ModL: exploring and restoring regularity when testing for positive selection.

Area of Science:

• Molecular evolution
• Bioinformatics
• Statistical genetics

Background:

• Likelihood ratio tests (LRTs) are standard for detecting positive selection in protein evolution.
• Current methods often rely on chi-square or mixture distributions for threshold determination, despite known statistical irregularities.
• This can lead to inaccurate Type I error rates, potentially misidentifying proteins under selection.

Purpose of the Study:

• To investigate the statistical validity of commonly used thresholds in LRTs for positive selection.
• To develop and evaluate a modified LRT that addresses statistical irregularities.
• To improve the accuracy of detecting positive selection in protein evolution.

Main Methods:

• Theoretical analysis of LRTs under positive selection models.
• Development of a modified LRT to restore statistical regularity.
• Extensive simulations to assess Type I error rates and statistical power.
• Investigation of parameter estimation challenges in mixture models of codon evolution.

Main Results:

• Commonly used LRT thresholds can result in inflated Type I error rates, contrary to expectations of conservatism.
• The modified LRT demonstrates approximately correct Type I error probabilities across various parameter settings.
• Simulations confirm that the modified LRT achieves accuracy without sacrificing statistical power.
• Parameter estimation for mixture models can be difficult with limited data (few taxa or short trees).

Conclusions:

• The modified LRT is a more statistically rigorous and reliable method for detecting positive selection in proteins.
• Accurate error rates are crucial for robust evolutionary inference.
• The study highlights challenges and offers solutions for mixture model applications in molecular evolution.
• The developed methods are broadly applicable to various problems in evolutionary biology.