Robustness of compound Dirichlet priors for Bayesian inference of branch lengths

We modified the phylogenetic program MrBayes 3.1.2 to incorporate the compound Dirichlet priors for branch lengths proposed recently by Rannala, Zhu, and Yang (2012. Tail paradox, partial identifiability and influential priors in Bayesian branch length inference. Mol. Biol. Evol. 29:325-335.) as a solution to the problem of branch-length overestimation in Bayesian phylogenetic inference. The compound Dirichlet prior specifies a fairly diffuse prior on the tree length (the sum of branch lengths) and uses a Dirichlet distribution to partition the tree length into branch lengths. Six problematic data sets originally analyzed by Brown, Hedtke, Lemmon, and Lemmon (2010. When trees grow too long: investigating the causes of highly inaccurate Bayesian branch-length estimates. Syst. Biol. 59:145-161) are reanalyzed using the modified version of MrBayes to investigate properties of Bayesian branch-length estimation using the new priors. While the default exponential priors for branch lengths produced extremely long trees, the compound Dirichlet priors produced posterior estimates that are much closer to the maximum likelihood estimates. Furthermore, the posterior tree lengths were quite robust to changes in the parameter values in the compound Dirichlet priors, for example, when the prior mean of tree length changed over several orders of magnitude. Our results suggest that the compound Dirichlet priors may be useful for correcting branch-length overestimation in phylogenetic analyses of empirical data sets.     

Effects of branch length uncertainty on Bayesian posterior probabilities for phylogenetic hypotheses

In Bayesian phylogenetics, confidence in evolutionary relationships is expressed as posterior probability--the probability that a tree or clade is true given the data, evolutionary model, and prior assumptions about model parameters. Model parameters, such as branch lengths, are never known in advance; Bayesian methods incorporate this uncertainty by integrating over a range of plausible values given an assumed prior probability distribution for each parameter. Little is known about the effects of integrating over branch length uncertainty on posterior probabilities when different priors are assumed. Here, we show that integrating over uncertainty using a wide range of typical prior assumptions strongly affects posterior probabilities, causing them to deviate from those that would be inferred if branch lengths were known in advance; only when there is no uncertainty to integrate over does the average posterior probability of a group of trees accurately predict the proportion of correct trees in the group. The pattern of branch lengths on the true tree determines whether integrating over uncertainty pushes posterior probabilities upward or downward. The magnitude of the effect depends on the specific prior distributions used and the length of the sequences analyzed. Under realistic conditions, however, even extraordinarily long sequences are not enough to prevent frequent inference of incorrect clades with strong support. We found that across a range of conditions, diffuse priors--either flat or exponential distributions with moderate to large means--provide more reliable inferences than small-mean exponential priors. An empirical Bayes approach that fixes branch lengths at their maximum likelihood estimates yields posterior probabilities that more closely match those that would be inferred if the true branch lengths were known in advance and reduces the rate of strongly supported false inferences compared with fully Bayesian integration.     

Tail paradox, partial identifiability, and influential priors in Bayesian branch length inference

Recent studies have observed that Bayesian analyses of sequence data sets using the program MrBayes sometimes generate extremely large branch lengths, with posterior credibility intervals for the tree length (sum of branch lengths) excluding the maximum likelihood estimates. Suggested explanations for this phenomenon include the existence of multiple local peaks in the posterior, lack of convergence of the chain in the tail of the posterior, mixing problems, and misspecified priors on branch lengths. Here, we analyze the behavior of Bayesian Markov chain Monte Carlo algorithms when the chain is in the tail of the posterior distribution and note that all these phenomena can occur. In Bayesian phylogenetics, the likelihood function approaches a constant instead of zero when the branch lengths increase to infinity. The flat tail of the likelihood can cause poor mixing and undue influence of the prior. We suggest that the main cause of the extreme branch length estimates produced in many Bayesian analyses is the poor choice of a default prior on branch lengths in current Bayesian phylogenetic programs. The default prior in MrBayes assigns independent and identical distributions to branch lengths, imposing strong (and unreasonable) assumptions about the tree length. The problem is exacerbated by the strong correlation between the branch lengths and parameters in models of variable rates among sites or among site partitions. To resolve the problem, we suggest two multivariate priors for the branch lengths (called compound Dirichlet priors) that are fairly diffuse and demonstrate their utility in the special case of branch length estimation on a star phylogeny. Our analysis highlights the need for careful thought in the specification of high-dimensional priors in Bayesian analyses.     

