The choice of an appropriate model is a critical aspect of a phylogenetic-likelihood analysis. Posada and Crandall (1998) argued that the use of their ModelTest program makes selection of the model objective. There are many models of molecular evolution, and the determination of which to use can significantly influence the results of an analysis. Models range in complexity from the relatively simple Jukes-Cantor model through the most complex GTR model. Currently there are at least 16 models that are commonly used in molecular systematics, most of which are simply special cases of the GTR model (Rodriguez et al. 1990). Each of the 16 basic models is varied with G (gamma distribution), I (proportion of invariable sites), and both (G +I) for a total of 56 different models (Posada and Crandall 1998). ModelTest ranks them in terms of their relative complexity (i.e., more complex = more parameters). The overall likelihood score of a tree increases with increasing complexity of the model, but the accuracy of the model decreases due to the increased number of estimated parameters (Huelsenbeck and Rannala 1997a). The program conducts pairwise comparison of the maximum likelihood trees generated under each model using hierarchical-likelihood ratio tests (Huelsenbeck and Crandall 1997; Huelsenbeck and Rannala 1997a; Posada and Crandall 1998; Johnson and Omland 2004). When no statistically significant difference between two trees is found, the simplest model of these is selected. Model selection based on the relative likelihood values is an ontological appeal to the principle of parsimony, because choosing the least complex explanation of the data rules out the possibility that evolution proceeded in a more complex manner (Huelsenbeck and Rannala 1997a).
Was this article helpful?