Divergent maximum-likelihood-branch-support values for polytomies.

Area of Science:

• Phylogenetics
• Computational Biology
• Evolutionary Biology

Background:

• Phylogenetic inference aims to reconstruct evolutionary history.
• Branch support values quantify the reliability of inferred evolutionary relationships.
• Conflicting evolutionary signals can lead to inaccurate branch support.

Purpose of the Study:

• To evaluate branch support methods in phylogenetic inference under conflicting data.
• To assess the performance of bootstrapping and likelihood-based tests for polytomies.
• To identify potential inflation of false positive branch support.

Main Methods:

• Simulations of 4-taxon datasets with 3-way character conflict (hard polytomy).
• Application and comparison of bootstrapping and SH-like approximate likelihood-ratio test (aLRT).
• Focus on likelihood-based phylogenetic inference programs, including RAxML.

Main Results:

• Bayesian MCMCMC posterior probabilities and RAxML rapid bootstrapping (GTRCAT model) can yield inflated support values for polytomies.
• The SH-like aLRT demonstrates superior performance over bootstrapping in identifying true polytomies.
• A 70% bootstrap cutoff does not reliably indicate a 5% error rate or 95% accuracy.

Conclusions:

• Rapid bootstrapping with the GTRCAT model in RAxML should be avoided due to inflated and imprecise support values.
• The SH-like aLRT is recommended for likelihood-based phylogenetic studies to complement bootstrapping by collapsing unreliable branches (≤ 10%).
• The assumption that a 70% bootstrap value equates to 95% accuracy should be abandoned.