The devil in the details: interactions between the branch-length prior and likelihood model affect node support and branch lengths in the phylogeny of the Psoraceae

In popular use of Bayesian phylogenetics, a default branch-length prior is almost universally applied without knowing how a different prior would have affected the outcome. We performed Bayesian and maximum likelihood (ML) inference of phylogeny based on empirical nucleotide sequence data from a family of lichenized ascomycetes, the Psoraceae, the morphological delimitation of which has been controversial. We specifically assessed the influence of the combination of Bayesian branch-length prior and likelihood model on the properties of the Markov chain Monte Carlo tree sample, including node support, branch lengths, and taxon stability. Data included two regions of the mitochondrial ribosomal RNA gene, the internal transcribed spacer region of the nuclear ribosomal RNA gene, and the protein-coding largest subunit of RNA polymerase II. Data partitioning was performed using Bayes' factors, whereas the best-fitting model of each partition was selected using the Bayesian information criterion (BIC). Given the data and model, short Bayesian branch-length priors generate higher numbers of strongly supported nodes as well as short and topologically similar trees sampled from parts of tree space that are largely unexplored by the ML bootstrap. Long branch-length priors generate fewer strongly supported nodes and longer and more dissimilar trees that are sampled mostly from inside the range of tree space sampled by the ML bootstrap. Priors near the ML distribution of branch lengths generate the best marginal likelihood and the highest frequency of "rogue" (unstable) taxa. The branch-length prior was shown to interact with the likelihood model. Trees inferred under complex partitioned models are more affected by the stretching effect of the branch-length prior. Fewer nodes are strongly supported under a complex model given the same branch-length prior. Irrespective of model, internal branches make up a larger proportion of total tree length under the shortest branch-length priors compared with longer priors. Relative effects on branch lengths caused by the branch-length prior can be problematic to downstream phylogenetic comparative methods making use of the branch lengths. Furthermore, given the same branch-length prior, trees are on average more dissimilar under a simple unpartitioned model compared with a more complex partitioned models. The distribution of ML branch lengths was shown to better fit a gamma or Pareto distribution than an exponential one. Model adequacy tests indicate that the best-fitting model selected by the BIC is insufficient for describing data patterns in 5 of 8 partitions. More general substitution models are required to explain the data in three of these partitions, one of which also requires nonstationarity. The two mitochondrial ribosomal RNA gene partitions need heterotachous models. We found no significant correlations between, on the one hand, the amount of ambiguous data or the smallest branch-length distance to another taxon and, on the other hand, the topological stability of individual taxa. Integrating over several exponentially distributed means under the best-fitting model, node support for the family Psoraceae, including Psora, Protoblastenia, and the Micarea sylvicola group, is approximately 0.96. Support for the genus Psora is distinctly lower, but we found no evidence to contradict the current classification.     

Dissolving the star-tree paradox

While Bayesian methods have become very popular in phylogenetic systematics, the foundations of this approach remain controversial. The star-tree paradox in Bayesian phylogenetics refers to the phenomenon that a particular binary phylogenetic tree sometimes has a very high posterior probability even though a star tree generates the data. I argue that this phenomenon reveals an unattractive feature of the Bayesian approach to scientific inference and discuss two proposals for how to address the star-tree paradox. In particular, I defend the polytomy prior as a solution (or rather dissolution) of the paradox and argue that it is preferable to a data-size dependent branch lengths prior from a methodological perspective. However, while this reply dissolves the star-tree paradox, the general challenge to Bayesian confirmation theory remains unmet.     

