Antibody Evolution
Point mutations to improve binding affinity
Protein language model which suggests point mutations to guide antibody evolution and improve binding affinity without requiring information about target antigens or protein structure. Uses ESM-1b and ESM-1v trained on UniRef50 and UniRef90.
Experimental validation
Experimentally validated to improve binding affinities of both mature and unmatured antibodies through minimal laboratory screening (20 or fewer variants across two rounds). The approach achieved up to sevenfold improvements for mature antibodies and up to 160-fold improvements for unmatured antibodies, while maintaining favorable thermostability and viral neutralization activity. Notably, the same models proved effective across diverse protein families and selection pressures, suggesting broad applicability beyond antibody optimization.