Robustness to divergence time underestimation when inferring species trees from estimated gene trees

Area of Science:

• Phylogenetics
• Computational Biology
• Evolutionary Genomics

Background:

• Inferring species trees from gene trees is crucial for understanding evolutionary relationships.
• Phylogenomic datasets with numerous loci necessitate computationally efficient inference methods.
• Existing methods like MP-EST, STAR, STEAC, STELLS, and STEM vary in performance based on gene tree estimation techniques.

Purpose of the Study:

• To evaluate the performance of several computationally efficient species tree inference methods.
• To compare the effectiveness of topology-based versus coalescent-based methods.
• To investigate the reasons behind the performance discrepancies, particularly for the STEM method.

Main Methods:

• Comparative analysis of five species tree inference methods: MP-EST, STAR, STEAC, STELLS, and STEM.
• Utilizing gene trees estimated via Maximum Likelihood (ML) and Bayesian approaches.
• Employing simulations and a great ape dataset to assess accuracy, bias, and resolution.

Main Results:

• Topology-based methods generally performed better with ML-estimated gene trees.
• Coalescent-based methods showed superior performance with Bayesian-estimated gene trees.
• MP-EST, STAR, STEAC, and STELLS consistently outperformed STEM across most conditions.
• STEM's reduced accuracy was linked to underestimation of coalescence times when using ML gene trees.

Conclusions:

• The choice of gene tree estimation method (ML vs. Bayesian) significantly impacts species tree inference performance.
• Coalescent-based methods, particularly MP-EST, STAR, STEAC, and STELLS, are recommended for robust species tree inference.
• Underestimation of coalescence times in ML gene trees can introduce bias and reduce resolution in species tree estimation.