Comparison of models for nucleotide substitution used in maximum- likelihood phylogenetic estimation

Using real sequence data, we evaluate the adequacy of assumptions made inevolutionary models of nucleotide substitution and the effects that theseassumptions have on estimation of evolutionary trees. Two aspects of theassumptions are evaluated. The first concerns the pattern of nucleotidesubstitution, including equilibrium base frequencies and thetransition/transversion-rate ratio. The second concerns the variation ofsubstitution rates over sites. The maximum-likelihood estimate of treetopology appears quite robust to both these aspects of the assumptions ofthe models, but evaluation of the reliability of the estimated tree byusing simpler, less realistic models can be misleading. Branch lengths areunderestimated when simpler models of substitution are used, but theunderestimation caused by ignoring rate variation over nucleotide sites ismuch more serious. The goodness of fit of a model is reduced by ignoringspatial rate variation, but unrealistic assumptions about the pattern ofnucleotide substitution can lead to an extraordinary reduction in thelikelihood. It seems that evolutionary biologists can obtain accurateestimates of certain evolutionary parameters even with an incorrectphylogeny, while systematists cannot get the right tree with confidenceeven when a realistic, and more complex, model of evolution is assumed.