 MLAEP is a machine learning-based method for predicting the evolution of SARS-CoV-2 virus strains. It uses structure models, multitask learning, and genetic algorithms to identify antigenic changes and correlate them with sampling times. This allows us to infer the order of mutation events along the virus's evolutionary trajectory. We also used MLAEP to analyze existing SARS-CoV-2 variants and found new mutations in immunocompromised patients and emerging variants such as XBB1.5. Furthermore, MLAEP was able to predict the effectiveness of these variants in evading immunity, allowing us to better understand how the virus evolves and potentially develop more effective vaccines. This article was authored by Wen Kai-han, Ningning Chen, Shinjishu, and others.