Biases of tree-independent-character-subsampling methods

Area of Science:

• Phylogenetics
• Computational Biology
• Evolutionary Biology

Background:

• Observed Variability (OV) and Tree Independent Generation of Evolutionary Rates (TIGER) are tree-independent methods for estimating evolutionary rates and excluding characters.
• These methods are often used to improve phylogenetic inference by removing rapidly evolving sites.
• Previous studies suggested OV and TIGER can yield better phylogenetic estimates than using all characters.

Purpose of the Study:

• To critically evaluate the reliability and potential biases of OV and TIGER methods in phylogenetics.
• To demonstrate the systematic biases inherent in OV and TIGER when applied to phylogenomic data.
• To provide a cautionary note on the interpretation of phylogenetic results derived from these character-exclusion methods.

Main Methods:

• Utilized four sets of simulations to test OV and TIGER methods.
• Applied OV and TIGER to an empirical phylogenomic dataset.
• Analyzed character distributions, character-state space, and clade support under character conflict.

Main Results:

• Demonstrated that OV and TIGER exhibit systematic bias against characters with symmetric state distributions and large character-state spaces.
• Showed these methods can favor convergence and reversals over synapomorphy, exacerbating long-branch attraction.
• Revealed that OV and TIGER can produce conflicting phylogenetic inferences dependent on taxon sampling.

Conclusions:

• Neither OV nor TIGER can be reliably used to enhance the phylogenetic signal-to-noise ratio in data matrices.
• Skepticism is warranted for phylogenetic results based on OV/TIGER character deletion, especially when small clades are supported post-deletion but contradicted by the full dataset.
• These methods may introduce systematic errors, potentially leading to incorrect evolutionary hypotheses.