Huot, M.; Rosenbaum, P.; Planchais, C.; Mouquet, H.; Monasson, R.; Cocco, S.

The evolutionary trajectory of SARS-CoV-2 variants is shaped by the selective pressures exerted by host immunity, in particular neutralizing antibodies targeting the receptor-binding domain (RBD). Here, we introduce a data-driven model that quantifies the impact of antibody pressure on RBD evolution and assesses antibody resilience beyond single mutations and known variants. We integrate deep mutational scanning of ACE2 and 31 antibodies with a generative model trained on pre-pandemic Coronaviridae sequences. We then use our generative model to design viable RBD variants under immune pressure from four monoclonal antibodies: SA55, S2E12, S309, and VIR-7229. Experimental validation of 22 variants, with up to 21 mutations from Wuhan wild-type, confirms 50% expression rate. Binding assays further reveal that S309 and VIR-7229 maintain binding across diverse mutational combinations, while SA55 is escaped by one variant and S2E12 exhibits lower resilience. In addition, our model captures correlated escape across antibody pairs (R = 0.65), guiding the selection of negatively correlated cocktails to reduce shared escape routes. By quantitatively linking viral adaptation to antibody resistance profiles, this framework provides a predictive foundation for optimizing therapeutic strategies and enhancing long-term pandemic preparedness.