Modelling molecular evolution of influenza

Poster number: 16

Richard A. Goldstein

  1. National Institute for Medical Research

Sequence evolution is central to both antigenic shift and antigenic drift. Unfortunately the standard models used to understand sequence evolution rely on assumptions and approximations that are extremely questionable for the modelling of pathogens evolving to evade an immune response or to infect a new species of host. In particular, the magnitude and nature of the selective pressure acting on different locations of viral proteins (such as haemagglutinin) can be quite different in ways that may be both changing and difficult to predict. We have developed a series of "site class" models which assume that various locations in the protein evolve according to different models, where neither the models nor their assignment to different locations is known a priori. Rather, the models and their assignments to different sites are determined through an iterative Expectation Maximisation (EM) procedure. In contrast to standard models of rate variation where a single substitution model is multiplied by a location-specific scaling factor (generally modelled with a Gamma distribution), we allow both the absolute and relative substitution rates to vary. This approach allows us to a) characterise the selective pressures acting on these proteins, b) determine which locations are under which forms of selective pressure, and c) understand where there are changes in the selective pressure at different locations, or where the selective pressure depends upon the subtype or host. Applications of this model to influenza haemagglutinin and neuraminidase are presented.