Predicting genetic evolution of viruses to identify suitable vaccines using artificial intelligence

Abstract

Abstract The evolution of the viruses is rapidly becoming a global challenge to the creation of vaccines since the new variants are often capable of escaping the immune system and decreasing the vaccine efficacy. The traditional methods of genomic epidemiology rely on the retrospective phylogenetic analysis, which can elucidate the previous mutations, but cannot predict the evolutionary trends in the future. In order to address these disadvantages, a new Refined Deep Evolutionary Learning Framework (R-DELF) is proposed that combines the genomic, structural, and temporal intelligence in predicting proactive viral mutations and assessing vaccine suitability. The methodology uses an ESM-2 Transformer that extracts structure-aware embeddings, merged with dual-attention Graph Neural Networks (GNNs) which learn phylogenetic and structural dependencies. Evolutionary learning maximiser improves adaptation modelling and an Explainable AI layer, which offers interpretability based on residue-level attribution. Tests indicate that experimentally it achieves 99.2% accuracy, 97.92% precision, 98.89% recall and 99.4% F1, which is higher than the current AI-based virology models. It is implemented in Python and with the help of TensorFlow and genomic and protein data obtained via Kaggle. The framework allows predicting the high-risk mutations in advance, facilitates the production of vaccines on time, and increases the preparedness to pandemics by making intelligent, data-driven predictions of viral evolution.