Likelihood-based tests of topologies in phylogenetics.

Area of Science:

• Evolutionary biology
• Statistical genetics
• Phylogenetic analysis

Background:

• Likelihood-based statistical tests for comparing evolutionary hypotheses (tree topologies) have existed for about ten years.
• The Kishino-Hasegawa test is the most frequently employed method for such comparisons.
• However, its validity is contingent upon specific assumptions that limit its practical application.

Purpose of the Study:

• To review the theoretical underpinnings of the Kishino-Hasegawa test.
• To highlight the limitations and potential biases of the Kishino-Hasegawa test in common applications.
• To advocate for the use of alternative statistical tests in phylogenetic analysis.

Main Methods:

• Theoretical review of the Kishino-Hasegawa test's assumptions and validity conditions.
• Analysis of scenarios where the Kishino-Hasegawa test may yield biased results, particularly when the maximum likelihood topology is tested.
• Review and discussion of appropriate nonparametric and parametric alternative statistical tests.

Main Results:

• The Kishino-Hasegawa test is only valid when tree topologies are specified a priori.
• The test is often severely biased when applied to cases where the topology was selected based on the data (e.g., maximum likelihood topology).
• Previously published results using invalid applications of the Kishino-Hasegawa test require cautious re-evaluation.

Conclusions:

• The Kishino-Hasegawa test should not be used in its majority of popular applications due to potential bias.
• Researchers should exercise extreme caution with prior results obtained from invalid applications of this test.
• Future phylogenetic analyses should employ validated alternative statistical tests for comparing evolutionary hypotheses.