Are Unequal Clade Priors Problematic for Bayesian Phylogenetics

摘要: Although Bayesian phylogenetic methodologies were first developed in the 1960s (Felsenstein, 1968, 2004), approach remained relatively obscure until initial release of software application MrBayes (Huelsenbeck and Ronquist, 2001). Since that time, popularity phylogenetics has increased tremendously, it now must be considered a primary method analysis on par with maximum likelihood, parsimony, distance methods. The can attributed to computational efficiencies allow for explicit model-based analyses large data sets real time simultaneous estimation nodal support form posterior probability values. Despite enthusiasm generated by availability fast likelihood-based approach, remains somewhat controversial. Much controversy is focused two related issues: (1) relationship between values nonparametric bootstrap proportions nagging suspicion probabilities are too liberal (e.g., Suzuki et al., 2002), (2) influence prior probabilities, especially so-called flat or uninformative priors, resulting posteriors 2004; Zwickl Holder, Pickett Randle, 2005). there been spate simulation studies published during past 2 years, most (Alfaro 2003; Cummings al. Douady Erixon Huelsenbeck Rannala, Wilcox 2002) have proportions. relative impact priors only recently received detailed study required determine if current implementations appropriate and, not, how they might corrected Lewis requires designation each parameter including those alternative tree topologies, branch lengths, nucleotide substitution model. In case, we usually little priori information would us select an informative distribution, thus researchers generally attempt accommodate their ignorance applying priors. Because proportional product truly should likelihood function drive outcome 2002; Lewis, 2001a; 2004). Unfortunately, notoriously difficult (see Kass Wasserman, 1996; advocates proceed hope will overwhelm inappropriately when cannot avoided. viability may depend inferences being robust these unavoidably recent article, Randle (2005; hereafter referred as “PR” sake brevity) provide one investigations also They correctly recognized topology does not result clade (we note distribution clades viewed either joint over all splits, marginal individual split. Here concerned former interpretation). This point was clearly illustrated PR simple example—if considers fully bifurcating five-taxon tree, 15 reconstructions linking possible pair taxa 9 any combination three taxa. Thus, rooted trees, larger smaller greater than intermediate size. All else equal, inflated presented examples this phenomenon analyzing both empirical DNA contrived sets. We focus because believe results indicate