Bioinformatics models for predicting antigenic variants of influenza A/H3N2 virus

Motivation: Continual and accumulated mutations in hemagglutinin (HA) protein of influenza A virus generate novel antigenic strains that cause annual epidemics. Results: We propose a model by incorporating scoring and regression methods to predict antigenic variants. Based on collected sequences of influenza A/H3N2 viruses isolated between 1971 and 2002, our model can be used to accurately predict the antigenic variants in 19992004 (agreement rate 91.67). Twenty amino acid positions identified in our model contribute significantly to antigenic difference and are potential immunodominant positions.