Computational Performance and Statistical Accuracy of *BEAST and Comparisons with Other Methods

Area of Science:

• Phylogenetics and molecular evolution
• Computational biology and bioinformatics

Background:

• Gene trees evolve independently within a species tree under the multispecies coalescent model.
• Supermatrix concatenation methods incorrectly assume a single genealogical history for all genes.
• Previous studies support the multispecies coalescent model and question concatenation's accuracy.

Purpose of the Study:

• To analyze the computational scaling and statistical accuracy of *BEAST, a Bayesian implementation of the multispecies coalescent.
• To quantitatively predict the impact of increasing the number of gene loci on *BEAST's performance.
• To compare *BEAST with concatenation methods using simulated and empirical phylogenomic data.

Main Methods:

• Characterized *BEAST's scaling behavior using simulation studies.
• Performed follow-up simulations varying parameters like branch length and number of loci.
• Compared *BEAST and concatenation methods using simulations based on estimated parameters from two phylogenomic datasets.

Main Results:

• *BEAST's statistical performance relative to concatenation improves with reduced branch lengths and increased numbers of loci.
• Using *BEAST with tens of loci can be more accurate than concatenation with thousands of loci.
• The study provides insights into the practicalities of Bayesian species tree estimation.

Conclusions:

• The fully Bayesian multispecies coalescent model, as implemented in *BEAST, can outperform supermatrix or summary methods.
• The number of loci required for accurate species tree estimation is dependent on the method used.
• This research clarifies the trade-offs between computational intensity and statistical accuracy in phylogenomic analyses.