Broadly neutralizing antibodies (bnAbs) are essential for the development of vaccines and therapeutics against rapidly evolving pathogens like HIV and SARS-CoV-2, yet traditional discovery methods remain technically challenging and time consuming. Here, we introduce ClonoDeep, an AI-powered platform that integrates public antibody clonotypes with a sequence-based deep learning model to directly identify bnAbs from a large-scale immune repertoire, independent of antigen-specific immunization. Applied to SARS-CoV-2 repertoires, ClonoDeep identified 18 clonotype-derived antibody candidates; 83% of the candidates were neutralizing antibodies, and 8 of these antibodies demonstrated broad neutralization across variants. Structural analysis revealed that somatic hypermutations at HCDR3 His107/Gly109 are key enhancers of the binding affinity and neutralizing breadth. Extending to HIV, ClonoDeep uncovered three previously unreported bnAbs from non-HIV cohorts, indicating that rare bnAb-like precursors exist in non-HIV cohort repertoires. ClonoDeep establishes a high-throughput computational approach for mining neutralizing antibodies from antibody repertoires shaped by non-pathogen-specific immunity and provides design principles to guide vaccine strategies against genetically diverse pathogens.

Link:https://doi.org/10.1016/j.celrep.2026.117582