Branch length estimation and divergence dating: estimates of error in Bayesian and maximum likelihood frameworks

Area of Science:

• Phylogenetics and Evolutionary Biology
• Molecular Evolution
• Computational Biology

Background:

• Estimating species divergence dates is crucial for understanding evolutionary processes.
• Accurate molecular genetic differences (branch lengths) are essential for reliable divergence date estimates.
• Previous studies have not fully explored factors affecting branch length estimation accuracy.

Purpose of the Study:

• To evaluate the impact of dataset size, branch length heterogeneity, branch depth, and analytical framework on branch length estimation accuracy.
• To assess how inaccurate branch length estimation affects divergence date estimates using empirical data.

Main Methods:

• Simulations were used to test branch length estimation under various conditions.
• Reanalysis of an empirical dataset (plethodontid salamanders) was performed using different estimation frameworks.
• Branch length estimation accuracy was assessed across different phylogenetic frameworks (Bayesian and Maximum Likelihood).

Main Results:

• Branch length estimation accuracy varied significantly with dataset size, complexity, and analytical framework.
• Bayesian frameworks tended to underestimate longer branches, while Maximum Likelihood (ML) frameworks showed slight overestimation.
• ML frameworks provided more accurate estimates for complex datasets, especially with longer sequences (> or = 1 kb).

Conclusions:

• Branch lengths are often misestimated in simple datasets using both Bayesian and ML frameworks.
• ML frameworks offer more accurate branch length estimates for complex datasets, even with shorter sequences (<1 substitution/site).
• Inaccurate branch length estimation, particularly from Bayesian methods, can significantly alter divergence date estimates, necessitating cautious interpretation of results outside optimal parameters.