Abstract

Influenza viruses contain 8 RNA segments coding for 10-11 proteins, including the two surface proteins Hemagluttinin (HA) and Neuraminidase (NA) which determine the viral subtype. In addition to mutations in the surface proteins (e.g. to enable binding to a mammalian receptor), mutations in the segments coding for internal proteins are also important in virulence and adapting the virus from avian to mammalian hosts.

In this talk I will describe how Bayesian Graphical Models (BGMs) can be used to detect site-site and site-phenotype associations in Influenza A viruses. BGMs represent the direct conditional dependencies between variables (mutations at amino acid sites or virulence phenotype), rather than just correlations, as edges in a network. Here BGMs were inferred using a Monte Carlo Markov Chain technique to sample from appropriate network structures, and cross validation, permutation tests and parametric bootstrapping was used to improve model robustness.

Firstly, I will describe the detection of the set of mutations directly associated with virulence in mammals in highly pathogenic avian influenza H5N1 (HPAI H5N1 is highly pathogenic in chickens, but not necessarily in mammals). Secondly an investigation into adaptations of influenza viruses from avian to mammalian hosts accounting for the role of founder effects, shared ancestry and epistatic interactions will be described. Here the variables are host change (avian from/to human or swine) and mutational histories of the amino acid sites (inferred using a phylogenetic tree and codon model). As well as changes in the Hemagglutinin receptor binding site for H1N1 and H5N1 subtypes, there were several sites directly associated with host change in the polymerases and NS1 . Networks of interactions between HA antigenic sites and the receptor binding site, and within the polymerase complex were also identified.