OpenCRISPR posts wet‑lab success in tests
- Profluent’s OpenCRISPR story is no longer just “AI made a CRISPR protein.” The important part is that OpenCRISPR-1 has now been shown working in real cells. - The concrete detail is how far from standard Cas9 this thing is — more than 400 mutations away from SpCas9 — while still editing human cells. - That matters because CRISPR design has been stuck between natural enzymes and slow lab tweaking; AI-designed editors could widen the menu fast.
CRISPR is a gene-editing tool. Usually, researchers start from enzymes nature already made, then spend years tweaking them. OpenCRISPR is different — Profluent used large language models to design a new editor from scratch, then had to prove it was not just a nice-looking sequence on a screen. That proof is the real story here: OpenCRISPR-1 has shown functional genome editing in human cells, and the work has now moved well past a purely computational claim. ### What is OpenCRISPR, exactly? OpenCRISPR is Profluent’s open release of AI-designed gene-editing systems. The first release, OpenCRISPR-1, is a Cas9-like nuclease plus guide RNA designed with the company’s language models rather than copied from a natural microbe. Profluent put the sequence on GitHub in April 2024 and describes it as available for ethical research and commercial use under license terms. (biorxiv.org) ### Why is the “wet-lab” part such a big deal? Because biology is full of plausible failures. A protein can look right in silico, fold badly in cells, miss its DNA target, or cut too many wrong places. So “wet-lab validation” is the moment an AI design stops being a software demo and starts becoming a biotechnology claim. In OpenCRISPR-1’s case, the key result is that it edited the human genome in cell-based experiments rather than only in computational benchmarks. (github.com) ### What did they actually show? The core paper says OpenCRISPR-1 achieved successful precision editing of the human genome and that several generated editors had activity and specificity comparable to, or better than, SpCas9. Profluent’s own technical page adds that, in HEK293T cells delivered by plasmid, OpenCRISPR-1 showed comparable on-target editing efficiency and higher specificity relative to SpCas9 across a range of on- and off-target sites. It also says the system is compatible with base editing formats. (biorxiv.org) ### Why does “400 mutations away” matter? Because it tells you this is not just Cas9 with a few knobs turned. Profluent says OpenCRISPR-1 is more than 400 mutations away from SpCas9 and nearly 200 mutations away from any other known natural CRISPR-associated protein. Basically, the company is arguing that AI can explore protein sequence space far beyond what researchers usually reach by incremental engineering. (biorxiv.org) ### So is this already better than normal CRISPR? Not in the simple “winner declared” sense. SpCas9 is deeply validated, widely tooled, and familiar to labs. OpenCRISPR-1’s early data are promising because they suggest you can get strong editing while potentially improving specificity, but that is still very different from proving broad superiority across delivery methods, cell types, and therapeutic settings. The catch is that CRISPR platforms live or die on repeatability, off-target behavior, manufacturability, and regulatory comfort. (profluent.bio) ### What changed beyond the first reveal? The biggest shift is maturity. In April 2024, OpenCRISPR-1 was a preprint and code release. By July 2025, the work appeared in Nature, and Addgene now distributes the plasmid, which means outside labs can get the actual construct more easily. That moves the project from “interesting claim” toward “tool other people can test.” ### What happens next? The next bottleneck is scale. (profluent.bio) If AI can generate lots of candidate editors, the hard part becomes testing enough of them in real assays to know which ones are robust, safe, and useful in production settings. That is where platform companies either become real infrastructure for biotech partners — or stall out as impressive design engines with too little validation throughput. That last step is still ahead. (github.com) ### Bottom line The news is not that OpenCRISPR merely exists. It is that an AI-designed editor has cleared the first credibility hurdle that matters in biotech — it worked in cells. Now the question is whether Profluent, and the labs that pick this up, can turn one convincing wet-lab success into a repeatable pipeline. (biorxiv.org)
