This company designed a drug candidate with help from ChatGPT and Gemini

Highlights, news stories and awards from ACS meetings.

Chicago—Artificial intelligence is all the rage in drug discovery as companies build sophisticated platforms to dream up new drugs. But Guibai Liang, cofounder and chief scientist of Sheo Pharmaceuticals, argues that drug discovery can also be done using off-the-shelf AI tools.

Liang, a veteran medicinal chemist who started Sheo after retiring from big pharma, used OpenAI’s ChatGPT and Google’s Gemini to develop a novel renin inhibitor for treating high blood pressure. He presented the work Aug. 24 during a talk in the Division of Medicinal Chemistry at the American Chemical Society’s Fall 2026 meeting (ACS publishes C&EN but is not involved in editorial decisions).

The enzyme renin plays a key role in the first step of the pathway that regulates blood pressure and was a hot target for drug hunters decades ago. “Every big pharma had a big project on renin inhibitors,” Liang told C&EN. But only one renin inhibitor—aliskiren, marketed by Novartis as Tekturna and Rasilez—was ever approved by the US Food and Drug Administration. It’s a mediocre drug, though, largely because of its poor bioavailability, Liang said.

Even though the landscape of abandoned drug candidates is littered with renin inhibitors, Liang thought renin was still a worthwhile target for developing high blood pressure treatments. He and colleagues at Sheo decided to use TAK-272, a renin inhibitor originally developed by Takeda, as a starting point for studies using AI.

To start, Sheo scientists asked ChatGPT and Gemini what Liang calls “critical and revealing questions” about TAK-272’s structure: What is the role of the hydrogen bond donor on the piperidine? Is the oxygen on the morpholine necessary for activity? Is metabolic stability a concern?

The AI tools sifted through the vast literature available on renin inhibitors and answered the Sheo scientists’ questions, essentially gathering information about all the structure-activity relationships and other data known about this class of compounds. “It’s impossible for human chemists to remember all those data to come up with a clear picture,” Liang said.

Using AI’s answers, Liang and his colleagues were able to design their own class of renin inhibitors, which they also used AI to query. They had a contract research organization synthesize 200 of them over several iterations, eventually landing on their lead compound, SHEO-054. Liang didn’t disclose SHEO-054’s precise structure at ACS Fall 2026, but he did include a general structure in his presentation. In tests in a monkey model of hypertension, the compound outperformed aliskiren.

Liang told C&EN that renin inhibitors may be a special case for this kind of AI-facilitated drug design because there’s so much published work on them. Nevertheless, his takeaway is: “If you are not using AI in your medicinal chemistry, start now.”

Samuel H. Gellman, who studies complex molecular phenomena in biological systems at the University of Wisconsin–Madison, attended Liang’s talk at ACS Fall 2026. He was Liang’s PhD mentor more than 30 years ago but was not involved with this work.

Gellman said chemists and biochemists need to figure out how to take advantage of AI tools, and Liang’s talk was a great example of how to do that. AI “is not like anything I have seen in my 40-year career,” Gellman said. “There have been lots of trends, lots of bandwagons to jump on, and I personally just hate doing that. But to me, this just seems totally different. Like if you don’t pick up these tools, you’re a fool.”

Bethany Halford is a Senior Correspondent for C&EN based in Boston.