 This study designed deep convolutional neural networks, CNNs, to predict influenza antigenicity, which outperforms other predictive models with a blind validation accuracy of 95.8%. The model was optimized using particle swarm optimization and applied to vaccine recommendations in the period 1997 to 2011. The results show that our model outperforms the WHO recommendation and other existing models and could potentially improve the vaccine recommendation process. The modeling framework is flexible and can be adopted to study other type of viruses. This article was offered by Eva K. Lee, Hashim Chin, and Helda I. Nakaya.