Branch-length prior influences Bayesian posterior probability of phylogeny

The Bayesian method for estimating species phylogenies from molecular sequence data provides an attractive alternative to maximum likelihood with nonparametric bootstrap due to the easy interpretation of posterior probabilities for trees and to availability of efficient computational algorithms. However, for many data sets it produces extremely high posterior probabilities, sometimes for apparently incorrect clades. Here we use both computer simulation and empirical data analysis to examine the effect of the prior model for internal branch lengths. We found that posterior probabilities for trees and clades are sensitive to the prior for internal branch lengths, and priors assuming long internal branches cause high posterior probabilities for trees. In particular, uniform priors with high upper bounds bias Bayesian clade probabilities in favor of extreme values. We discuss possible remedies to the problem, including empirical and full Bayesian methods and subjective procedures suggested in Bayesian hypothesis testing. Our results also suggest that the bootstrap proportion and Bayesian posterior probability are different measures of accuracy, and that the bootstrap proportion, if interpreted as the probability that the clade is true, can be either too liberal or too conservative.     

Fair-balance paradox, star-tree paradox, and Bayesian phylogenetics

The star-tree paradox refers to the conjecture that the posterior probabilities for the three unrooted trees for four species (or the three rooted trees for three species if the molecular clock is assumed) do not approach 1/3 when the data are generated using the star tree and when the amount of data approaches infinity. It reflects the more general phenomenon of high and presumably spurious posterior probabilities for trees or clades produced by the Bayesian method of phylogenetic reconstruction, and it is perceived to be a manifestation of the deeper problem of the extreme sensitivity of Bayesian model selection to the prior on parameters. Analysis of the star-tree paradox has been hampered by the intractability of the integrals involved. In this article, I use Laplacian expansion to approximate the posterior probabilities for the three rooted trees for three species using binary characters evolving at a constant rate. The approximation enables calculation of posterior tree probabilities for arbitrarily large data sets. Both theoretical analysis of the analogous fair-coin and fair-balance problems and computer simulation for the tree problem confirmed the existence of the star-tree paradox. When the data size n --> infinity, the posterior tree probabilities do not converge to 1/3 each, but they vary among data sets according to a statistical distribution. This distribution is characterized. Two strategies for resolving the star-tree paradox are explored: (1) a nonzero prior probability for the degenerate star tree and (2) an increasingly informative prior forcing the internal branch length toward zero. Both appear to be effective in resolving the paradox, but the latter is simpler to implement. The posterior tree probabilities are found to be very sensitive to the prior.     

A Bayesian perspective on a non-parsimonious parsimony model

Several stochastic models of character change, when implemented in a maximum likelihood framework, are known to give a correspondence between the maximum parsimony method and the method of maximum likelihood. One such model has an independently estimated branch-length parameter for each site and each branch of the phylogenetic tree. This model--the no-common-mechanism model--has many parameters, and, in fact, the number of parameters increases as fast as the alignment is extended. We take a Bayesian approach to the no-common-mechanism model and place independent gamma prior probability distributions on the branch-length parameters. We are able to analytically integrate over the branch lengths, and this allowed us to implement an efficient Markov chain Monte Carlo method for exploring the space of phylogenetic trees. We were able to reliably estimate the posterior probabilities of clades for phylogenetic trees of up to 500 sequences. However, the Bayesian approach to the problem, at least as implemented here with an independent prior on the length of each branch, does not tame the behavior of the branch-length parameters. The integrated likelihood appears to be a simple rescaling of the parsimony score for a tree, and the marginal posterior probability distribution of the length of a branch is dependent upon how the maximum parsimony method reconstructs the characters at the interior nodes of the tree. The method we describe, however, is of potential importance in the analysis of morphological character data and also for improving the behavior of Markov chain Monte Carlo methods implemented for models in which sites share a common branch-length parameter.     

