Improving in silico prediction of influenza vaccine effectiveness by genome analysis incorporating epitope information

Abstract

AbstractFast evaluation of vaccine effectiveness (VE) is valuable for facilitating vaccine development and making vaccination strategy. In previous studies, we developed the computational model linking molecular variations and VE for the influenza and COVID-19, through which VE prediction prior to mass vaccination and infection is possible. In this study, we perform a complete survey of the predictive effect of major functional regions of the influenza virus for VE. Interestingly, we found that the genetic distance measured on the antigenic sites being also the effective mutations for epidemics is a strong predictor for influenza VE. Based on the identified optimal predictor codon set, we develop the improved VE-Genetic Distance model for influenza (VE-GD flu). The prediction accuracy of the new model is R-square 87.1% for H3N2 (p-value < 0.001) on VE data of the United States. Leave-one-out cross validation shows that the concordance correlation coefficient of the predicted and observed VE is 90.6% (95% CI: 73.8-96.9). Significant prediction improvement is also found for pH1N1. Accurate prediction of influenza VE before vaccine deployment may facilitate reverse vaccinology to optimize vaccine antigen design and government preparedness of influenza epidemics.