Bodega Phylogenetics Workshop Day 3
Rich Glor, Character Evolution
Official slides and tutorial
- Weakness of ancestral state reconstruction
- Cunningham 1999
- Losos 1999
- Testing hypotheses with ancestral reconstruction
- Farris method. Downward/upward pass. Gives a maximum parsimony reconstruction. May miss other ones.
- Apparently this Parsimony analysis can be done for continuous traits as well, needs a measure of distance
- Pagel 1994
- Schluter 1997
- Discrete trait ancestral state reconstruction
- Wouldn't it be nice to have functions that applied Rich's awesome plots to any data? Segway to writing functions
- Wouldn't a zoomable updatable plot interface be nice? update plot fn and cairo pdf.
- Testing for phylogenetic signal
- Lambda: rescale all internal branches by a factor of lambda.
- Compare likelihood of output under rescaled and original trees.
- The model for these is a single constant rate of transition between all states. Wondering what would happen if we didn't refit the model to each tree but simply compared the likelihood of the model fit to the original tree when analyzed with the dataset on the modified tree.
Carl Boettiger: useful software
- I gave a 20 minute introduction to subversion and Mendeley.
- Gave a demonstration on using subversion with MrBayes. Useful example, but should have emphasized the generality. Also didn't have a chance to demonstrate the rapidsvn GUI.
- I've added information about using subversion to the Bodega wiki.
- Settled on rapidsvn as the suggested GUI, even though I'm not totally happy with it. The kdesvn gui seems a better choice for linux, but rapidsvn works on all platforms and has an excellent guide.
Justin Whittall: cts character traits
Official slides and tutorial
- Check out the Charlie Nunn Papers.
- Compare incoperates uncertainty in the species means.
- Are contrasts standardized by branch length? Absolute value of contrast independent of standard deviation of contrast? If not, transform data..
My Thoughts and Explorations for the Day
- Many analyses are not explicit about the model used and the assumptions it involves!
- Not happy with the independence assumption on columns of the alignment in DNA data. We can quantify the predictive capacity of that information...
- Rich Glor's likelihood test under Pagel's lambda. seems a strange thing to do, what does it actually mean?
- Need to explore the later applications of independent contrasts further.
- Figuring out strange behavior in ouch. the value kept at model@alpha is actually the Cholesky decomposition of the alpha matrix, just as the values of sigma are the Cholesky decomposition of the σ2 matrix. This seems like a strange convention, surely the user would rather specify the matrices as full matrices and let the software do the decomposition? Because this decomposes matrix M into M = LLT, for single trait value alpha is a scalar and this is the square root. sym.par() function in ouch is often called in the source to assemble M from the decomposition...