Antigenic characterization of circulating influenza A virus (IAV) strains, via haemagglutination inhibition (HI) assays, is routinely performed by the World Health Organization (WHO) and affiliated laboratories for influenza surveillance and vaccine strain selection. However, global antigenic characterization of IAV is confronted with high costs, animal (ferret) availability, logistical issues with sample sharing and coordination, and other practical challenges. This motivates the development of computational methods for performing antigenic characterization using influenza haemagglutinin (HA) protein sequence data, which is easily measured and widely available. This thesis presents research contributions to address this problem. To facilitate model training and evaluation, we generated...[ Read more ]

Antigenic characterization of circulating influenza A virus (IAV) strains, via haemagglutination inhibition (HI) assays, is routinely performed by the World Health Organization (WHO) and affiliated laboratories for influenza surveillance and vaccine strain selection. However, global antigenic characterization of IAV is confronted with high costs, animal (ferret) availability, logistical issues with sample sharing and coordination, and other practical challenges. This motivates the development of computational methods for performing antigenic characterization using influenza haemagglutinin (HA) protein sequence data, which is easily measured and widely available. This thesis presents research contributions to address this problem. To facilitate model training and evaluation, we generated a comprehensive, up-to-date, genetically matched antigenic dataset for IAV H3N2. This dataset was developed by pooling together and processing HI assay data from periodic WHO surveillance reports, along with HA sequence data from genetic sequence databases. Dataset generation involved multiple steps of curation and filtering to provide genetically matched HI assay data that resolves ambiguities in the raw data and includes relevant metadata. Using this dataset, we developed a robust machine learning-based computational method for accurately predicting the seasonal outcomes of HI assays for IAV H3N2, using only HA sequence data and associated metadata of circulating virus strains. Application of the model reveals the key factors involved with seasonal prediction of HI assays and the dynamics of HA protein sites that are the most important antigenic contributors in different influenza seasons, bringing new insights into the antigenic evolution of IAV H3N2. The results of this thesis have the potential to offer public health benefits by providing computational tools for more rapid and comprehensive global influenza monitoring and potentially informing improved vaccine strain selection.