Professor Sangsu Bae

This research introduces a novel method for enhancing prime editing by using artificial intelligence to design specialized proteins. Scientists utilized RFdiffusion and AlphaFold 3 to create a compact binder that effectively inhibits mismatch repair, a cellular process that typically hinders editing success. Because of its minimal size, this binder can be easily integrated directly into the prime editing architecture, streamlining the delivery process. Experimental results demonstrate that this AI-generated protein significantly boosts the precision and efficiency of genomic modifications in both human cells and animal models. Ultimately, the study highlights how generative protein design can overcome biological barriers to advance the future of gene therapy.

SNU Medicine Researchers Harness AI to Revolutionize Prime Editing Efficiency

In a groundbreaking fusion of artificial intelligence and genome editing, a research team from Seoul National University College of Medicine has developed a novel approach to significantly boost the efficiency of prime editing, bringing highly precise gene therapies one step closer to clinical reality. The study, led by Professor Sangsu Bae alongside researchers Ju-Chan Park and Heesoo Uhm from our Genomic Medicine Institute and Department of Biomedical Sciences, was recently published in the prestigious journal Cell.

Prime editing (PE) is widely celebrated as the "word processor" of genome editing, allowing for precise search-and-replace DNA edits without causing harmful double-strand breaks. However, the cell's natural mismatch repair (MMR) pathway has historically been a major roadblock, as it frequently recognizes and removes these desired engineered edits, severely limiting the system's efficiency. Previously, researchers attempted to suppress this repair mechanism by co-delivering a dominant-negative MLH1 protein (MLH1dn). Unfortunately, MLH1dn is a massive protein comprising 753 amino acids, making it exceedingly difficult to package into the limited payload space of viral delivery vehicles used in gene therapy, such as adeno-associated viruses (AAVs).

To overcome this critical size limitation, the SNU Medicine team turned to cutting-edge generative AI technologies, specifically RFdiffusion and AlphaFold 3. The team generated thousands of potential protein structures and utilized a highly innovative AlphaFold 3 competition test to identify which candidates could most effectively disrupt natural cellular interactions.

The result was the creation of a de novo MLH1 small binder (MLH1-SB) that is incredibly compact—just 82 amino acids long. This AI-generated micro-protein precisely targets the dimeric interface of the MLH1 and PMS2 proteins, successfully preventing the formation of the MutLα complex and effectively hitting the "pause button" on the MMR pathway.

The experimental results are striking. Because of its tiny footprint, the MLH1-SB can be seamlessly integrated directly into existing prime editor architectures using a 2A self-cleaving peptide system. When incorporated into the advanced PE7 system (creating the "PE7-SB2" platform), the researchers observed an astonishing 18.8-fold increase in prime editing efficiency compared to the standard PEmax system in human HeLa cells. Furthermore, this new architecture achieved a 3.4-fold increase over the baseline PE7 system in living mice, demonstrating its robust in vivo capabilities with limited cellular toxicity and no significant unwanted transcriptome alterations.

This remarkable achievement not only solves a major bottleneck in CRISPR technologies but also highlights a paradigm shift in how biomedical researchers can utilize AI. "Our findings demonstrated that AI-generated SBs can successfully inhibit the function of target protein complexes," the authors noted, expanding the horizons for developing compact, highly effective therapeutic tools.

By utilizing AI to shrink a massive inhibitory protein into a functional, 82-amino-acid binder, our SNU College of Medicine researchers are paving the way for the next generation of safe, ultra-efficient in vivo genome editing.