Abstract:
|
Each year, influenza virus causes over 250,000 deaths. Vaccination remains the most effective way to limit epidemic spread. To select a vaccine strain, the World Health Organization uses antigenic cartography, a technique based on Multidimensional Scaling, to analyze serological data from thousands of virus isolates. Here, we extended this method to enable systematic clustering of viruses into discrete antigenic phenotypes. We explicitly leverage information from virus phylogeny to relate viral evolution to the emergence of major antigenic phenotypes. Our method extends a recently introduced Bayesian Multidimensional Scaling technique, thereby capturing uncertainty in clustering and cartographic location. In this framework, we used Bayesian Stochastic Search Variable Selection to probabilistically infer branches associated with antigenic mutation. We developed model-specific proposals to address the challenge of mixing when computing posterior estimates using MCMC. We analyzed 43 years of historical H3N2 data and successfully inferred circulating antigenic phenotypes. We believe this method will be useful in vaccine strain selection for antigenically variable pathogens.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.