Alignment errors strongly impact likelihood-based tests for comparing topologies.

Area of Science:

• Phylogenetics
• Computational Biology
• Statistical Modeling

Background:

• Estimating phylogenetic trees from sequence data is crucial in evolutionary biology.
• The maximum-likelihood paradigm provides a point estimate for the best tree.
• Hypothesis testing is required to assess data support for evolutionary scenarios, leading to the development of the Kishino-Hasegawa (KH) test.

Purpose of the Study:

• To investigate the biases of the Kishino-Hasegawa (KH) test in phylogenetic tree selection.
• To identify the impact of alignment errors on the reliability of the KH test.
• To propose and evaluate methods for correcting these biases.

Main Methods:

• Simulations were used to demonstrate the bias in the KH test.
• The effect of alignment error on topology selection was analyzed.
• Two bias correction methods were proposed: alignment column filtering and a novel parametric test.

Main Results:

• The KH test is shown to be biased by alignment errors, often favoring the guide tree used for alignment.
• Branch length optimization makes the KH test overly conservative.
• Alignment column removal reduces bias but also decreases statistical power.
• The proposed parametric test effectively eliminates bias without data filtering.

Conclusions:

• The standard Kishino-Hasegawa (KH) test is unreliable when sequence alignment contains errors.
• A new parametric hypothesis testing framework is introduced to accurately compare phylogenetic trees.
• This method accounts for alignment uncertainty, providing more robust phylogenetic inference.