Online Bayesian Phylodynamic Inference in BEAST with Application to Epidemic Reconstruction

Area of Science:

• Computational Biology
• Epidemiology
• Phylogenetics

Background:

• Traditional Bayesian phylogenetic inference packages require complete recalculation for new genetic data, hindering real-time outbreak analysis.
• Sequential data arrival during epidemics necessitates "online" inference methods for timely pathogen dynamics reconstruction.

Purpose of the Study:

• To develop an efficient methodology for updating Bayesian phylogenetic posterior distributions with newly acquired genetic data.
• To enable rapid, online inference of pathogen evolutionary dynamics during outbreaks.

Main Methods:

• Introduced a novel distance-based method to insert new taxa into existing phylogenies.
• Developed imputation strategies for new model parameters to handle increasing data dimensionality.
• Implemented the procedure in the BEAST 1.10 software package, leveraging informed starting values and optimized transition kernels.

Main Results:

• Demonstrated a considerable reduction in time to obtain posterior estimates for evolving datasets.
• Successfully applied the framework to genetic data from the West African Ebola virus epidemic, showing improved efficiency.
• Validated the utility of online inference for real-time epidemic monitoring and phylogenetics.

Conclusions:

• The proposed online Bayesian phylogenetic framework significantly accelerates inference for sequentially arriving genetic data.
• This methodology is crucial for timely outbreak reconstruction and has broad applicability in phylogenetics beyond epidemic surveillance.