Phycas

Phycas performs Bayesian phylogenetic analysis of nucleotide sequence data to estimate evolutionary relationships and compare phylogenetic models using marginal likelihoods and Bayes Factors.

Key Features:

• Marginal likelihood estimation: Focuses on estimating marginal likelihoods for Bayes Factor model comparison.
• Thermodynamic Integration and Generalized Steppingstone: Implements Thermodynamic Integration and Generalized Steppingstone estimators as alternatives to the Harmonic Mean estimator for marginal likelihoods.
• Posterior predictive approaches: Supports Gelfand-Ghosh and Conditional Predictive Ordinates for model selection and assessment.
• Substitution models: Supports the General Time Reversible (GTR) family and a codon model.
• Data partitioning: Allows data partitioning with all parameters unlinked except for tree topology and edge lengths.
• Tree topology options: Supports analyses permitting polytomous and fully resolved trees.
• Edge length priors: Provides several options for edge length priors.
• Hierarchical and compound Dirichlet prior: Includes a hierarchical model and the compound Dirichlet prior to mitigate overly informative induced priors on tree length.
• Implementation: Implemented primarily in C++ with a Python interface.

Scientific Applications:

• Evolutionary relationship inference: Inferring phylogenetic relationships from nucleotide sequence data.
• Bayesian model selection: Comparing phylogenetic models using marginal likelihoods and Bayes Factors.
• Model fit assessment: Evaluating model fit using Gelfand-Ghosh and Conditional Predictive Ordinates posterior predictive approaches.
• Partitioned-data analyses: Modeling complex evolutionary scenarios with partitioned data and unlinked parameters.
• Tree resolution and prior sensitivity: Investigating polytomous versus fully resolved trees and the effects of different edge length priors.

Methodology:

Uses Thermodynamic Integration and Generalized Steppingstone estimators (with Harmonic Mean estimator noted for comparison), Gelfand-Ghosh and Conditional Predictive Ordinates posterior predictive approaches, supports GTR family and codon substitution models, data partitioning with parameters unlinked except tree topology and edge lengths, options for polytomous or fully resolved trees, multiple edge length priors, a hierarchical model and a compound Dirichlet prior; implemented primarily in C++ with a Python interface.