The Effects of Sampling Location and Predictor Point Estimate Certainty on Posterior Support in Bayesian...

Area of Science:

• * Computational Biology
• * Epidemiology
• * Genetics

Background:

• * Generalized linear models in Bayesian phylogeography allow simultaneous reconstruction of viral spatiotemporal history and quantification of predictor variable contributions.
• * The impact of discrete state partitioning on these analyses remains largely unexplored.

Purpose of the Study:

• * To investigate the sensitivity of Bayesian phylogeography to the choice of discrete state partitions.
• * To determine the optimal level of spatial aggregation for predictor variables in viral diffusion studies.

Main Methods:

• * Analysis of 299 West Nile virus envelope gene sequences from the United States.
• * Inclusion of fifteen predictor variables aggregated at four spatial levels.
• * Application of Bayesian phylogeography with generalized linear models.

Main Results:

• * Viral phylogeny topology remained consistent across different spatial aggregation levels.
• * Predictor variable support varied significantly based on the level of spatial aggregation.
• * Variance in predictor support metrics was minimized at the most precise spatial level and maximized at sparser levels.

Conclusions:

• * The partitioning of regions into discrete locations significantly impacts Bayesian phylogeography results.
• * Utilizing the most precise discrete states is recommended to minimize posterior variance and obtain interpretable estimates.
• * Precise spatial partitioning aids in accurately identifying key drivers of viral diffusion, as demonstrated with West Nile virus.