PREDICTING AVIAN-TO-HUMAN TRANSMISSION OF INFLUENZA A VIRUSES USING MACHINE LEARNING APPROACHES

Abstract

Zoonotic influenza A viruses occasionally emerge from the avian population to cause human infections, such as the H5N1 and H7N9 outbreaks. Current influenza surveillance focuses on genomic markers, yet cannot reliably predict zoonotic strains prior to an outbreak. The main objective is therefore to develop tools using machine learning approaches to detect avian-to-human transmission of influenza viruses from protein sequences. A host tropism protein prediction system was first constructed to predict avian or human tropism of 11 influenza proteins, with the results combined into a host tropism protein signature. This led to the discovery of distinct zoonotic signatures with mixed avian and human protein tropisms, motivating the construction of a zoonotic strain prediction model to classify avian, human or zoonotic strains with high accuracy. This may prove useful in improving influenza surveillance to identify high risk zoonotic strains prior to an outbreak, providing an early warning to an impending outbreak.