Probabilistic principal component analysis for phylogenetic comparative studies.

Area of Science:

• Evolutionary Biology
• Bioinformatics
• Statistical Genetics

Background:

• Principal Component Analysis (PCA) is a standard technique for multivariate data analysis in biology.
• Phylogenetic PCA extends PCA to account for evolutionary relationships, but parameter estimation can be improved.
• Existing methods often ignore the error introduced by dimensionality reduction.

Purpose of the Study:

• To develop a probabilistic framework for estimating PCA parameters, enhancing accuracy.
• To introduce a statistically rigorous method for selecting the number of principal components.
• To improve the analysis of trait evolution and multivariate patterns in biological datasets.

Main Methods:

• Implemented explicit probability modeling for PCA parameter estimation.
• Integrated multiple models of trait evolution (Brownian motion, Ornstein-Uhlenbeck, Early Burst, Pagel's λ).
• Utilized Akaike Information Criterion (AIC) for model selection and introduced a probabilistic component selection approach.

Main Results:

• Probabilistic PCA successfully incorporates dimensionality reduction error, unlike standard PCA.
• The method demonstrated superior performance in simulations and an empirical dataset with 35 traits.
• The R package "do3PCA" provides a user-friendly implementation of the novel approach.

Conclusions:

• Probabilistic PCA offers a more robust and accurate approach to multivariate data analysis in biology.
• The framework enhances the study of trait evolution and pattern discovery by accounting for statistical uncertainty.
• The "do3PCA" package facilitates the application of these advanced methods in evolutionary and biological research.