AI‑designed gene‑editing enzymes expand the CRISPR toolbox
tags:
Scientists have made many advances using traditional CRISPR technology, especially in medicine, but they are now seeking ways to create genuinely new gene-editing enzymes with properties that have not already evolved naturally. A new study, published in Science, describes a new AI-designed synthetic TnpB enzyme, called SynTnpBs, that has outperformed the natural reference enzyme.
Creating new gene editors
CRISPR tools use a programmable guide RNA to direct an enzyme to a specific target in the genome to edit (cut, insert or correct) DNA. The TnpB enzyme is a compact ancestor of certain CRISPR enzymes, called CRISPR-Cas12 enzymes. Researchers think its small size could make it useful in situations where delivery space is limited, like some kinds of gene editing in plants. However, these enzymes can be difficult to redesign.
While AI has been useful for automating complex genome editing and predicting DNA repair outcomes, most AI methods used for generating gene-editing enzymes have produced versions that are still very similar to natural proteins. When researchers have attempted to create new editors with novel, useful properties, they have found it challenging to change the protein without breaking the molecular contacts needed for DNA editing.
The authors of the new study write, "Although this approach has been successful for the generation of dynamic switches and DNA binders, the design of complex enzymes, such as RNA-guided nucleases bearing multiple functional domains and conformational states, remains an attractive but unresolved challenge."
A new method for diverging variants
Starting with the known TnpB enzyme structure, the researchers used an AI protein-design model called the ESM Inverse Folding (ESM-IF1) model to propose new sequences. Using evolutionary data, they selected which amino acids were important for RNA or DNA recognition and kept those unchanged. They tested these designs in bacteria, and then tested the most useful candidates in human cells and Arabidopsis plant cells. These active, divergent variants of TnpB were termed "SynTnpBs."
In the bacterial screen, 466 of 1,980 designed protein-part combinations showed detectable activity, with about 8% outperforming the natural reference enzyme. When the new variants were tested in human cells, two variants edited a test gene more efficiently than the natural enzyme, reaching 46% and 50%, versus 28% for the original enzyme. Other new variants were comparable in efficiency to the original. At some human DNA targets, the best designs delivered nearly fourfold higher editing than the natural TnpB.
The team also measured how much the new variants diverged from the original to ensure they were not too similar. They write, "Unlike LMs that generated proteins that retained wild-type DNA binding domains with >99% identity to natural homologs, our structure- and evolution-guided design approach created DNA- and RNA-interacting lobes with AI-generated contacts that had 83% and 72% identity to their closest counterparts in nature, respectively."
The study focused on only one compact nuclease family, so the approach may not work equally well for all gene editors. However, the results are promising, and further refinement and testing may yield better, smaller editing tools that eventually improve genetic research methods and agricultural innovation.
The study authors write, "Establishing the foundation for design of RNA-guided systems and flexible nucleic acid binders with an evolution- and structure-guided model, we envision extending their functions beyond natural sequence constraints through generative biology and toward de novo design, expanding the designable protein space."
Written for you by our author Krystal Kasal, edited by Gaby Clark, and fact-checked and reviewed by Andrew Zinin—this article is the result of careful human work. We rely on readers like you to keep independent science journalism alive. If this reporting matters to you, please consider a donation (especially monthly). You'll get an ad-free account as a thank-you.
Publication details
Petr Skopintsev et al, Structure and evolution-guided design of minimal RNA-guided nucleases, Science (2026). DOI: 10.1126/science.aed6123
Who's behind this story?
Freelance science writer with Master's in physics. Five years clinical research and physics education experience. Science communicator. Full profile →
MA in English, copy editor since 2021 with experience in higher education and health content. Dedicated to trustworthy science news. Full profile →
Master's in physics with research experience. Long-time science news enthusiast. Plays key role in Science X's editorial success. Full profile →
© 2026 Science X Network
Citation: AI‑designed gene‑editing enzymes expand the CRISPR toolbox (2026, July 17) retrieved 17 July 2026 from https://phys.org/news/2026-07-aidesigned-geneediting-enzymes-crispr-toolbox.html
This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no part may be reproduced without the written permission. The content is provided for information purposes only.