User:Carl Boettiger/Notebook/Teaching/2009/11/03

From OpenWetWare

Jump to: navigation, search
Teaching Main project page
Previous entry      Next entry

CPB Talk

Inferring Adaptive Landscapes from Phylogenetic Trees was the title of the talk I just gave to the Center for Population Biology. This is the first time I’ve given a talk to my entire program; the next time might be my exit seminar. I’ve given an earlier version of this talk, about 10 minutes long, in the SIAM student conference last April, and a 30 minute version to Ulf Dieckmann’s EEP group at IIASA, but this was the first hour-long talk I was to give in any setting outside of a lab group. CPB has made a push for more student talks in its weekly seminar series, and for the most part it has been an excellent shift, if for no other reason than it is more fun to hear about the work of people you know, and also offers the rest of the students more benchmarks.

My goals for this talk were: (a) to give an engaging and stylized talk (b) provide an accessible description of how the existing methods work, what kinds of things they show us, and what shortcomings they have, (c) introduce the need for nonlinear models, (d) introduce the biological questions we seek to answer (e) give some suggestion of how our approach works with examples. I think I did significantly better on the early goals then on the later ones.

The talk uses my minimalistic chalkboard style latex template and begins with a story about how Chris Martin got me started on this project, as told through a couple emoticons, most memorably Chris himself: Q}-< (with ponytail)*, making fun of us both. Judging by the laughter, this part was well received and eased me into the talk. The next section reviews comparative methods, and employs “dramatic coin tosses” as Turelli disapprovingly calls them, while I try to explain the underlying idea behind Brownian motion as a model for evolution. It was memorable and helped me think, but sounds like it might have been a bit grating, particularly to anyone who is inclined to resent presentations that try to be more entertainment than science. Guess I’ll need to learn some more subtlety, I think this kind of talk can easily go awry if it seems the presenter is trying to hard to be entertaining, rather than simply explaining the science in a way that is almost accidentally entertaining. Something to ask more people about at some point.

At two points I presented a slide with a commentary phrase on the method which I didn’t state verbally or turn to look at, my only hint that the audience got the slide were the occasional giggles when it appeared. These slides tried to draw attention to the over-interpretation of the parameters of simple or even null models, without me making the critique. Not sure what my verdict is on this one, but will probably try it again sometime, maybe when it’s simpler commentary.

Overall, it sounds like the explanation of the comparative methods was as clear as I could hope for. I transition into the next section by taking a second look at the example tree that I’ve been using to illustrate the method thus far. This tree has trait values from two disparate clusters, and both BM and OU models infer that most of the tree should be intermediate to these values, while a postulate of two peaks seems more likely. I think this made the problem with the existing approaches quite clear, though my solution of adding a multi-peaked adaptive landscape was less popular. I introduced the Anolis data as a multipeak problem, and showed we can infer multiple peaks even though you don’t see them in the original data. I guess this worked, but didn’t carry the impact I might have hoped. Demonstrating we could recover a model we used to generate our own data would have been a nice second example. I then introduced the Labrid data set, which again went over okay, though might have used some more detail and more visuals.

The weakest part seemed to be my explanation of the forward process, described as “three loops.” It’s a technical discussion without a clear example, and all done in one slide. Almost worthless, though some responsive audience members asked enough questions to help clarify. I guess it was good in that it stimulated some challenging questions which I was happy for, though I’m afraid the ending tone may have been one of ‘too many questions’ or doubts about the method. I concluded with a couple suggestions of further nonlinear models, such as Peter’s bounded evolution and a nod to comparative methods, but I think neither of these had enough material or discussion accompanying them to really carry.

Great to hear from faculty and students that I did well on goals (a) and (b), and for those who work on different kinds of things those are the main take away anyway. Very happy that I seem to have gotten Graham Coop and some other theoretical genetics types interested; sounds like we’re forming an informal coffee group to discuss these ideas further. Graham has pointed out a couple statistical tricks I should be implementing, and also identified the two chief concerns — the low probability of any given outcome and the interpretation of transitioning between peaks. Michael Turelli let me know that he thinks the entire approach is nonsense, because in my model things only change peaks by evolving against the gradient to cross the valley. This question of how evolution crosses between peaks is definitely an interesting one and one in which I’m looking forward to hearing the explanations of different evolutionary biologists. I think that their descriptions aren’t as inconsistent with the general SDE on multi-peaked landscape as they’d like to suggest, and a derivation of the macroscopic SDE from several possible microscopic descriptions might be insightful to show them.

  • Inspiration from poet John Rives on TED.



Personal tools