Phylogenetic minimum description length: an optimality criterion based on algorithmic complexity.

Area of Science:

• Computational Biology
• Evolutionary Biology
• Bioinformatics

Background:

• Phylogenetic analysis aims to reconstruct evolutionary histories.
• Existing optimality criteria (e.g., parsimony, likelihood) have limitations in data weighting and model selection.
• There is a need for a unified criterion that accounts for phylogenetic graph complexity.

Purpose of the Study:

• To introduce Phylogenetic Minimum Description Length (PMDL) as a novel optimality criterion for phylogenetic analysis.
• To demonstrate PMDL's ability to generate natural weighting functions for diverse phylogenetic data and models.
• To show PMDL's capacity to incorporate phylogenetic graph complexity for hypothesis comparison.

Main Methods:

• PMDL is based on algorithmic (Kolmogorov) information theory and the minimum description length principle.
• Heuristic methods are developed to estimate the uncomputable algorithmic information content of phylogenetic hypotheses.
• The criterion is shown to generalize and converge with existing inference methods under specific conditions.

Main Results:

• PMDL provides a unified framework that naturally incorporates data and model weighting.
• The criterion allows for the comparison of hypotheses involving various phylogenetic graph types (trees, networks, forests).
• PMDL facilitates analytical model choice alongside phylogenetic graph selection, preventing over-parameterization.

Conclusions:

• PMDL represents a generalized and powerful optimality criterion for phylogenetic inference.
• It offers a principled approach to model selection and hypothesis testing in phylogenetics.
• The heuristic methods provide practical means to apply PMDL despite its theoretical uncomputability.