Statistical modeling for adaptive trait evolution in randomly evolving environment

TL;DR

This paper extends models of adaptive trait evolution by incorporating the Cox-Ingersoll-Ross process for the rate of evolution, providing a new framework for phylogenetic analysis with intractable likelihoods.

Contribution

It introduces a novel model where the rate of trait evolution follows the CIR process and employs ABC for inference, advancing adaptive evolution modeling.

Findings

Simulation results validate the model's effectiveness.

Empirical analysis demonstrates practical applicability.

The framework accommodates multiple predictors and interactions.

Abstract

In past decades, Gaussian processes has been widely applied in studying trait evolution using phylogenetic comparative analysis. In particular, two members of Gaussian processes: Brownian motion and Ornstein-Uhlenbeck process, have been frequently used to describe continuous trait evolution. Under the assumption of adaptive evolution, several models have been created around Ornstein-Uhlenbeck process where the optimum $\theta^y_t$ of a single trait $y_t$ is influenced with predictor $x_t$. Since in general the dynamics of rate of evolution $\tau^y_t$ of trait could adopt a pertinent process, in this work we extend models of adaptive evolution by considering the rate of evolution $\tau_t^y$ following the Cox-Ingersoll-Ross (CIR) process. We provide a heuristic Monte Carlo simulation scheme to simulate trait along the phylogeny as a structure of dependence among species. We add a framework to incorporate multiple regression with interaction between optimum of the trait and its potential predictors. Since the likelihood function for our models are intractable, we propose the use of Approximate Bayesian Computation (ABC) for parameter estimation and inference. Simulation as well as empirical study using the proposed models are also performed and carried out to validate our models and for practical applications.