Motivation Influenza is an infectious disease associated with excess human deaths. For influenza A virus (IAV), the surface antigens hemagglutinin (HA) and neuraminidase (NA) define the specific subtype. The high mutation rate of IAV can make clinical testing and assigning subtype after sequencing challenging. Accurate subtype classification is important for tracking circulating IAV strains that may impact human health.

Results We analyzed a large IAV protein sequence dataset with known HA and NA subtypes. Using logistic regression and random forest approaches, we identified a small set of subtype-associated amino acids, which were then used to develop a subtype characterization method with near-optimal accuracy. Further analysis indicated that signal sequence peptide variation and indels in HA and NA explained the unique combination of subtype-associated amino acids.