Rapid protein evolution by few-shot learning with a protein language model

Researchers have developed a new method called EVOLVEpro that combines protein language models (PLMs) and protein activity predictors to accelerate protein evolution. This few-shot active learning framework can enhance protein activity with just four rounds of evolution, outperforming current methods and achieving up to 100-fold improvement in desired properties. The study showcases EVOLVEpro's effectiveness in evolving proteins for various applications, including RNA production, CRISPR technology, genome editing, and antibody binding. The approach demonstrates the advantages of using small amounts of experimental data over zero-shot predictions, paving the way for AI-guided protein engineering in biology and medicine. The authors have disclosed competing interests and filed patents related to this work.