Model design for nonparametric phylodynamic inference and applications to pathogen surveillance.

Area of Science:

• Genomics
• Epidemiology
• Statistical modeling

Background:

• Effective population size inference from genomic data reveals demographic and epidemiological dynamics.
• Phylodynamic inference combines nonparametric population models with molecular clock models for time-stamped genetic data.
• Existing methods for nonparametric effective population size inference are primarily Bayesian.

Purpose of the Study:

• To develop a novel frequentist approach for inferring effective population size using nonparametric latent process models.
• To optimize model parameters based on out-of-sample prediction accuracy.
• To implement this methodology in a user-friendly R package (mlesky).

Main Methods:

• Developed a frequentist statistical framework for population size dynamics.
• Utilized nonparametric latent process models to capture population size fluctuations over time.
• Employed out-of-sample prediction accuracy for parameter optimization.
• Applied the method to HIV-1 and SARS-CoV-2 genomic sequence data.

Main Results:

• Demonstrated the flexibility and computational speed of the new frequentist approach through simulations.
• Successfully applied the methodology to infer HIV-1 population dynamics in the USA.
• Estimated the impact of non-pharmaceutical interventions (e.g., UK lockdown) on COVID-19 epidemic reproduction number using SARS-CoV-2 data.

Conclusions:

• The new frequentist method provides a robust and efficient alternative for phylodynamic inference.
• The mlesky R package facilitates the application of advanced statistical techniques to genomic and epidemiological data.
• This approach enhances our ability to understand pathogen evolution and the impact of public health interventions.