Bayesian Inference of Species Trees using Diffusion Models

Abstract

<jats:title>Abstract</jats:title> <jats:p>We describe a new and computationally efficient Bayesian methodology for inferring species trees and demographics from unlinked binary markers. Likelihood calculations are carried out using diffusion models of allele frequency dynamics combined with novel numerical algorithms. The diffusion approach allows for analysis of datasets containing hundreds or thousands of individuals. The method, which we call Snapper, has been implemented as part of the BEAST2 package. We conducted simulation experiments to assess numerical error, computational requirements and accuracy recovering known model parameters. A re-analysis of soybean SNP data demonstrates that the models implemented in Snapp and Snapper can be difficult to distinguish in practice, a characteristic which we tested with further simulations. We demonstrate the scale of analysis possible using a SNP dataset sampled from 399 fresh water turtles in 41 populations.</jats:p>

Demographic decline and lineage-specific adaptations characterize New Zealand kiwi

“…A single species tree was then estimated from the UCE trees using Astral-III v. 5.7.7 [35]. Divergence times between lineages were estimated in SNAPPER v. 1.0.1 [36] using 1000 randomly selected SNPs from one individual per lineage (electronic supplementary material, table S2), with divergence of the kiwi basal split calibrated using a previous estimate of 5.96 Ma [29].…”

Quantitatively defining species boundaries with more efficiency and more biological realism

“…However, these additional complexities in the model are met with highly efficient proposal kernels [28, 17, 25], and much like the Yule-skyline collapse model, is expected to converge quite efficiently in MCMC. Lastly, we demonstrated how the collapse model can be used for molecular sequence analysis in conjunction with StarBeast3 [25] and for SNP analysis in conjunction with SNAPPER [26] - each of which are reported to be significantly more efficient than their predecessors. We demonstrated that StarBeast3 outperforms STACEY at achieving convergence during Bayesian MCMC ( Fig.…”

“…Using this method, we can run a single MCMC analysis and test a large number of hypotheses, whereas BFD* requires a path sampling run for each hypothesis under consideration, and each of these path sampling runs are at least as computationally intensive as a single MCMC run. By using SNAPPER instead of SNAPP, a further order of magnitude in performance gain is accumulated [26] [4]. For each hypothesis, the number of species, the log e marginal likelihood ML (averaged across 5 replicates), the Bayes factor BF, and the total rank are reported (with a Yule tree prior).…”

Drainage basins serve as multiple glacial refugia for alpine habitats in the Sierra Nevada Mountains, California

“…Thefully-filtered dataset was used, but due to computational constraints, one random individual was selected from each of eight populations. Individuals were chosen from populations with relatively pure ancestry coefficients to avoid violating assumptions in the species tree approach, namely the effect of gene flow from other populations (see the result of sNMF analysis; Bryant et al, 2012, Stoltz et al, 2019. SNAPP was run for 30 million Markov chain Monte Carlo (MCMC) generations, with the custom program SNAPPER employed to run SNAPP at faster speed (Stoltz et al, 2019).…”

“…Individuals were chosen from populations with relatively pure ancestry coefficients to avoid violating assumptions in the species tree approach, namely the effect of gene flow from other populations (see the result of sNMF analysis; Bryant et al, 2012, Stoltz et al, 2019. SNAPP was run for 30 million Markov chain Monte Carlo (MCMC) generations, with the custom program SNAPPER employed to run SNAPP at faster speed (Stoltz et al, 2019). Two independent runs were conducted and the output trees were visualized after 10% burnin using DensiTree version 2.01 (Bouckaert 2010).…”