The rapid evolution of influenza Viruses is a significant challenge to vaccine production. To predict influenza antigenicity, the study created deep convolutional neural networks (CNNs). The configuration of the CNNs was optimised using particle swarm optimization. With a blind validation accuracy of 95.8 percent, optimal neural networks outperform other predictive models. This model exceeds the WHO and other current models, with the findings showing that it could theoretically boost the vaccine recommendation process. Findings show that WHO often chooses viral strains with small annual variations, and slowly learns and recovers once the coverage falls very short.

The influenza strains selected by the CNN model, however, will vary considerably each year and often have good coverage. In brief, the researchers built an extensive computer pipeline to optimise CNN for influenza A modelling and recommendations on vaccines. Compared with conventional hemagglutination inhibition research, this is more cost-effective and time-effective. The modelling system for the study of other types of viruses is versatile and can be adopted.

Reference: https://www.tandfonline.com/doi/full/10.1080/21645515.2020.1734397