Phylogenetic Analyses

Abstract

The main task of phylogenetics is the reconstruction of trees, which can be understood as a representation of evolutionary relationships. Different tree representations include cladograms (branch lengths have no meaning), phylograms (branch lengths correlate with evolutionary change) and ultrametric trees (the path of branch lengths from root to tip are equal; can be used for molecular clocks). Consensus methods, such as strict consensus or majority rule, can be used to summarize the information of multiple trees. In some cases, networks might be a better choice of representation, as they allow reticulate relationships to visualize conflict or horizontal gene transfer. Many tree-building methods rely on explicit models of sequence evolution. Different models of nucleotide substitution are nested and can be derived from a general model (GTR) by restricting free parameters (e.g. different types of substitution, base frequencies). All these models assume homogeneity of evolutionary rates across alignment sites. As this assumption is often unrealistic, rate heterogeneity can be incorporated by classifying differently evolving sites according to a gamma distribution, incorporating information by constant sites or using site-heterogeneous CAT models. Amino acid models are broadly classified into empirical and mechanistic models. Empirical models are derived from large compilations of sequence databases and specify transitions probabilities of all possible amino acid changes within a matrix (e.g. DAYHOFF or WAG). Mechanistic models include assumptions about biological processes or are formulated at the codon level. If models are nested, a selection of the most appropriate model according to hierarchical likelihood ratio tests is available. Other criteria for model selection are based on information criteria such as the AIC or BIC, which penalize the inclusion of more parameters. The most widely applied methods of phylogenetic reconstruction are neighbour joining (NJ), maximum parsimony (MP), maximum likelihood (ML) and Bayesian inference (BI). Whereas NJ is a clustering method based on pairwise distances, all other methods are character based and rely on an optimality criterion. All these methods will turn any alignment into a tree, and metrics are needed to measure the support of certain branches within each tree. Bootstrapping relies on the generation of pseudoreplicates to measure branch support, whereas other measures are based on the likelihood ratio test (e.g. aLRT).