But the proteins in my universal data set have relatively high sequence similarities.
For the real, universal data set, the models are ordered as follows, best to worst: common ancestry > independent ancestry > star tree > profile model.
A universal data bus for everything important now.
Utilizing the quality information in consolidating data sets demands a universal data-quality measure that is not yet existent.
CA, common ancestry; IA, independent ancestry; BF, Bayes factor Log marginal likelihoods and Bayes factors for the real, universal protein data set.
Therefore, contrary to K&W's hypothesis, similarity alone cannot explain the complex sequence correlations observed in the universal protein data set.
In the real, universal protein data set, there are only 12 sequences, so each column has only 12 observations from which to estimate the 19 parameters, giving a data-to-parameter ratio of 12/19 = 0.63.
The strong phylogenetic structure in the universal protein data set I used (and in similar data sets in the literature) is well established, and trees based on this dataset have local regions of high topological support [ 103, 104].
For the real, universal protein data set, this hierarchical structure is the overriding factor for the superiority of the UCA hypothesis over competing hypotheses, including various independent ancestry models.
However, for the universal protein data set used in my paper, the profile models are extremely poor compared to common ancestry (log-likelihood differences 30,000, Table 4) — much worse even than independent ancestry models.
