User:Carl Boettiger/Notebook/Comparative Phylogenetics/2010/03/15
|Comparative Phylogenetics||Main project page|
Previous entry Next entry
The failure of the Anoles data set to robustly select the more common model can easily be explained by the likelihood distribution: Not surprising since this dataset consists of only 23 taxa. The Labrid tree of 114 taxa clearly resolves the difference between the different models. Bootstrapping should be a more reliable guide than comparing AIC scores.
Compare to the Labrid data set:
These use 200 bootstrap simulations, repeating with 2000 for better resolution.
Breakthrough! This approach has far greater implications than I first realized. This provides an inverse approach which solves the hardest part of my original problem, and then tests that under the phylogenetic context. Solving the partition problem independent of the phylogeny, reconstructing the ancestral states as a purely discrete problem, and then scoring the resulting continuous time model on the phylogeny is likely to be far more effective, efficient, and robust, then the direct calculation of the joint probability that I've been pursuing. While the three steps: partitioning, discrete inference, and continuous inference; all need refining, I think I'm very close to my first implementation of the full problem!
From the git log
commit ddf6db6fca626c8f6ed31f4ab2a7cb78ed70f226 Author: Carl Boettiger <email@example.com> Date: Mon Mar 15 09:40:08 2010 -0700
Updated doxygen comments
commit 17dd3775004741c9c590828b498dacd75c5bfc49 Author: Carl Boettiger <firstname.lastname@example.org> Date: Mon Mar 15 09:39:34 2010 -0700
added license and multimin fn
commit 02b9bfc02cb830297fc7fad81cfcd05598d482f1 Author: Carl Boettiger <email@example.com> Date: Mon Mar 15 09:22:21 2010 -0700
mcmc is basics now implemented.
commit 220ccae12fbbdf78cd5e082555c04b99af92c4af Author: Carl Boettiger <firstname.lastname@example.org> Date: Sun Mar 14 20:51:50 2010 -0700
adding the mcmc routine
commit 94642c8d406fb0d6e8970dc31cfd5304dda2c007 Author: Carl Boettiger <email@example.com> Date: Sun Mar 14 15:33:39 2010 -0700