Ancestral state reconstruction with large numbers of sequences and edge-length estimation (Q2113521)

The main goal of the present work using likelihood-based methods is to study the consistency of the maximum likelihood and empirical Bayes estimators of the ancestral state of discrete traits in such settings under a star tree. The authors prove that the likelihood-based reconstruction is consistent under symmetric models but can be inconsistent under non-symmetric models. More preciously speaking they show that in this setting the maximum likelihood estimator (MLE) and empirical Bayes estimator (sometimes also referred to as the maximum a posteriori (MAP) estimator) converge upon the same solution. Consequently, it suffices to consider the MLE. The MLE is shown to be consistent under symmetric models but there exists a zone of inconsistency under non-symmetric models. They show, also, that a simple consistent estimator for the ancestral states is available under non-symmetric models. It is noted by them that in some species, for which sequence length is much smaller than the number of species, edge lengths are not likely to be accurately estimated, whatever is the tree topology is reasonable or not. The results illustrate that the likelihood methods can unexpectedly have undesirable properties as the number of sequences considered gets very large. Broader implications of the results are discussed.