Algorithmic improvements to species delimitation and phylogeny estimation under the multispecies coalescent

Abstract

The focus of this article is a Bayesian method for inferring both species delimitations and species trees under the multispecies coalescent model using molecular sequences from multiple loci. The species delimitation requires no a priori assignment of individuals to species, and no guide tree. The method is implemented in a package called STACEY for BEAST2, and is a extension of the author's DISSECT package. Here we demonstrate considerable efficiency improvements by using three new operators for sampling from the posterior using the Markov chain Monte Carlo algorithm, and by using a model for the population size parameters along the branches of the species tree which allows these parameters to be integrated out. The correctness of the moves is demonstrated by tests of the implementation. The practice of using a pipeline approach to species delimitation under the multispecies coalescent, has been shown to have major problems on simulated data (Olave et al. in Syst Biol 63:263-271. doi: 10.1093/sysbio/syt106 , 2014). The same simulated data set is used to demonstrate the accuracy and improved convergence of the present method. We also compare performance with *BEAST for a fixed delimitation analysis on a large data set, and again show improved convergence.

Integrative taxonomy by molecular species delimitation: multi-locus data corroborate a new species of Balkan Drusinae micro-endemics

“…(IV) We performed combined species tree estimation and species delimitation analysis as available via Species Tree And Classification Estimation, Yarely [STACEY] in BEAST 2 [90]. Simplified, this method (as its predecessor, DISSECT [91]) is an extension of the multispecies coalescence model used in *BEAST [75], in which a birth-death-collapse model is used to estimate the species tree [90, 91].…”

“…Simplified, this method (as its predecessor, DISSECT [91]) is an extension of the multispecies coalescence model used in *BEAST [75], in which a birth-death-collapse model is used to estimate the species tree [90, 91]. Further, specialized operators are included that model population sizes along branches, prune and regraft subtrees, modify node heights, and merge tips to minimal clusters.…”

“…Further, specialized operators are included that model population sizes along branches, prune and regraft subtrees, modify node heights, and merge tips to minimal clusters. The method aims to maximize tree likelihood over a Bayesian tree space by using a MCMC model in which single tips can be merged to minimal clusters to estimate a species or minimal cluster (SMC) tree, while specific priors ensure compatibility between species and gene trees [90]. User-supplied priors define behaviour of MCMC moves, and provide a probability space for the expected number of species primarily through the Collapse Weight prior in combination with a Collapse Height parameter [90, 91].…”

“…The method aims to maximize tree likelihood over a Bayesian tree space by using a MCMC model in which single tips can be merged to minimal clusters to estimate a species or minimal cluster (SMC) tree, while specific priors ensure compatibility between species and gene trees [90]. User-supplied priors define behaviour of MCMC moves, and provide a probability space for the expected number of species primarily through the Collapse Weight prior in combination with a Collapse Height parameter [90, 91]. While not extensively tested, a wide range of values [1e-4–1e-7] for Collapse Height was found to provide similar species delimitation results [91].…”

Molecular association and morphological characterisation of Himalopsyche larval types (Trichoptera, Rhyacophilidae)

“…GMYC has, for example, been shown to be an accurate and conservative method under conditions where effective sample sizes are low, and divergence times between species are high, since such conditions yield monophyletic species (Esselstyn et al 2012, Fujisawa and Barraclough 2013). To resolve the species complexes more genes should be used, ideally under a multi-species coalescence approach, such as BPP (Yang 2015), or STACEY (Jones 2017). These methods combine data from several genes to estimate the species borders, accounting for incomplete lineage sorting and conflicting information among genes.…”