Stochastic Phylogenetic Models of Shape
TL;DR
A new framework modeling morphological shape evolution as a diffusion process using Metropolis-Hastings MCMC improves ancestral shape estimation at the root and internal nodes of a phylogeny.
Problem / question
Modeling the evolution of 2D and 3D morphological shapes is statistically difficult because landmark data are inherently non-linear and spatially correlated, causing naive phylogenetic methods to fail.
Methods
The authors developed a framework that models shape evolution as a diffusion process parameterized by evolutionary rate and spatial autocorrelation among landmarks. Operating on a fixed phylogenetic tree, the method uses a Metropolis-Hastings Markov Chain Monte Carlo (MCMC) sampling scheme to infer ancestral shapes. It was evaluated using simulations and applied to a real-world dataset of butterfly wing images.
Key findings
Simulation studies demonstrated that the new MCMC-based diffusion model produces improved estimates of ancestral shapes at the root of the tree and yields well-calibrated credible sets for shapes at internal nodes. Additionally, the model's diffusion parameter for spatial autocorrelation was shown to quantify morphological integration in a manner similar to existing integration metrics.
Why it matters
This approach allows evolutionary biologists to accurately reconstruct ancestral morphologies and map stochastic characters while respecting the biological reality that nearby physical landmarks evolve in a correlated manner.
Limitations
The proposed framework assumes that the underlying phylogenetic tree is fixed and does not account for uncertainty in the tree topology itself.
Takeaway
Modeling morphological evolution as a diffusion process with MCMC sampling successfully captures the spatial correlation of landmarks, enabling more accurate reconstructions of ancestral shapes like butterfly wings.