https://www.cell.com/cell/abstract/S0092-8674(25)00680-4

https://www.cas.cn/syky/202507/t20250708_5075564.shtml

A team from the CAS Institute of Genetics and Developmental Biology has proposed a universal protein engineering method based on artificial intelligence.

Their method AiCE (AI-informed Constraints for protein Engineering) is based on a universal inverse folding model that integrates structural and evolutionary constraints and can achieve efficient protein evolution simulation and functional design without training a dedicated artificial intelligence model.

The team achieved AiCE functional verification of 8 structurally and functionally diverse proteins including deaminases, nuclear localization sequences, nucleases and reverse transcriptases at the wet experimental level, demonstrating its simplicity, efficiency and versatility. With the help of optimized deaminases, the team developed new base editors that can be used for precision medicine and molecular breeding, including a new cytosine base editor enABE8e with a nearly halved editing window, a new adenine base editor enSdd6-CBE with a 1.3-fold increase in fidelity, and a new mitochondrial base editor enDdd1-DdCBE with a 13-fold increase in activity.