A Penalized Likelihood Framework for High-Dimensional Phylogenetic Comparative Methods and an Application to...

Area of Science:

• Evolutionary Biology
• Phylogenetics
• Quantitative Genetics

Background:

• High-dimensional phylogenetic comparative data (many traits, few species) pose statistical and computational challenges for traditional methods.
• Existing methods for large p, small n scenarios have limitations in performance and applicability.
• Accurate estimation of evolutionary parameters and model selection are crucial for understanding trait evolution.

Purpose of the Study:

• To develop a penalized likelihood (PL) framework for high-dimensional phylogenetic comparative data.
• To improve the estimation of evolutionary parameters, including the trait covariance matrix and model parameters.
• To enable efficient model comparison and analysis of complex evolutionary patterns.

Main Methods:

• Development of a penalized likelihood (PL) framework with various penalization strategies.
• Application to Brownian motion, Early-burst, Ornstein-Uhlenbeck, and Pagel's lambda models.
• Utilizing generalized information criterion (GIC) for model comparison.
• Implementation in R packages RPANDA and mvMORPH for broader accessibility.

Main Results:

• PL framework significantly enhances the accuracy of estimating evolutionary trait covariance matrices and model parameters, especially when the number of traits approaches or exceeds the number of species.
• PL models demonstrate efficient comparison using GIC.
• Simulations confirm the improved performance of the PL approach in high-dimensional scenarios.
• Analysis of New World monkey brain shape data supports an Early-burst model for brain morphology diversification.

Conclusions:

• The penalized likelihood (PL) framework provides an efficient and powerful solution for analyzing high-dimensional multivariate phylogenetic comparative data.
• This approach overcomes limitations of traditional methods, enabling robust parameter estimation and model selection.
• The implemented methods facilitate deeper insights into evolutionary processes, trait integration, and ancestral state reconstruction.