Profluent‑Lilly $2.25B generative protein deal

Genetic medicine is trying to fix a hard problem — not just how to cut DNA, but how to rewrite bigger chunks of it cleanly enough to become real drugs. That is the backdrop for the new Profluent-Lilly deal. On April 28, Profluent said Lilly signed a multi-program collaboration worth up to $2.25 billion to develop AI-designed recombinases, which are enzymes that can cut and rejoin DNA at specific sites. Lilly gets exclusive rights to take selected candidates into development and commercialization. (biospace.com) ### What did they actually sign? This is not a one-drug licensing deal. It is a platform partnership. Profluent will use its generative AI models to design and optimize custom site-specific recombinases across multiple genomic targets, and Lilly will choose candidates to push through in vivo work, preclin(biospace.com)2.25 billion, and tiered royalties on future sales. (biospace.com) ### What is a recombinase? A recombinase is basically a DNA editor with a different trick from CRISPR. Standard CRISPR systems usually make a cut and rely on the cell’s repair machinery to finish the job. Recombinases are built to recognize particular DNA sequences and then rearrange, insert, or remove DNA more directly. That matters because some diseases may need more than a tiny edit — they may need a full gene inserted or a larger broken segment replaced. (precisionmedicineonline.com) ### Why is “kilobase-scale” the big phrase here? Because that is the promise that makes pharma pay attention. A lot of first-generation editing tools are strongest at small changes. But many genetic diseases would be easier to treat if you could place long stretches of DNA exactly where you want them. Profluent and Lil(precisionmedicineonline.com)tional payloads. Think less typo correction, more swapping in a missing paragraph. (biospace.com) ### Why does AI matter here? The pitch is not just speed. It is search. Biology has an absurdly large design space, and natural proteins cover only a sliver of it. Profluent’s whole model is that large language models trained on protein and CRISPR-related sequence data can generate useful enzymes that do (biospace.com)natural CRISPR proteins while still working in human cells. (github.com) ### Has Profluent shown this is more than a slide deck? At least enough to get Lilly to write a very large option check. Profluent is still early, but it has published Nature work around OpenCRISPR-1 and has been positioning itself as an AI-first enzyme design company rather than a standard drug startup built around one lead asset. That distinction matters — Lilly is buying access to a design engine it hopes can produce many editing tools over time. (nature.com) ### Why Lilly? Lilly has been spending aggressively beyond its obesity franchise, and this fits that pattern. The company has recently piled into business-development deals across genetic medicines and adjacent platforms, using its cash flow to buy shots on goal in areas that could define the next decade of drugmaking. So this deal reads less like a one-off experiment and more like portfolio construction. (fiercebiotech.com) ### What is the catch? The catch is that designed enzymes are not approved medicines. They still have to hit the right DNA sites, avoid the wrong ones, get delivered into the right tissues, and stay safe enough for human use. Large edits are usually more powerful, but they are also the harder engineering problem. A flashy collaboration number mostly tells you Lilly thinks the upside is worth chasing. It does not tell you the biology is solved. (precisionmedicineonline.com) ### Bottom line? This deal is really a bet that AI-designed proteins are graduating from demo projects into core biotech infrastructure. If that bet is right, the biggest winners may not be the first flashy enzyme, but the companies that can keep generating the next one. (biospace.com)