Researchers led by Nobel laureate Jennifer Doudna at the Innovative Genomics Institute and the University of California, Berkeley, have successfully used generative AI to design novel RNA-guided nucleases, a breakthrough in synthetic biology. By combining evolutionary data with inverse protein-folding models, the team created "SynTnpBs"—synthetic variants of TnpB proteins—that match or exceed the editing performance of wild-type enzymes, according to findings published in Science.
Engineering Proteins with Hybrid AI Models
The design process for SynTnpBs diverged from traditional methods by utilizing a hybrid approach. According to structural biologist Petr Skopintsev of the Doudna Lab, the team sought to determine if modern generative models could handle the complexity of RNA-guided nucleases.
The researchers partitioned the design task into two distinct workflows: the DNA-binding interface and the guide RNA–binding interface. By anchoring these designs with fixed sequence information derived from evolutionary data, they allowed AI models to generate sequences that fold into specific, functional structures. This method addresses a common limitation in protein engineering: knowing not just what changes to make, but where to make them, as noted by Eli Bixby, cofounder and head of machine learning at Cradle.
Scaling Experimental Validation
Computational design generates millions of potential sequences, but physical testing remains a significant bottleneck. Biochemist Isabel Esaín-Garcia of the Doudna Lab explains that the team screened approximately 50 top candidates for each interface using Escherichia coli to measure editing activity.
The most promising variants were then assessed for efficacy within human and plant genomes. To confirm the structural accuracy of these synthetic proteins, the researchers utilized advanced imaging techniques, including cryo-electron microscopy. The results revealed that the most effective SynTnpB shared only 77% sequence identity with wild-type TnpB, yet maintained superior functional characteristics. This research also allowed the team to observe a conformational state of TnpBs that had been hypothesized previously but never captured in a laboratory setting.
The researchers used methods including cryo-electron microscopy to characterize some candidates structurally, and the research team was able to identify a conformational state of TnpBs that had previously been proposed but never observed.
Implications for Personalized Medicine
While the current SynTnpBs are experimental, the project signals a broader shift toward bespoke protein engineering. The ability to tailor enzymes with specific properties is a prerequisite for the next generation of personalized medicine.
As the field moves away from relying solely on naturally occurring gene editors, researchers are increasingly marrying structural and evolutionary data to create synthetic tools. This trend suggests a future where gene-editing enzymes can be optimized for individual patient needs, potentially increasing the precision and safety of genomic therapies.
Frequently Asked Questions
What are SynTnpBs?
SynTnpBs are synthetic RNA-guided nucleases designed by AI. They are based on the TnpB family of CRISPR-Cas12-like proteins and were developed to test if generative models could create functional, complex gene-editing systems.
How does the AI design process work?
The researchers used a hybrid approach that combined evolutionary data with inverse protein-folding models. They split the design process into two parts—the DNA-binding interface and the guide RNA–binding interface—to generate specific, functional protein sequences.
Why is this research important for medicine?
This work demonstrates that AI can create custom-designed enzymes. In the future, this could lead to the development of personalized gene-editing tools tailored to specific patient requirements or unique genomic properties.
How were the synthetic proteins tested?
The team screened top candidates in Escherichia coli for editing activity, followed by further validation in human and plant genomes. Structural characterization was performed using methods like cryo-electron microscopy.
Are you interested in the intersection of artificial intelligence and synthetic biology? Subscribe to our newsletter for the latest updates on genomic research and biotechnology breakthroughs.
Related reading