Estimation of confidence in phylogeny: the complete-and-partial bootstrap technique

Area of Science:

• Phylogenetics
• Statistical methods
• Computational biology

Background:

• The bootstrap is a common statistical method for assessing confidence in phylogenetic trees.
• Existing bootstrap tests may have reduced power and underestimate confidence levels, especially with multiple alternative topologies.

Purpose of the Study:

• To evaluate the accuracy of the bootstrap method in estimating confidence limits for phylogenies.
• To propose a novel technique for obtaining unbiased confidence estimates in phylogenetic analysis.

Main Methods:

• Comparison of bootstrap test power against Felsenstein's C and S tests.
• Development and application of the complete-and-partial bootstrap technique.
• Utilizing a multinomial model for estimating confidence levels and competing topologies.

Main Results:

• Standard bootstrap proportions underestimate confidence levels when three or more alternative topologies exist.
• The degree of underestimation increases with the number of competing topologies.
• The complete-and-partial bootstrap technique provides an unbiased estimate of confidence levels.

Conclusions:

• The standard bootstrap method requires careful selection of critical values to avoid underestimation of confidence.
• The complete-and-partial bootstrap technique offers a robust and unbiased approach for confidence estimation in phylogenetics.
• This new method is compatible with maximum parsimony and neighbor-joining tree reconstruction.