Phylogenetic Factor Analysis.

Area of Science:

• Evolutionary Biology
• Phylogenetics
• Quantitative Genetics

Background:

• Phylogenetic comparative methods analyze trait evolution considering shared ancestry.
• Traditional methods using Brownian diffusion struggle with high-dimensional traits due to complex correlation structures.
• Inferring pairwise correlations in multivariate diffusion models is computationally limiting for large trait sets.

Purpose of the Study:

• To introduce Phylogenetic Factor Analysis (PFA) as a method for analyzing high-dimensional traits in an evolutionary context.
• To address the limitations of multivariate diffusion models in inferring trait correlations.
• To develop a Bayesian framework that accounts for trait number uncertainty, mixed data types, missing data, and phylogenetic uncertainty.

Main Methods:

• Developed a Bayesian Phylogenetic Factor Analysis (PFA) model.
• Utilized Gibbs samplers with dynamic programming for efficient posterior distribution estimation.
• Proposed a novel marginal likelihood estimator for models including discrete traits.

Main Results:

• PFA efficiently estimates evolutionary factors and trait groupings, outperforming multivariate diffusion.
• The method demonstrates a significant speed increase (over 3-fold) compared to multivariate diffusion, especially with latent traits.
• PFA shows a better model fit across diverse evolutionary case studies: columbine flower development, placental reproduction, and triggerfish fin morphometry.

Conclusions:

• Phylogenetic Factor Analysis (PFA) provides a powerful and efficient framework for studying high-dimensional trait evolution.
• PFA overcomes limitations of existing methods by handling complex correlations, mixed data types, and phylogenetic uncertainty.
• The proposed methods offer improved computational efficiency and model fit for evolutionary analyses.