Using deep learning to discover functional antibodies for SARS-CoV-2 with picomolar binding affinity

Therapeutic SARS-CoV-2 antibody is desired to target S1 whilst addressing its viral evolution and escape mutations. A deep learning-based approach was developed to create, optimize, and rank mini-protein binders for drug-like properties including binding affinity and developability. The approach was used to discover single domain VHH antibodies that bind to S1 Ancestral, Delta and Omicron simultaneously with picomolar to sub-nanomolar affinity and block S1 interaction with human ACE2 in vitro.