Gene-editing proteins now come from language models

What it does

OpenCRISPR-1 is a Cas9-like nuclease and guide RNA pair generated entirely by Profluent’s large language models, not harvested from bacteria. It preserves the classic Type II Cas9 architecture and NGG PAM preference, so it slots into existing protocols and even accepts canonical SpCas9 gRNAs. The protein sits hundreds of mutations away from any known natural CRISPR-associated protein and has been described in a 2025 Nature paper.

The interesting bit

Profluent is treating its protein design engine as a generative compiler for biology: its LLMs designed a functional enzyme distant from every known natural CRISPR protein, and the company is releasing the sequence under a free commercial license—provided you sign an ethical-use agreement for therapeutic applications.

Key highlights

• Generated from scratch by LLMs, not discovered in natural genomes
• Drop-in replacement for SpCas9 workflows, compatible with standard gRNAs
• Validated in peer-reviewed Nature research with protocols detailed in a preprint
• Free for research and commercial use, though therapeutic use requires a lightweight license agreement
• Can be adapted into deactivated or nickase forms for base, prime, or epigenome editing

Caveats

• Only one editor (OpenCRISPR-1) is released so far; future versions are promised but not scheduled
• IP has been filed, so “open” means freely licensed rather than unpatented
• Commercial therapeutic use requires executing a license agreement that includes ethical-use obligations

Verdict

Gene-therapy researchers and bioinformaticians seeking a Cas9 alternative with a fully synthetic pedigree should bookmark this; traditional software developers looking for a pip-installable package will find only sequences and citations.