Current research focuses on designing multi-epitope vaccines via in silico methods to address the limitations of traditional approaches constructed on real pathogens. Yang et al. developed the AI-based method (DeepVacPred) that integrates a computational approach with deep neural networks to improve the speed, increase efficiency, and narrow the number of initial candidates compared to standard in silico methods. In the study, the DeepVacPred framework identified 26 potential vaccine subunits of SARS-CoV-2 spike protein out of 132 candidates. Then, the epitopes of B-cells, cytotoxic T lymphocytes (CTL), helper T lymphocytes (HTL), which are responsible for inducing immune response, were predicted via individual online databases and identified 14 key vaccine subunit candidates. Additional analysis in a population coverage database verified a wide range of coverage for all 14 subunits. The final multi-epitope vaccine was designed with 11 subunits and an adjuvant. The antigenicity, allergenicity, and other properties were further analyzed by different servers to validate the finalized vaccine design. Moreover, the additional study on the three most frequent RNA mutation occurring area from the spike protein resulted in no effect on the final vaccine design, indicating the effectiveness of the vaccine to the known mutations of SARS-CoV-2.