Poor Fit to the Multispecies Coalescent is Widely Detectable in Empirical Data

Abstract

Model checking is a critical part of Bayesian data analysis, yet it remains largely unused in systematic studies. Phylogeny estimation has recently moved into an era of increasingly complex models that simultaneously account for multiple evolutionary processes, the statistical fit of these models to the data has rarely been tested. Here we develop a posterior predictive simulation-based model check for a commonly used multispecies coalescent model, implemented in *BEAST, and apply it to 25 published data sets. We show that poor model fit is detectable in the majority of data sets; that this poor fit can mislead phylogenetic estimation; and that in some cases it stems from processes of inherent interest to systematists. We suggest that as systematists scale up to phylogenomic data sets, which will be subject to a heterogeneous array of evolutionary processes, critically evaluating the fit of models to data is an analytical step that can no longer be ignored.

Coalescent methods for estimating species trees from phylogenomic data

“…The coalescent inference is based on the assumption that the multispecies coalescent model is a good approximation to the real biological process that causes incongruent gene trees. There are preliminary attempts to assess the goodness of fit of the multispecies coalescent model, but they were either limited to small data sets (Reid et al, 2014) or they were based on the distance between gene trees and the species tree (Song et al, 2012), which may not have the power to reject the multispecies coalescent model. Reid et al (2014) evaluated the multispecies coalescent model in a Bayesian framework using posterior predictive simulation (PPS), in which the estimated gene trees were compared with the predictive distribution of gene trees.…”

“…There are preliminary attempts to assess the goodness of fit of the multispecies coalescent model, but they were either limited to small data sets (Reid et al, 2014) or they were based on the distance between gene trees and the species tree (Song et al, 2012), which may not have the power to reject the multispecies coalescent model. Reid et al (2014) evaluated the multispecies coalescent model in a Bayesian framework using posterior predictive simulation (PPS), in which the estimated gene trees were compared with the predictive distribution of gene trees. Since the predictive distribution of gene trees is generated from a Bayesian coalescent approach (i.e., Ã BEAST), PPS is not able to evaluate the multispecies coalescent model for genome-scale sequence data.…”

Extant diversity of bryophytes emerged from successive post-Mesozoic diversification bursts

“…Indeed, simulation studies have shown that combining data can provide high statistical support for an incorrect species tree because of processes potentially leading to incongruence among data partition, including incomplete lineage sorting, gene duplications/extinctions (paralogy), non-neutral evolution and hybridization (see ref. 42 for review), questioning the use of a total evidence approach [43][44][45] . Incongruence was operationally defined as the presence of incompatible bipartitions found in branches supported with bootstrap proportions 495% from separate analyses of two data sets.…”

Multilocus Species Trees Show the Recent Adaptive Radiation of the Mimetic Heliconius Butterflies

“…Considering that our focal group is intensively studied and exceptionally well represented in research, museum, and private collections due to its aesthetic appeal , it would be considerably more challenging to obtain a complete sampling of many other groups. Another difficulty stems from the fact that the advanced coalescent techniques like BUCKy and *BEAST perform best with multiple samples per species, which should capture intraspecific diversity, and may require complete taxon coverage (Heled and Drummond 2010;Reid et al 2013;Steel and Velasco 2014). Much of the uncertainty in the estimates can be attributed to missing data, which can negatively affect the estimation of both individual gene trees and the encompassing species tree (Wiens and Morrill 2011;Roure et al 2013).…”