Interior-branch and bootstrap tests of phylogenetic trees

Area of Science:

• Phylogenetics
• Computational Biology
• Statistical Genetics

Background:

• Phylogenetic trees are crucial for understanding evolutionary relationships.
• Interior-branch and bootstrap tests are commonly used to assess the statistical support of branches in phylogenetic trees.
• These tests provide confidence values (PC and PB) often interpreted similarly, but their statistical properties differ.

Purpose of the Study:

• To compare the statistical properties of interior-branch and bootstrap tests in phylogenetic tree construction using the neighbor-joining method.
• To analyze the relationship between confidence values (PC and PB) under varying tree topologies and branch lengths.
• To evaluate the reliability of these tests, particularly concerning the number of sequences analyzed.

Main Methods:

• Utilized the neighbor-joining tree-building method.
• Analyzed statistical properties of interior-branch (PC) and bootstrap (PB) tests.
• Developed a modified interior-branch test (P'C) for estimated trees.
• Employed a model where evolutionary distances follow a multivariate normal distribution to enable analytical comparisons.

Main Results:

• The interior-branch test's PC value is the complement of the standard statistical P-value, while PB is not.
• The bootstrap test (PB) often underestimates statistical support for species clusters.
• The difference between PC and PB is most pronounced in starlike trees and increases with the number of sequences.
• The interior-branch test is statistically consistent regardless of sequence number for predetermined trees.
• The standard PC value can overestimate confidence for estimated trees; the modified P'C provides a conservative estimate.

Conclusions:

• The interior-branch test is generally preferable to the bootstrap test for assessing phylogenetic tree reliability when unbiased evolutionary distance estimators are used.
• The modified interior-branch test (P'C) offers a more reliable, conservative confidence estimate for trees derived from data.
• Understanding the distinct statistical behaviors of these tests is vital for accurate phylogenetic inference.