Empowering precise base editing: Chinese scientists establish an AI framework based on AlphaFold3 contact probability ContactSeek

IT House reported on July 25 that the School of Life Sciences of Peking University published a blog post announcing that its Yichengqi research group jointly published a paper in Nature on July 22, proposing an AI framework ContactSeek based on AlphaFold3 contact probability. The paper is titled "Precise DNA base editing using AlphaFold3-based contact modeling" and was jointly published by the research group of Yi Chengqi from the School of Life Sciences of Peking University, the National Key Laboratory of Gene Function Research and Manipulation, and the Beijing Research Center for RNA, and the research group of Li Dali from the School of Life Sciences of East China Normal University. The research team integrated the contact probability (CP) predicted by AlphaFold3 with the off-target editing information obtained by high-throughput sequencing to develop the AI-driven ContactSeek framework. AlphaFold3 is an artificial intelligence model launched by DeepMind, a subsidiary of Google, which is mainly used to predict the three-dimensional structure of proteins. Updated versions can also analyze protein-nucleic acid interactions, as well as multimolecular complexes. This framework systematically identifies key amino acid residues in Cas proteins and deaminases that determine editing specificity, and engineers base editing tools accordingly, significantly improving editing specificity and providing a new paradigm for the accurate design of gene editing tools. Schematic diagram of the ContactSeek framework The research team pointed out that base editors have become important tools for basic research and disease treatment, but maintaining high editing activity while reducing off-target activity has always been a core issue in the field of gene editing. The team used AlphaFold3 to predict DNA-RNA-protein ternary complexes at target sites and off-target sites, and compared the interaction differences between the two types of complexes. Figure 2. Schematic diagram of AlphaFold3’s prediction process. AlphaFold3 first sorts the input