Implementing and testing the multispecies coalescent model: A valuable paradigm for phylogenomics

In recent articles published in Molecular Phylogenetics and Evolution, Mark Springer and John Gatesy (S&G) present numerous criticisms of recent implementations and testing of the multispecies coalescent (MSC) model in phylogenomics, popularly known as "species tree" methods. After pointing out errors in alignments and gene tree rooting in recent phylogenomic data sets, particularly in Song et al. (2012) on mammals and Xi et al. (2014) on plants, they suggest that these errors seriously compromise the conclusions of these studies. Additionally, S&G enumerate numerous perceived violated assumptions and deficiencies in the application of the MSC model in phylogenomics, such as its assumption of neutrality and in particular the use of transcriptomes, which are deemed inappropriate for the MSC because the constituent exons often subtend large regions of chromosomes within which recombination is substantial. We acknowledge these previously reported errors in recent phylogenomic data sets, but disapprove of S&G's excessively combative and taunting tone. We show that these errors, as well as two nucleotide sorting methods used in the analysis of Amborella, have little impact on the conclusions of those papers. Moreover, several concepts introduced by S&G and an appeal to "first principles" of phylogenetics in an attempt to discredit MSC models are invalid and reveal numerous misunderstandings of the MSC. Contrary to the claims of S&G we show that recent computer simulations used to test the robustness of MSC models are not circular and do not unfairly favor MSC models over concatenation. In fact, although both concatenation and MSC models clearly perform well in regions of tree space with long branches and little incomplete lineage sorting (ILS), simulations reveal the erratic behavior of concatenation when subjected to data subsampling and its tendency to produce spuriously confident yet conflicting results in regions of parameter space where MSC models still perform well. S&G's claims that MSC models explain little or none (0-15%) of the observed gene tree heterogeneity observed in a mammal data set and that MSC models assume ILS as the only source of gene tree variation are flawed. Overall many of their criticisms of MSC models are invalidated when concatenation is appropriately viewed as a special case of the MSC, which in turn is a special case of emerging network models in phylogenomics. We reiterate that there is enormous promise and value in recent implementations and tests of the MSC and look forward to its increased use and refinement in phylogenomics.