Comparative Analysis of Principal Components Can be Misleading

Area of Science:

• Evolutionary Biology
• Phylogenetics
• Quantitative Genetics

Background:

• Most trait evolution models are univariate, yet multivariate analyses are crucial for understanding complex evolutionary patterns.
• Principal Component Analysis (PCA) is frequently used to reduce multivariate data dimensionality for univariate phylogenetic modeling.
• Standard PCA on phylogenetically structured data is known to cause inferential issues but remains widely applied.

Purpose of the Study:

• To demonstrate how standard PCA can lead to erroneous conclusions in phylogenetic comparative studies.
• To evaluate the effectiveness of phylogenetic PCA (pPCA) in mitigating PCA-induced artifacts.
• To highlight the need for and discuss alternatives to univariate analyses of multivariate trait data in phylogenetics.

Main Methods:

• Simulated trait evolution under multivariate Brownian motion.
• Application and analysis of standard PCA and phylogenetic PCA (pPCA) on simulated datasets.
• Comparison of inferred evolutionary patterns (e.g., early burst) from univariate models fit to principal components.

Main Results:

• Standard PCA can incorrectly suggest an 'early burst' evolutionary process for traits evolving under constant-rate multivariate Brownian motion.
• pPCA also produces similar artifacts when the assumed model for axis calculation deviates from the true evolutionary model.
• High effective dimensionality of datasets exacerbates these inferential errors, stemming from analyzing a biased subset of the multivariate data.

Conclusions:

• Current PCA-based approaches for phylogenetic comparative analysis can yield misleading results, particularly with complex trait evolution.
• The reliance on a few principal components represents a biased sampling of the underlying multivariate evolutionary pattern.
• Development and application of truly multivariate phylogenetic comparative methods are essential for accurate evolutionary inference.