Base editors, fusion proteins comprising a catalytically impaired Cas9 nuclease and a nucleobase deaminase, frequently cause unwanted Cas9-dependent off-target mutations due to mismatch tolerance between gRNA and the target sequence. This study generated large datasets of off-target editing efficiencies for adenine base editors (ABEs) and cytosine base editors (CBEs) by stably integrating gRNA-off-target pairs into human cells. Deep learning models, ABEdeepoff and CBEdeepoff, were trained using these datasets (54,663 and 55,727 off-targets for ABEs and CBEs, respectively) to predict off-target sites. These models demonstrated strong predictive capabilities (Spearman correlation 0.710–0.859 for endogenous loci) and are accessible via an online web server (http://www.deephf.com/#/bedeep/bedeepoff) to facilitate minimizing off-target effects in base editing.

Chengdong Zhang, Yuan Yang, Tao Qi, Yuening Zhang, Linghui Hou, Jingjing Wei, Jingcheng Yang, Leming Shi, Sang-Ging Ong, Hongyan Wang, Hui Wang, Bo Yu, Yongming Wang