Hominoid phylogeny estimated by model selection using goodness of fit significance tests

Phylogeny estimation from nucleotide sequence data may be thought of as a problem of choosing between different evolutionary models that vary with the branching pattern of the phylogeny and with the stochastic process of nucleotide sequence change occurring on the branches of the phylogenetic tree. Thus, each evolutionary model consists of both a particular stochastic process and a particular phylogeny. Such models produce multinomial distributions of nucleotide character patterns. As first suggested by Cavalli-Sforza and Edwards [Evolution 21: 550-570 (1967)] the distribution of patterns expected under each model can be compared to the actual observed distribution of patterns by a goodness of fit statistic such as the loglikelihood ratio G2 or Pearson's X2 after the numerical parameters for the model have been chosen to minimize the respective statistic. For each evolutionary model, the probability P of getting a value of the goodness of fit statistic greater than the observed value is computed. A very small P value means that either a rare event has occurred or that the model is false. Employing for each of 16 models a stochastic process which has 12 parameters to describe the mode of nucleotide change on each branch of each putative phylogenetic tree, we examined all 15 unrooted dichotomously branching arrangements of orthologous noncoding sequences from the gamma hemoglobin genomic region of the five hominoids (gibbon, orangutan, gorilla, chimpanzee, and human) plus the branching arrangement with a trichotomous separation of gorilla, chimpanzee, and human. Of these 16 models, all had P values less than 0.01, except for the arrangement of human joined by chimpanzee, in turn joined by gorilla, and then orangutan and gibbon. This analysis allows convincing claims to be made about hominoid phylogenetic relationships by testing the applicability of the assumed stochastic process for nucleotide sequence evolution at the same time as testing the inferred phylogenetic branching arrangement.