Thursday, December 8, 2011

Testing common ancestry again

At last, the semester winds to a close here at Bryan! What a relief (as my students will no doubt attest). I have four more exams to give and grade, and then I'm done. That's a lot of grading in the next week, but I'm looking forward to getting back to research projects of great interest to me (especially my response to Senter). And in the next few weeks, I'll be cleaning out my backlog of "interesting things I ought to blog about."

Longtime readers might recall an interesting paper by Doug Theobald on "A Formal Test of the Theory of Universal Common Ancestry." In my assessment of the original paper, I expressed some doubts about his methodology, especially since there is no good model for "independent ancestry." Soon after, Koonin and Wolf published a response in which they dismissed Theobald's claims as a trivial consequence of sequence similarity. I initially reacted optimistically to this, but after giving it some thought, I realized that Koonin and Wolf's argument wasn't as good as I initially thought it was. Here's what I wrote about a year ago:
The problem as I've come to see it is this notion of eliminating "phylogenetic signal." By scrambling the sequences the way they do, Koonin and Wolf eliminate the "signal" that supports a bifurcating tree. BUT that's not the only tree possible. If their sequences are truly randomized as they describe, then they are modeling a star phylogeny, which is still a phylogeny.
Now Theobald has a response in the same journal, and here's a snippet:
...lacking hierarchical structure does not necessarily imply that the sequences are independent evolutionary inventions (though a lack of phylogenetic structure may decrease the probability of common ancestry relative to competing hypotheses that predict a lack of structure). As mentioned earlier, star trees are bona fide common ancestry models that lack hierarchical structure. A star tree is a phylogeny, after all.
Hey, that sounds familiar! His conclusion:
It is always possible that a biological model may be proposed in the future that explains the data better than the UCA models. Clearly I have not tested all possible models, especially those yet to be developed. I emphasize again here, as I have elsewhere [120], that I have not provided absolute "proof" of UCA. Proof is for mathematics and whiskey; it is not found in science. Nevertheless, these results provide strong evidence for UCA, given the hypotheses and sequence data currently available.
I can live with that.

The paper contains a fascinating look at basic statistics, including a critique of frequentist statistical tests. As one educated in the frequentist school, I'm going to have to ponder that one. If you're into this kind of thing, I strongly recommend reading the full article.

Theobald. 2011. On universal common ancestry, sequence similarity, and phylogenetic structure: The sins of P-values and the virtues of Bayesian evidence. Biology Direct 6:60.

Feedback? Email me at toddcharleswood [at] gmail [dot] com.