Horseshoe-based Bayesian nonparametric estimation of effective population size trajectories

Area of Science:

• Population genetics
• Phylodynamics
• Computational biology

Background:

• Phylodynamics infers past population dynamics from genetic data.
• Current Bayesian nonparametric methods include change-point models and Gaussian process priors.
• Both methods have limitations: computational issues with change-point models and lack of local adaptivity in Gaussian processes.

Purpose of the Study:

• To develop a novel, locally adaptive Bayesian nonparametric method for phylodynamic inference.
• To overcome limitations of existing methods in accurately recovering population size trajectories.
• To provide a flexible approach accommodating diverse functional behaviors.

Main Methods:

• Proposed a novel method modeling log-transformed effective population size using a horseshoe Markov random field.
• The horseshoe Markov random field blends properties of change-point and Gaussian process models.
• Assessed model performance using simulated data and real-world case studies.

Main Results:

• The proposed method demonstrated reduced bias and increased precision compared to contemporary phylodynamic methods.
• Successfully reconstructed past genetic diversity changes in Hepatitis C virus in Egypt.
• Accurately estimated population size changes for ancient and modern steppe bison.

Conclusions:

• The novel horseshoe Markov random field approach offers enhanced local adaptivity for phylodynamic inference.
• This method captures complex population size trajectory features missed by state-of-the-art techniques.
• The approach provides a more accurate and flexible tool for analyzing population genetic data.