Most influenza surveillance utilizes antibody responses from ferrets (outnumbering the amount of human data by 10-fold). While it is known that ferret responses can differ from human responses, it is not clear when or how their responses will differ. Using ferret data, we will predict the value±error for human experiments, which will help refine influenza vaccine selection (that is currently determined through ferret studies).

The influenza vaccine changes every 1-2 years. While we can separately model each vaccine, we would like to train a model on the combined data from all prior studies. Our current approach describes each virus according to its interactions with antibodies. By adding sequence information, we will develop a single unified model that can predict all past vaccines and explore the space of potential future vaccines.