Scientists at Profluent Bio trained a protein language model on more than a million CRISPR systems mined from microbial genomes, then asked it to write entirely new Cas9-like proteins that had never existed in nature.
The model generated 4.8 times as many families of CRISPR proteins as are found in nature. Testing selected designs in human cells, the team found OpenCRISPR-1, a protein about 400 mutations away from natural Cas9, that edited DNA about as efficiently while making far fewer stray cuts.
In lab tests, OpenCRISPR-1 matched standard Cas9's editing rate in human cells while cutting off-target edits by about 95 percent, and because its sequence differs so much from any natural protein, it may trigger a weaker immune reaction in future therapies. Profluent released the design publicly. It has not been tested in animals or people as an actual treatment.
Read the original study
Ruffolo et al., Nature, 2025 · doi.org/10.1038/s41586-025-09298-z
A protein with no ancestor, built letter by letter.
Source
- Ruffolo, J. A., Nayfach, S., Gallagher, J., Bhatnagar, A., Beazer, J., Hussain, R., Russ, J., Yip, J., Hill, E., Pacesa, M., Meeske, A. J., Cameron, P., & Madani, A. (2025). Design of highly functional genome editors by modelling CRISPR-Cas sequences. Nature, 645(8080), 518–525. https://doi.org/10.1038/s41586-025-09298-z