Reliability of Bayesian posterior probabilities and bootstrap frequencies in phylogenetics

Area of Science:

• Phylogenetics
• Computational Biology
• Statistical Inference

Background:

• Nonparametric bootstrapping and Bayesian posterior probabilities are widely used for phylogenetic tree inference.
• Empirical studies report discrepancies in branch support values between these two methods.
• Theoretical predictions suggest approximate equivalence, which is often not observed.

Purpose of the Study:

• To investigate the differences between nonparametric bootstrapping and Bayesian posterior probabilities in phylogenetic analysis.
• To evaluate the impact of model parameterization on these differences.
• To compare error rates associated with each method.

Main Methods:

• Simulated phylogenetic datasets were generated.
• Analyses were performed using maximum likelihood bootstrapping and Bayesian phylogenetic inference.
• Identical evolutionary models and reoptimized parameter values were used for both methods.

Main Results:

• Bayesian posterior probabilities yielded significantly higher support values for true clades than nonparametric bootstrap frequencies.
• Higher rates of erroneous conclusions were observed with Bayesian posterior probabilities.
• Underparameterized models exacerbated these errors, while correct models led to more conservative estimates for both methods, with Bayesian being less so.

Conclusions:

• Bayesian posterior probabilities and nonparametric bootstrapping provide different measures of phylogenetic support.
• The choice of evolutionary model significantly influences the reliability of these support values.
• Careful model selection and interpretation of support values are crucial in phylogenetic studies.