A comparative study of SVDquartets and other coalescent-based species tree estimation methods

Area of Science:

• Phylogenetics
• Computational Biology
• Evolutionary Biology

Background:

• Species tree estimation is complicated by incomplete lineage sorting (ILS), which causes gene trees to differ from the species tree.
• Coalescent-based summary methods offer theoretical robustness to ILS but can struggle with short gene sequences.
• SVDquartets was introduced as a coalescent-based method using quartet analysis, but its accuracy relative to other methods was unassessed.

Purpose of the Study:

• To compare the accuracy of SVDquartets against leading coalescent-based methods (ASTRAL-2, NJst) and concatenation.
• To evaluate method performance across varying levels of incomplete lineage sorting (ILS), taxon numbers, and gene sequence lengths.

Main Methods:

• Comparative analysis using simulated datasets.
• Evaluation of SVDquartets, ASTRAL-2, NJst, and maximum likelihood concatenation.
• Testing across diverse ILS levels, taxon numbers, and sequence lengths (down to 10 sites).

Main Results:

• ASTRAL-2 generally demonstrated the highest accuracy under higher ILS conditions.
• Concatenation was the most accurate method under the lowest ILS conditions.
• SVDquartets performed competitively with leading methods under low ILS and short sequence lengths.

Conclusions:

• ASTRAL-2 is recommended for high ILS scenarios, while concatenation is best for low ILS.
• SVDquartets offers a viable alternative for specific conditions (low ILS, short sequences).
• ASTRAL-2's robustness to short sequences was notable and somewhat unexpected.