BIONEWS
A small protein fragment encoded in the human genome, not a plant extract or supplement, that in mice suppresses appetite like Ozempic without nausea or muscle loss; the senior author has co founded a company to test it in humans.
How AI helped Stanford find a 'natural Ozempic' — and why it's years from a prescription
A Stanford Medicine team has identified a small protein fragment called BRP (brain-restricted peptide) that, in mice, suppresses appetite and drives weight loss in a way that resembles semaglutide, the GLP-1 drug sold as Ozempic, while sidestepping several of its most common side effects. The molecule is the worked example. The pipeline that found it is the story.
The pipeline is a machine-learning screen of prohormones, the inactive precursor proteins that the body cleaves into shorter, biologically active peptides. Stanford's group fed that prohormone library to an algorithm that predicted which fragments were most likely to behave like hormones. BRP was the standout hit. The Nature paper describing the work was published March 5; Stanford Medicine publicized it on July 19, and the study has since circulated through science-news outlets as a candidate "natural Ozempic."
"Natural" here has a specific meaning that the popular framing tends to flatten. BRP is not a plant extract, a supplement, or a synthetic analogue tweaked in a lab. It is a peptide encoded in the human genome, a piece of a larger prohormone that the body already makes. That distinction matters because the molecule is, in principle, the same chemistry a future drug would have to mimic or deliver, which is part of why the result is being read as more than a curiosity.
Where BRP appears to differ from semaglutide is in where it acts. Semaglutide mimics GLP-1, a hormone that binds receptors in the gut, pancreas, brain, and other tissues. That broad distribution is also why the drug causes nausea, constipation, and the gastroparesis that drives some of the more persistent complaints. BRP, by contrast, acts mainly in the hypothalamus, the small brain region that governs hunger and metabolic rate. In the mouse experiments, that narrower footprint translated into weight loss without the digestive symptoms and without the muscle wasting that has become a focus of long-term GLP-1 use.
The method is worth dwelling on because the same scaffold can be re-run. Prohormone libraries exist for many endocrine pathways, and a screen of that kind is exactly the task a tuned classifier can chew through cheaply. The researchers framed the result in those terms: the AI's job was not to design a new molecule from scratch but to triage a finite list of candidates so wet-lab effort could be spent only on the ones most likely to behave like hormones. If the same pipeline surfaces additional peptides in the next round, the lens scales. If it does not, this stays a single result.
The caveats are load-bearing. The data is preclinical. BRP has not been tested in a single human being. Mouse appetite circuits and human appetite circuits overlap, but they do not match, and the history of obesity-drug translation is full of molecules that worked in rodents and failed in people. The team said human trials are planned but have not started; no timeline has been disclosed.
The team has also co-founded a company intended to run those trials, a financial interest that does not invalidate the work but does change how readers should weight any forward-looking timeline. Senior authors promoting their own commercial pipeline is a routine feature of academic biotech, not a scandal. It is also a reason to be specific about what the press material does and does not establish.
The honest read is that the result is two things at once. It is a single animal result in a field where the bar for a new obesity drug is human efficacy and safety. And it is a documented example of an AI-driven discovery pipeline that produced a candidate worth testing, which is the more reusable of the two stories. The molecule may or may not become a prescription. The pipeline is already running again.