Scaling Bayesian Models for Large-Scale Infectious Disease Surveillance

• Activity Number: 154
• Type: Topic Contributed
• Date/Time: Monday, August 4, 2014 : 10:30 AM to 12:20 PM
• Sponsor: International Society for Bayesian Analysis (ISBA)
• Author(s): Marc Suchard *+
• Companies: University of California, Los Angeles
• Keywords: Bayesian nonparametrics ; Statistical computing ; Phylogenetics

Abstract

Influenza viruses undergo continual antigenic evolution allowing mutant viruses to evade host immunity. Antigenic phenotype is often assessed through pairwise measurement of cross-reactivity between influenza strains using the hemagglutination inhibition (HI) assay. Large-scale experimentation is generating HI measurements between thousands of strains. Semi-parametric clustering of these measurements guides annual influenza vaccine selection, but current methods fail to account for uncertainty in cluster identification and the unobserved shared evolutionary history between strains. We propose a semi-parametric Bayesian model that uses multidimensional scaling to map pairwise measurements into a latent low-dimensional space. Viral coordinates in this space identify clusters and the viral shared history provides a prior distribution over coordinates through a Browian diffusion process along the history. Central to the success of this model lies a novel sum-product algorithm that transforms an O(N^3 K^2 + K^3) into an O(N K^2) computation for the Brownian diffusion, where N counts the number of viruses and K counts the dimension of the phenotypes in the database. The algorithm exp