Model selection for mixtures of mutagenetic trees

Area of Science:

• Computational biology
• Evolutionary genetics
• Virology

Background:

• HIV drug resistance arises from accumulating mutations.
• Mutagenetic trees model the order and rate of these mutations.
• Mixture models of mutagenetic trees offer greater flexibility for complex mutational patterns.

Purpose of the Study:

• To address the model selection challenge in K-mutagenetic trees mixture models.
• To evaluate existing model selection criteria and propose a novel approach.
• To improve the accuracy and efficiency of modeling HIV evolution.

Main Methods:

• Evaluation of classical criteria: cross-validation, Bayesian Information Criterion (BIC), Akaike Information Criterion.
• Application of empirical Bayes methods to derive posterior model probabilities.
• Development of a new model selection criterion based on mixture model redundancy and BIC modification.

Main Results:

• Classical criteria often select overly complex models with too many tree components.
• Cross-validation and the modified BIC effectively recover the correct number of trees and their topologies.
• The modified BIC offers comparable accuracy to cross-validation but with significantly reduced computational runtime.

Conclusions:

• The proposed modified BIC is a robust and efficient criterion for selecting K-mutagenetic trees mixture models.
• This method improves the understanding of HIV evolution and aids in antiretroviral therapy design.
• The modified BIC is suitable for large datasets where cross-validation is computationally prohibitive.