Adding time-calibrated branch lengths to the Asteraceae supertree

New inference techniques,such as supertrees,have improved the construction of large phylogenies,helping to reveal the tree of life.In addition,these large phylogenies have enhanced the study of other evolutionary questions,such as whether traits have evolved in a neutral or adaptive way,or what factors have influenced diversification.However,supertrees usually lack branch lengths,which are necessary for all these issues to be investigated.Here,divergence times within the largest family of flowering plants,namely the Asteraceae,are reviewed to estimate time-calibrated branch lengths in the supertree of this lineage.An inconsistency between estimated dates of basal branching events and the earliest asteraceous fossil pollen record was detected.In addition,the impact of different methods of branch length assignment on the total number of transitions between states in the reconstruction of sexual system evolution in Asteraceae was investigated.At least for this dataset,different branch length assignation approaches influenced maximum likelihood(ML)reconstructions only and not Bayesian ones.Therefore,the selection of different branch length information is not arbitrary and should be carefully assessed,at least when ML approaches are being used.The reviewed divergence times and the estimated time-calibrated branch lengths provide a useful tool for future phylogenetic comparative and macroevolutionary studies of Asteraceae.     

The accuracy of species tree estimation under simulation: a comparison of methods

Numerous simulation studies have investigated the accuracy of phylogenetic inference of gene trees under maximum parsimony, maximum likelihood, and Bayesian techniques. The relative accuracy of species tree inference methods under simulation has received less study. The number of analytical techniques available for inferring species trees is increasing rapidly, and in this paper, we compare the performance of several species tree inference techniques at estimating recent species divergences using computer simulation. Simulating gene trees within species trees of different shapes and with varying tree lengths (T) and population sizes (), and evolving sequences on those gene trees, allows us to determine how phylogenetic accuracy changes in relation to different levels of deep coalescence and phylogenetic signal. When the probability of discordance between the gene trees and the species tree is high (i.e., T is small and/or is large), Bayesian species tree inference using the multispecies coalescent (BEST) outperforms other methods. The performance of all methods improves as the total length of the species tree is increased, which reflects the combined benefits of decreasing the probability of discordance between species trees and gene trees and gaining more accurate estimates for gene trees. Decreasing the probability of deep coalescences by reducing also leads to accuracy gains for most methods. Increasing the number of loci from 10 to 100 improves accuracy under difficult demographic scenarios (i.e., coalescent units ≤ 4N(e)), but 10 loci are adequate for estimating the correct species tree in cases where deep coalescence is limited or absent. In general, the correlation between the phylogenetic accuracy and the posterior probability values obtained from BEST is high, although posterior probabilities are overestimated when the prior distribution for is misspecified.     

Long-Branch Attraction Bias and Inconsistency in Bayesian Phylogenetics

Bayesian inference (BI) of phylogenetic relationships uses the same probabilistic models of evolution as its precursor maximum likelihood (ML), so BI has generally been assumed to share ML''s desirable statistical properties, such as largely unbiased inference of topology given an accurate model and increasingly reliable inferences as the amount of data increases. Here we show that BI, unlike ML, is biased in favor of topologies that group long branches together, even when the true model and prior distributions of evolutionary parameters over a group of phylogenies are known. Using experimental simulation studies and numerical and mathematical analyses, we show that this bias becomes more severe as more data are analyzed, causing BI to infer an incorrect tree as the maximum a posteriori phylogeny with asymptotically high support as sequence length approaches infinity. BI''s long branch attraction bias is relatively weak when the true model is simple but becomes pronounced when sequence sites evolve heterogeneously, even when this complexity is incorporated in the model. This bias—which is apparent under both controlled simulation conditions and in analyses of empirical sequence data—also makes BI less efficient and less robust to the use of an incorrect evolutionary model than ML. Surprisingly, BI''s bias is caused by one of the method''s stated advantages—that it incorporates uncertainty about branch lengths by integrating over a distribution of possible values instead of estimating them from the data, as ML does. Our findings suggest that trees inferred using BI should be interpreted with caution and that ML may be a more reliable framework for modern phylogenetic analysis.     

Constraining prior probabilities of phylogenetic trees

Although Bayesian methods are widely used in phylogenetic systematics today, the foundations of this methodology are still debated among both biologists and philosophers. The Bayesian approach to phylogenetic inference requires the assignment of prior probabilities to phylogenetic trees. As in other applications of Bayesian epistemology, the question of whether there is an objective way to assign these prior probabilities is a contested issue. This paper discusses the strategy of constraining the prior probabilities of phylogenetic trees by means of the Principal Principle. In particular, I discuss a proposal due to Velasco (Biol Philos 23:455–473, 2008) of assigning prior probabilities to tree topologies based on the Yule process. By invoking the Principal Principle I argue that prior probabilities of tree topologies should rather be assigned a weighted mixture of probability distributions based on Pinelis’ (P Roy Soc Lond B Bio 270:1425–1431, 2003) multi-rate branching process including both the Yule distribution and the uniform distribution. However, I argue that this solves the problem of the priors of phylogenetic trees only in a weak form.     

