Approximate likelihood-ratio test for branches: A fast, accurate, and powerful alternative

Area of Science:

• Computational Biology
• Evolutionary Biology
• Bioinformatics

Background:

• Statistical tests are crucial for reconstructing evolutionary trees from molecular data.
• Existing methods like nonparametric bootstrap and Bayesian estimation have limitations in speed or accuracy.

Purpose of the Study:

• To introduce a novel, fast, and accurate statistical test for evaluating branches in molecular evolutionary trees.
• To present the approximate likelihood-ratio test (aLRT) as a competitive alternative to current branch support methods.

Main Methods:

• Developed an approximate likelihood-ratio test (aLRT) based on the conventional LRT, with a null hypothesis of zero branch length.
• Approximated the test statistic as twice the difference between maximum log-likelihood values of the best and second-best tree topologies.
• Implemented the aLRT within the PHYML program for fast maximum likelihood tree estimation.

Main Results:

• The aLRT statistic is asymptotically distributed as a maximum of three random variables from the chi(0)2 + chi(1)2 distribution.
• The test is computationally efficient as it optimizes only a subset of parameters for the second-best topology.
• Evaluated on simulated data, the aLRT demonstrated accuracy, power, and robustness to model assumption violations.

Conclusions:

• The aLRT is a fast, accurate, and robust statistical method for assessing branch support in molecular phylogenetics.
• It serves as a strong alternative to traditional methods, enhancing the reliability of evolutionary tree reconstruction.