Page 232 - Applied Probability
P. 232

10. Molecular Phylogeny
                              218


                                                         6
                                                    5                root            2
                                                                 1

                                                4
                                                        3



                                            eubacterium  halobacterium  eukaryote  eocyte
                                     FIGURE 10.7. Best Rooted Tree for the Evolution of Eukaryotes
                              with the tree depicted in Figure 10.7.
                                Without the assumption of stationarity, two other rooted trees reducing
                              to the E tree have nearly the same loglikelihoods (−4537.3 and −4537.4)
                              as the best rooted tree. Apparently, distinguishing rooted trees in practice
                              is much harder than distinguishing them in theory. On the other hand, the
                              best rooted tree corresponding to either an F or G tree has a much lower
                              loglikelihood than that (−4544.2) of the best rooted tree.


                              10.8 Codon Models


                              Except for regulatory regions, codons rather than nucleotides are the units
                              of evolution. Because there are 61 non-stop codons compared to 4 nu-
                              cleotides, codon models are much more computationally demanding than
                              nucleotide models. Nonetheless, taking variation at the codon level into ac-
                              count can substantially improve reconstruction of evolutionary trees [9, 25].
                              It also helps identify conserved domains within proteins. The most natural
                              method of turning a nucleotide substitution model into a codon substitu-
                              tion model is to penalize nonsynonymous codon changes. For example,
                              suppose ω bd is the infinitesimal transition rate from base b to base d. On the
                              codon level, the infinitesimal transition rate from codon (a, b, c)tocodon
                              (a, d, c) is still ω bd, assuming sites mutate independently. If two codons are
                              nonsynonymous, then one can penalize transitions between them by replac-
                              ing ω bd by ρè bd for ρ< 1. Here we can view ρ as the probability that a
                              proposed evolutionary change is accepted. If the destination codon is one
                              of the three stop codons, the acceptance probability is 0. The acceptance
                              probability for a synonymous change is 1. In the presence of strong positive
                              selection, a value of ρ> 1 is plausible. In this case, we simply interpret the
                              product ρè bd as a rate.
                                Besides accounting for nonsynonymous codon changes in a parsimonious
                              manner, this codon model possesses the attractive property of turning a
   227   228   229   230   231   232   233   234   235   236   237