On reduced amino acid alphabets for phylogenetic inference

Area of Science:

• Evolutionary biology
• Computational biology
• Bioinformatics

Background:

• Markov models of evolution are crucial for phylogenetic analysis.
• Model misspecification and saturation can bias evolutionary inferences.
• Reduced amino acid alphabets offer a potential solution to these issues.

Purpose of the Study:

• To investigate the effectiveness of Markov models using reduced amino acid alphabets (bins).
• To develop and evaluate algorithms for automated bin construction.
• To assess the impact of binning on phylogenetic tree estimation.

Main Methods:

• Development of algorithms for binning amino acids based on rate matrices and alignment properties.
• Simulations to compare binning approaches with standard methods and missing data approaches.
• Application of binning methods to real biological datasets exhibiting compositional heterogeneity and saturation.

Main Results:

• Binning amino acids results in minimal information loss, even without model misspecification.
• Markov models utilizing binned amino acids perform comparably to more complex missing data methods.
• Binning demonstrably improves topological accuracy in phylogenetic estimation for real datasets.

Conclusions:

• Reduced amino acid alphabets provide a practical strategy to mitigate model misspecification and saturation in evolutionary analyses.
• Automated binning algorithms offer an efficient way to implement this approach.
• This method enhances the reliability of phylogenetic tree reconstruction in challenging biological scenarios.