Efficient and precise base editors (BEs) for C-to-G transversion are highly desirable. This study reports engineered C-to-G BEs with high efficiency and fidelity, and sequence context predictability via machine learning. Optimized C-to-G BEs (OPTI-CGBEs) were created by altering the species origin and relative position of uracil-DNA glycosylase and deaminase, along with codon optimization. A deep-learning model accurately predicted OPTI-CGBE editing outcomes based on sequence context, using data from a sgRNA library. Efficient base editing in mouse embryos was demonstrated, generating Tyr-edited offspring. These engineered CGBEs offer efficient and precise base editing with predictable outcomes based on target site sequence context.

Tanglong Yuan, Nana Yan, Tianyi Fei, Jitan Zheng, Juan Meng, Nana Li, Jing Liu, Haihang Zhang, Long Xie, Wenqin Ying, Di Li, Lei Shi, Yongsen Sun, Yongyao Li, Yixue Li, Yidi Sun, Erwei Zuo