Abstract
The COVID-19 pandemic has profoundly affected global health, driven by the remarkable transmissibility and mutational adaptability of the SARS-CoV-2 virus. Although five variants of concern, Alpha, Beta, Gamma, Delta, and Omicron, have been identified, the classification task in this study is formulated using four classes: Alpha, Delta, Omicron, and Else, reflecting the sequence availability and temporal coverage of the dataset. Here, we develop an integrative framework that combines direct coupling analysis (DCA), Circos-based visualization, and convolutional neural networks (CNNs) to characterize lineage-specific epistatic signatures from large-scale SARS-CoV-2 genomic sequences. DCA-inferred pairwise mutational couplings were transformed into Circos images, which were then used as inputs for CNN-based classification models. The proposed framework achieved robust variant classification, with the best-performing model reaching a weighted-average $$F_1\text {-score}$$ of $$98.68\pm 0.75\%$$ and an AUC close to 1. Additional temporal holdout analyses showed that the framework retained reasonable predictive capability across evolutionary time, yielding a weighted-average $$F_1\text {-score}$$ of 87.85%.