, however, he admitted that some historical problems may be too complicated to ever be accurately retraced.

Although only time will tell, it seems likely that this limitation will apply to many of the questions that interest us most.

Liam Revell said...

What a great title for the Graur and Martin article though.