Deep mutational learning for the selection of therapeutic antibodies resistant to the evolution of Omicron variants of SARS-CoV-2.

Summary

Using a deep mutational library of Omicron BA.1 RBD and ensemble deep-learning models, the authors predicted antibody binding/escape and identified complementary two-antibody combinations resilient to viral evolution. This strategy enables prospective selection of therapeutic antibodies robust to emerging SARS-CoV-2 variants.

Key Findings

• Constructed a high-mutational-distance Omicron BA.1 full-length RBD library and screened for ACE2 and antibody binding.
• Trained ensemble deep-learning models to predict binding/escape for eight therapeutic antibody candidates across diverse epitopes.
• In silico evolution across millions of sequences identified two-antibody combinations with complementary, enhanced resistance to viral evolution.

Clinical Implications

Supports development of durable antibody cocktails for prophylaxis/treatment, informing stockpiles and rapid response to new variants. It guides epitope complementarity to minimize escape.

Limitations

• Computational predictions require continuous experimental validation against emerging variants
• Focus on RBD-targeting antibodies; non-RBD epitopes were not explored

Future Directions

Prospective clinical translation of AI-selected antibody cocktails; expand to non-RBD epitopes and polyclonal mixtures; open benchmarks for model generalization across viral families.