Comparison of Information Criteria and Simulation Statistics Method for Topology Selection in Molecular Phylogenetic Tree

In evolutionary studies, a number of methods have been proposed so far for reconstruction of phylogenetic tree. But, it has been pointed out, that even maximum likelihood method which is considered as the most rigorous one still has some problems for estimating multifurcate tree. Two kinds of approaches are thought to be appropriate to resolve this problem. One is the approach by information criteria such as Akaike information criterion (AIC) and Rissanen’s minimum description length (MDL) criterion by which we can deal with the tree topology selection as a sort of model selection problem. Another is to construct statistical framework to test whether a certain length of branch is 0 or not, and thus multifurcate node is detected. In our previous studies [1, 2], we have investigated the qualitative characteristics of AIC and our model-based complexity (MBC) method, which is a variant of MDL method, in estimating the appropriate tree topology. But we have not rigorously examined the effects of modeling error in base substitution. In this study, we will take this kind of estimation error into account and strictly compare the efficiency of MBC method with ML and AIC methods by computer simulation and the results will be examined by statistical hypothesis tests.