Deflating Trees: Improving Bayesian Branch-Length Estimates using Informed Priors

Abstract

Prior distributions can have a strong effect on the results of Bayesian analyses. However, no general consensus exists for how priors should be set in all circumstances. Branch-length priors are of particular interest for phylogenetics, because they affect many parameters and biologically relevant inferences have been shown to be sensitive to the chosen prior distribution. Here, we explore the use of outside information to set informed branch-length priors and compare inferences from these informed analyses to those using default settings. For both the commonly used exponential and the newly proposed compound Dirichlet prior distributions, the incorporation of relevant outside information improves inferences for data sets that have produced problematic branch- and tree-length estimates under default settings. We suggest that informed priors are worthy of further exploration for phylogenetics.

Assessing the performance of DNA barcoding using posterior predictive simulations

“…; Nelson et al . ). The use of independent and identically distributed exponential priors has been shown to cause overly long branch lengths to be inferred for certain data sets, particularly ‘intraspecific’ data sets where true branch lengths are likely short (as is often the case for barcoding data sets).…”

“…), although it is also clear that this prior has not entirely alleviated the problem (Nelson et al . ). Informing priors based on ML parameter estimates from the data set to be analysed has been criticized for being non‐Bayesian and for artificially reducing uncertainty in the posterior distribution (Zhang et al .…”

“…Informing priors based on other ‘similar’ data sets has also been proposed (Nelson et al . ). However, even when data sets are available for comparison, assessing data set ‘similarity’ can be difficult and does not guarantee that estimates obtained using those data sets will be appropriate for the focal data set being analysed (Nelson et al .…”

“…We used the Nelder–Mead method as implemented in the ‘ optim ’ function in r 3.0 and the r package ‘ bbmle ’ v1.0.17 for ML parameter estimation (Nelson et al . ). Under the compound Dirichlet tree‐length prior, the tree length is considered drawn from a gamma distribution with mean α T /λ T .…”

“…Branch length estimation in Bayesian phylogenetics is currently an active area of research, and the most appropriate way to set priors on branch lengths is a subject of considerable focus (Brown et al 2010;Marshall 2010;Rannala et al 2012;Zhang et al 2012;Nelson et al 2015). The use of independent and identically distributed exponential priors has been shown to cause overly long branch lengths to be inferred for certain data sets, particularly 'intraspecific' data sets where true branch lengths are likely short (as is often the case for barcoding data sets).…”