Computational predicting the human infectivity of H7N9 influenza viruses isolated from avian hosts

Abstract

The genome composition of a given avian influenza virus is the primary determinant of its potential for cross-species transmission from birds to humans. Here, we introduce a viral genome-based computational tool that can be used to evaluate the human infectivity of avian isolates of influenza A H7N9 viruses, which can enable prediction of the potential risk of these isolates infecting humans. This tool, which is based on a novel class weight-biased logistic regression (CWBLR) algorithm, uses the sequences of the eight genome segments of an H7N9 strain as the input and gives the probability of this strain infecting humans (reflecting its human infectivity). We examined the replication efficiency and the pathogenicity of several H7N9 avian isolates that were predicted to have very low or high human infectivity by the CWBLR model in cell culture and in mice, and found that the strains with high predicted human infectivity replicated more efficiently in mammalian cells and were more infective in mice than those that were predicted to have low human infectivity. These results demonstrate that our CWBLR model can serve as a powerful tool for predicting the human infectivity and cross-species transmission risks of H7N9 avian strains.