Prof. Farrell is spending one year at Universidad Autónoma de Santo Domingo (UASD) in the Dominican Republic as a Fulbright Scholar.
Current research extends the evolution of insect-plant interactions to other trophic levels through a broad collaboration in the beetle Tree of Life project.
A new research dimension in the lab concerns the acoustic signals produced for mating and territory defense. More
We often want to test our hypotheses of evolution. For example, we may ask if the tree we get from our analysis is significantly different than the traditional phylogeny, or whether the observed paraphyly of a particular group is strongly supported, or other questions. There are several ways to do this. The first step in most of them is to create a constraint tree. Open the data file in MacClade and go to the tree window. Create a random tree. Then use MacClade's tree drawing tools to create the constraint tree. A constraint tree for the monophyly of a certain group would have only one resolved internode branch, that between the group and the rest of the taxa. A constraint tree for a previously-existing phylogenetic hypothesis would be more resolved. Export these trees from MacClade, one at a time, naming each appropriately.
Open PAUP and execute the data file. Under analysis, choose load constraints (don't load as backbone constraint). Then, depending on the analysis and optimization criterion:
For Bayesian analyses:
Load the trees found after stationarity. Filter the trees to find those consistent with the constraint tree. The command for this is "filter constraint=[constraint name]", where [constraint name] is replaced by the name of your constraint tree. Record the number of these trees. This number, divided by the total number of post-stationarity trees, is the posterior probability of the hypothesis represented by the constraint tree. For example, if 60 out of 1000 trees are left after filtering for trees with a certain group monophyletic, the probability of monophyly of this group is 6%.
Under the parsimony criterion:
Perform a thorough tree search with the constraint enforced [if using a batch file, insert "enforce=yes constraint=[constraint name]" within the hsearch command (somewhere between "hsearch" and the semicolon), replacing [constraint name] with the name of your constraint]. This generates the most parsimonious tree[s] which meet this constraint. Save these trees. Then load your most parsimonious trees found without the constraint without replacing the trees you just found [go to options in the get trees window and fill in both circles, or use "gettrees mode=7 file=[tree file name];", replacing [tree file name] with the name of your tree file]. The Templeton and winning-sites tests are the appropriate tests to use in this situation and can be selected under Trees -> Tree Scores... -> Parsimony, then choosing nonparametric tests. The command for this is "pscores /nonparamtest=yes;". The Kishino-Hasegawa test, which is also available, is inappropriate when comparing trees which were not specified a priori. We often report this value anyway.
Under the likelihood criterion:
Perform a thorough tree search with the constraint enforced [if using a batch file, insert "enforce=yes constraint=[constraint name]" within the hsearch command (between "hsearch" and the following semicolon), replacing [constraint name] with the name of your constraint]. This generates the likeliest tree which meets this constraint. Save this tree. Then load all likely trees without replacing the trees you just found [go to options in the get trees window and fill in both circles, or use "gettrees mode=7 file=[tree file name];", replacing [tree file name] with the name of your tree file]. What "all likely trees" means is difficult. Some people include the maximum likelihood tree and the most parsimonious trees. You could also include all trees with a length no greater than 5 more than the most parsimonious trees or something like that. Then go to Trees -> Tree Scores... -> Likelihood, then click on the button for topology based tests. Choose the Shimodaira-Hasegawa test. This test measures whether some trees are better than others under likelihood. The Kishino-Hasegawa test, one many people use, is inappropriate (non-conservative) unless all the trees have been specified a priori (specified without using any of the data for a tree search). We often report the value anyway, though noting it's not valid.