Reconstructible Phylogenetic Networks: Do Not Distinguish the Indistinguishable

Phylogenetic networks represent the evolution of organisms that have undergone reticulate events, such as recombination, hybrid speciation or lateral gene transfer. An important way to interpret a phylogenetic network is in terms of the trees it displays, which represent all the possible histories of the characters carried by the organisms in the network. Interestingly, however, different networks may display exactly the same set of trees, an observation that poses a problem for network reconstruction: from the perspective of many inference methods such networks are indistinguishable. This is true for all methods that evaluate a phylogenetic network solely on the basis of how well the displayed trees fit the available data, including all methods based on input data consisting of clades, triples, quartets, or trees with any number of taxa, and also sequence-based approaches such as popular formalisations of maximum parsimony and maximum likelihood for networks. This identifiability problem is partially solved by accounting for branch lengths, although this merely reduces the frequency of the problem. Here we propose that network inference methods should only attempt to reconstruct what they can uniquely identify. To this end, we introduce a novel definition of what constitutes a uniquely reconstructible network. For any given set of indistinguishable networks, we define a canonical network that, under mild assumptions, is unique and thus representative of the entire set. Given data that underwent reticulate evolution, only the canonical form of the underlying phylogenetic network can be uniquely reconstructed. While on the methodological side this will imply a drastic reduction of the solution space in network inference, for the study of reticulate evolution this is a fundamental limitation that will require an important change of perspective when interpreting phylogenetic networks.     

Adding time‐calibrated branch lengths to the Asteraceae supertree

Abstract New inference techniques, such as supertrees, have improved the construction of large phylogenies, helping to reveal the tree of life. In addition, these large phylogenies have enhanced the study of other evolutionary questions, such as whether traits have evolved in a neutral or adaptive way, or what factors have influenced diversification. However, supertrees usually lack branch lengths, which are necessary for all these issues to be investigated. Here, divergence times within the largest family of flowering plants, namely the Asteraceae, are reviewed to estimate time‐calibrated branch lengths in the supertree of this lineage. An inconsistency between estimated dates of basal branching events and the earliest asteraceous fossil pollen record was detected. In addition, the impact of different methods of branch length assignment on the total number of transitions between states in the reconstruction of sexual system evolution in Asteraceae was investigated. At least for this dataset, different branch length assignation approaches influenced maximum likelihood (ML) reconstructions only and not Bayesian ones. Therefore, the selection of different branch length information is not arbitrary and should be carefully assessed, at least when ML approaches are being used. The reviewed divergence times and the estimated time‐calibrated branch lengths provide a useful tool for future phylogenetic comparative and macroevolutionary studies of Asteraceae.     

Are you my mother? Bayesian phylogenetic inference of recombination among putative parental strains

Reconstructing evolutionary relationships using Bayesian inference has become increasingly popular due to the ability of Bayesian inference to handle complex models of evolution. In this review we concentrate on inference of recombination events between strains of viruses when these events are sporadic, ie rare relative to point mutations. Bayesian inference is especially attractive in the detection of recombination events because it allows for simultaneous inferences about the presence, number and location of crossover points and the identification of parental sequences. Current frequentist recombination identification falls into a sequential testing trap. The most likely parental sequences and crossover points are identified using the data and then the certainty of recombination is assessed conditional on this identification. After briefly outlining basic phylogenetic models, Bayesian inference and Markov chain Monte Carlo (MCMC) computation, we summarise three different approaches to recombination detection and discuss current challenges in applying Bayesian phylogenetic inference of recombination.     

