Predictive modeling of immune escape and antigenic grouping of SARS-CoV-2 variants

The ongoing adaptive evolution of Severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2) is characterized by the continued emergence of variants with increased transmissibility and the ability to escape infection- and/or vaccine-induced immunity. This sustained antigenic evolution has necessitated updates to COVID-19 vaccine compositions to better match circulating viral variants. To optimize protection against emerging variants, a reliable means of predicting the immune escape of novel variants is needed to enable at-risk preparation of new vaccine strain compositions. Herein, we describe the development and applications of a quantitative risk calculator that predicts relative immune escape of SARS-CoV-2 variants using a statistical modeling framework. The approach integrates large-scale, experimentally derived spike-antibody epitope and escape maps with serum neutralization data generated using pseudotyped viruses and clinical sera. By aggregating site-level escape information into a strain-level metric, the calculator enables the grouping of antigenically related SARS-CoV-2 variants to guide strain selection for at-risk vaccine design and preparation, in anticipation of seasonal strain change recommendations by global public health agencies and the WHO. Here, we demonstrate the utility of this framework through retrospective and prospective strain selection exercises for the XBB.1.5-, JN.1/KP.2-, and LP.8.1-adapted mRNA-1273 COVID-19 vaccines during the 2023-2026 seasons, respectively. In all cases, model predictions were largely supported by clinical immunogenicity data and aligned with subsequent recommendations by global public health agencies.IMPORTANCEWe present a framework to estimate the relative immune escape potential of emerging variants by integrating previously published experimental epitope-level escape data with serum neutralization measurements. By consolidating mutation-level effects into a strain-level metric, this approach enables classification of antigenically similar variants. Retrospective and prospective applications demonstrate that model-based assessments are consistent with observed immunogenicity data. This framework provides a practical tool to support preparedness efforts by informing at-risk vaccine development activities in advance of seasonal strain selection guidance.