Efficient evolution of human antibodies from general protein language models
Problem
Efficiently improving human antibodies by finding rare beneficial mutations in a vast sequence landscape, normally requiring costly screening.
Why it matters
Improved antibodies can yield better therapeutics against diseases such as Ebola and SARS-CoV-2.
What the AI did
General protein language models suggested evolutionarily plausible mutations, guiding affinity maturation with no antigen, specificity or structure information given.
Results
Screening 20 or fewer variants per antibody over two rounds improved binding affinity up to sevenfold (four mature antibodies) and up to 160-fold (three unmatured); designs showed favorable thermostability and viral neutralization.
Impact
Achieved large affinity gains with minimal laboratory screening across seven antibodies.
Record
| Venue | Nature Biotechnology |
|---|---|
| Published | 2023 |
| Work type | article |
| Authors | Brian Hie, Varun R. Shanker, Duo Xu, Theodora U. J. Bruun, Payton A. B. Weidenbacher, Shaogeng Tang, Wesley C. H. Wu, John E. Pak et al. |
| Citations | 449 (OpenAlex, at import time) |
| References | 107 |
| Model or method | general protein language models (not named in source) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41587-023-01763-2 |
| Open access | hybrid |
| Keywords | protein language models, affinity maturation, antibody engineering, neutralization, SARS-CoV-2, Ebola |
Importance rating
significant — thousands to millions affected
Improved clinically relevant antibodies neutralizing Ebola and SARS-CoV-2; results generalize to antibiotic resistance and enzyme activity, a broad protein-engineering capability.