Concordance analysis in mitogenomic phylogenetics

Here I advocate the utility of Bayesian concordance analysis as a mechanism for exploring the magnitude and source of phylogenetic signal in concatenated mitogenomic phylogenetic studies. While typically applied to the study of independently evolving gene trees, Bayesian concordance analysis can also be applied to linked, but individually analyzed, gene regions using a prior probability that reflects the expectation of similar phylogenetic reconstructions. For true branches in the mitogenomic tree, concordance factors should represent the number of gene regions that contain phylogenetic signal for a particular clade. As a demonstration of the application of Bayesian concordance analysis to empirical data, I analyzed two different salamander (Hynobiidae and Plethodontidae) mitogenomic data sets using a gene-based partitioning strategy. The results revealed many strongly supported clades in the concatenated trees that have high concordance factors, permitting the inference that these are robustly resolved through phylogenetic signal distributed across the mitogenome. In contrast, a number of strongly supported clades in the concatenated tree received low concordance factors, indicating that their reconstruction is either driven primarily by phylogenetic signal in a small number of gene regions, or that they are inconsistent reconstructions influenced by properties of the data that can produce inaccurate trees (e.g., compositional bias, selection, etc.). Exploration of the Bayesian joint posterior distribution of trees highlighted partitions that contribute phylogenetic information to similar clade reconstructions. This approach was particularly insightful in the hynobiid data, where different combinations of genes were identified that support alternative tree reconstructions. Concatenated analysis of these different subsets of genes highlighted through Bayesian concordance analysis produced strongly supported and contrasting trees, demonstrating the potential for inconsistency in concatenated mitogenomic phylogenetics. The overall results presented here suggest that Bayesian concordance analysis can serve as an effective exploration of the influence of different gene regions in mitogenomic (and other organellar genomic) phylogenetic studies.     

Comparison of site-specific rate-inference methods for protein sequences: empirical Bayesian methods are superior

The degree to which an amino acid site is free to vary is strongly dependent on its structural and functional importance. An amino acid that plays an essential role is unlikely to change over evolutionary time. Hence, the evolutionary rate at an amino acid site is indicative of how conserved this site is and, in turn, allows evaluation of its importance in maintaining the structure/function of the protein. When using probabilistic methods for site-specific rate inference, few alternatives are possible. In this study we use simulations to compare the maximum-likelihood and Bayesian paradigms. We study the dependence of inference accuracy on such parameters as number of sequences, branch lengths, the shape of the rate distribution, and sequence length. We also study the possibility of simultaneously estimating branch lengths and site-specific rates. Our results show that a Bayesian approach is superior to maximum-likelihood under a wide range of conditions, indicating that the prior that is incorporated into the Bayesian computation significantly improves performance. We show that when branch lengths are unknown, it is better first to estimate branch lengths and then to estimate site-specific rates. This procedure was found to be superior to estimating both the branch lengths and site-specific rates simultaneously. Finally, we illustrate the difference between maximum-likelihood and Bayesian methods when analyzing site-conservation for the apoptosis regulator protein Bcl-x(L).     

Statistical measures of uncertainty for branches in phylogenetic trees inferred from molecular sequences by using model-based methods

In recent years, the emphasis of theoretical work on phylogenetic inference has shifted from the development of new tree inference methods to the development of methods to measure the statistical support for the topologies. This paper reviews 3 approaches to assign support values to branches in trees obtained in the analysis of molecular sequences: the bootstrap, the Bayesian posterior probabilities for clades, and the interior branch tests. In some circumstances, these methods give different answers. It should not be surprising: their assumptions are different. Thus the interior branch tests assume that a given topology is true and only consider if a particular branch length is longer than zero. If a tree is incorrect, a wrong branch (a low bootstrap or Bayesian support may be an indication) may have a non-zero length. If the substitution model is oversimplified, the length of a branch may be overestimated, and the Bayesian support for the branch may be inflated. The bootstrap, on the other hand, approximates the variance of the data under the real model of sequence evolution, because it involves direct resampling from this data. Thus the discrepancy between the Bayesian support and the bootstrap support may signal model inaccuracy. In practical application, use of all 3 methods is recommended, and if discrepancies are observed, then a careful analysis of their potential origins should be made.