Quantification and relative severity of inflated branch-support values generated by alternative methods: an empirical...

Area of Science:

• Phylogenetics
• Molecular Evolution
• Bioinformatics

Background:

• Phylogenetic analyses aim to reconstruct evolutionary relationships using molecular data.
• Parametric methods like Bayesian MCMC and maximum likelihood are widely used but susceptible to various error sources.
• Supermatrices with missing data present challenges for accurate phylogenetic reconstruction.

Purpose of the Study:

• To quantify sources of error in heuristic parametric phylogenetic analyses using a large supermatrix.
• To evaluate the impact of tree search strategies and resampling methods on phylogenetic resolution and support.
• To investigate the reliability of data partitioning and missing data handling in likelihood analyses.

Main Methods:

• Construction of a supermatrix from 272 terminals of the Rubiaceae tribe Spermacoceae across 10 gene regions.
• Application of Bayesian MCMC and maximum likelihood phylogenetic methods.
• Examination of error sources including tree search quality, resampling artifacts (undersampling-within-replicates, frequency-within-replicates), and missing data.

Main Results:

• Clades lacking unambiguous synapomorphies are less robust to added data in parametric analyses.
• Tree search methods prone to artifacts often yield high resolution and support, irrespective of true phylogenetic signal.
• Data partitioning does not consistently resolve issues caused by missing data in likelihood analyses with linked branch lengths.
• Resampling artifacts are present in both parsimony and likelihood analyses, necessitating specific handling strategies.

Conclusions:

• The robustness of phylogenetic inference depends on the presence of unambiguous synapomorphies.
• Certain tree search and resampling methods can generate artifactual resolution and support.
• Missing data and linked branch lengths in likelihood models can lead to unreliable phylogenetic outcomes.
• Appropriate bootstrap methods and saving multiple trees are recommended to mitigate resampling artifacts in parametric phylogenetic analyses.