Bayesian Model Averaging of Parametric Coalescent Models for Phylodynamic Inference.

Area of Science:

• Population genetics
• Computational biology
• Epidemiology

Background:

• Accurate reconstruction of population history from genetic data is crucial for understanding evolutionary dynamics.
• Bayesian phylodynamic models rely heavily on the choice of appropriate demographic models, introducing uncertainty.
• Existing methods often require pre-specifying a single demographic model, potentially limiting inference.

Purpose of the Study:

• To develop a Bayesian model averaging (BMA) framework to integrate multiple parametric coalescent models for demographic history inference.
• To address uncertainty in model selection for phylodynamic analyses.
• To provide a unified approach for inferring population histories without restrictive model pre-selection.

Main Methods:

• Introduction of a Bayesian model averaging (BMA) framework integrating constant, exponential, logistic, and Gompertz growth models with expansion variants.
• Implementation using Metropolis-coupled Markov chain Monte Carlo (MCMC) for seamless switching among candidate growth functions.
• Validation through simulation studies and application to real-world datasets (Hepatitis C virus and colorectal cancer).

Main Results:

• The BMA framework successfully captures demographic histories by integrating multiple growth models.
• Simulation studies confirmed the well-calibrated joint inference of genealogy and population parameters.
• Analysis of Hepatitis C virus data supported founder population models with rapid Gompertz-like expansions.
• Colorectal cancer data suggested plausible exponential-like growth in tumor subclones, even in advanced stages.

Conclusions:

• The unified BMA framework reduces the need for restrictive model selection, enhancing the inference of population histories.
• This approach offers deeper biological insights into epidemic spread and tumor evolution.
• The method provides a powerful, robust tool for inferring population dynamics across diverse biological domains by avoiding overfitting.