Topology testing of phylogenies using least squares methods

Area of Science:

• Phylogenetics
• Computational Biology
• Bioinformatics

Background:

• Least squares (LS) methods are used for constructing confidence sets of phylogenetic trees by fitting patristic distances to observed distances.
• Generalized least squares (GLS) is computationally intensive due to covariance matrix calculations, often requiring approximations.
• Weighted least squares (WLS) simplifies calculations by omitting covariances, offering an efficient approximation.

Purpose of the Study:

• To evaluate the applicability of LS approaches for building confidence sets of trees.
• To compare the accuracy and reliability of WLS and GLS methods in phylogenetic analyses.

Main Methods:

• Assessed the WLS method's performance in analyzing biological sequences and DNA-DNA hybridization data.
• Investigated the computational feasibility and statistical properties of the GLS method.
• Utilized simulated sequence data to evaluate tree inclusion in confidence sets.

Main Results:

• WLS approximations did not compromise accuracy or reliability for biological sequence and DNA-DNA hybridization data.
• GLS faced computational challenges, failing to calculate statistics for many datasets.
• GLS included all trees in confidence sets for some datasets and showed undercoverage for simulated sequences.

Conclusions:

• WLS serves as a computationally efficient approximation to GLS for confidence set construction.
• WLS is particularly valuable for exploratory phylogenetic analyses and assessing phylogenetic signal.
• WLS is a viable alternative when other phylogenetic tree-testing methods are unavailable.