Performance of criteria for selecting evolutionary models in phylogenetics: a comprehensive study based on simulated...

Area of Science:

• Evolutionary biology
• Computational phylogenetics
• Bioinformatics

Background:

• Maximum-likelihood and Bayesian inference are standard phylogenetic methods requiring explicit evolutionary models.
• Nucleotide substitution model selection is crucial; incorrect models can lead to erroneous phylogenetic inference.
• Evaluating model-selection criteria performance is essential for reliable phylogenetic analyses.

Purpose of the Study:

• To assess the accuracy, precision, dissimilarity, and biases of different model-selection criteria.
• To identify the most appropriate criteria for selecting nucleotide substitution models in phylogenetic studies.

Main Methods:

• Analysis of 33,600 simulated DNA sequence datasets.
• Comparison of four model-selection criteria: hierarchical likelihood-ratio test (hLRT), Akaike information criterion (AIC), Bayesian information criterion (BIC), and decision theory (DT).
• Evaluation metrics included accuracy, precision, dissimilarity, and bias.

Main Results:

• Bayesian information criterion (BIC) and decision theory (DT) demonstrated superior accuracy and precision.
• High dissimilarity in model selection was observed between hLRT and AIC.
• hLRT performed poorly with models including a proportion of invariable sites; BIC and DT showed consistent performance.

Conclusions:

• Bayesian information criterion (BIC) and decision theory (DT) are recommended for model selection in phylogenetic analyses.
• Accurate model selection, combined with model-adequacy tests, enhances the reliability of phylogenetic inference.
• Improved model selection practices will advance the accuracy of evolutionary studies using DNA sequence data.