User:Carl Boettiger/Notebook/Comparative Phylogenetics/2010/04/18

Comparative Phylogenetics

Main project page Previous entry      Next entry

MATICCE Testing out Andrew Hipp's maticce package, doi:10.1093/bioinformatics/btp625. Goal of package is to model-average over trees and regimes. Takes a brownie approach to regimes, identifying nodes at which a transition may have occurred. The Approach Takes a brute force approach to avoid user-defined paintings in the ouch framework. Users identify nodes of interest, software tries combinations of possible shifts. runBatchHansen() User identifies a set of nodes of interest, and optionally a maximum number of transitions. What models/regimes are tested? You choose a set of candidate nodes and a maximum number of transitions that can occur (defaults to all nodes). For instance, you may specify four candidate nodes but set a max at 2 changes (three peaks). Then the batch will evaluate 12 models: the 4 choose 2 = 6 three-peak models, the 4 choose 1 = 4 two peak models, a single peak, and a brownian motion model. If you didn't specify a max, it would allow the 1 five-peak (4 choose 4), and the 4 four-peak (4 choose 3) models, for a total of 17 models. Has summary function. Model averaging is a weighted average by BIC characterStates -- the trait data, currently must be one-dimensional, seems like it can be in ape or ouch format. nodes of interest have to be specified by listing all the species descendants. somewhat cumbersome. Takes an ouchtree or list of ouchtrees. Output See loglik, sigma and alpha values of the models tested (doesn't include brownian motion by default) <syntaxhighlight lang="rsplus"> trialBayesConsensus <- runBatchHansen(ape2ouch(carex$ovales.tree), carex$ovales.data, carex$ovales.nodes[1:4], maxNodes = 2) > trialBayesConsensus$hansen 1 loglik dof sigma.squared theta / alpha 1 52.72744 5 5.31005363 331.626872 2 50.54000 5 5.86424413 337.252821 3 50.21561 5 6.04474990 343.369492 4 49.95720 4 5.14354118 289.387746 5 51.75330 5 5.12434568 308.473054 6 51.81077 5 4.84664447 292.440242 7 51.75142 4 5.16258108 310.762680 8 44.31518 5 6.55448190 297.993592 9 44.17736 4 8.49647580 384.329498 10 42.89766 4 6.87292796 296.238909 11 39.57676 3 0.09810907 3.341806 See which nodes involved a shift: > trialBayesConsensus$regMatrix 1 1 2 3 4 1 1 1 0 0 2 1 0 1 0 3 1 0 0 1 4 1 0 0 0 5 0 1 1 0 6 0 1 0 1 7 0 1 0 0 8 0 0 1 1 9 0 0 1 0 10 0 0 0 1 11 0 0 0 0 </syntaxhighlight> Where the four nodes are shown as separate columns, and there numbers correspond to their indices specified in ovales.nodes. multiModel() For a focal node, there are 5 hypotheses we want to distinguish: BM OU Two OU Two BM, censored as separate sub trees (data of non-descendants is not considered) Two OU, censored. Note that the first three of these cases are among the cases covered in a runBatchHansen() over the node. Other functions paintBranches() turns a maticce specification of a focal node into an OUCH regimes list, identifying the regime of each branch. Other Notes Evaluation A useful package for batch runs of hansen hypotheses, reasonably polished. Two chief drawbacks: Method of specifying which nodes are of interest seems very annoying. Should require only two tips, and hence specify their common ancestor as the node. Should have an option to identify the node merely by id number. Having to specify nodes doesn't really get around the painting problem. The parameter space of paintings is too big!! A new approach is needed to thinking about this; one that does not consider all possible outcomes as different models. (but of course I'd say that). Notes / Reading Linked Data a new approach to semantic web ideas. Newsweek on why scientists are losing PR battle. Recently started wikipedia page on Science 2.0 lots of Stack Overflow models. Came across: Semantic Overflow (thanks Rod Page).