Nearly every blockbuster obesity drug of the past five years — Ozempic, Wegovy, Zepbound, Mounjaro — works by activating hormone receptors in the gut and brain. On September 30, 2026, Hong Kong-based Insilico Medicine announced it had used generative AI to design a molecule that does the reverse: it blocks one of those same receptors. The company nominated the compound, called ISM1354, as a preclinical candidate for obesity and related metabolic disease, betting that turning a receptor off rather than on could solve a problem that has dogged the entire GLP-1 drug class — the muscle loss that comes bundled with the fat loss.
Agonize or antagonize: a real scientific fight
ISM1354 targets GIPR, the receptor for glucose-dependent insulinotropic polypeptide, a gut hormone involved in metabolism. Eli Lilly’s blockbuster Zepbound activates GIPR alongside GLP-1 in a dual-agonist approach. Insilico’s molecule does the opposite, blocking GIPR instead. This is not a settled scientific question — it is an active paradox that researchers openly debate in journals like Nature Metabolism, since both strategies have shown success in the clinic: Lilly’s dual-agonist tirzepatide works, and so does Amgen’s GIPR-antagonist-based candidate MariTide. Some lab studies in obese mice have found that both GIPR agonism and antagonism reduce food intake and body weight, with antagonism in some models producing longer-lasting appetite suppression, while other research suggests that sustained agonism of the receptor can eventually desensitize it, functionally turning an agonist into something that behaves like an antagonist over time. In other words, the field doesn’t yet know definitively which approach is better, which is exactly why a company betting heavily on the antagonist path represents a real strategic wager rather than an obvious choice.
Why Insilico says blocking the receptor helps preserve muscle
The company’s argument, articulated by founder and co-CEO Alex Zhavoronkov, centers on a side effect that has become a growing concern as GLP-1 drugs reach millions of patients: substantial lean muscle loss alongside fat loss, a problem of particular concern in older or frail patients. “Obesity is not an isolated health issue; it is a trigger for many diseases,” Zhavoronkov said in the company’s announcement, framing the push for a next-generation approach as a response to unmet needs in the existing drug class. Insilico’s strategy is to eventually combine a GIPR antagonist like ISM1354 with an existing GLP-1 therapy, aiming to cut fat while sparing muscle — addressing a side effect that has become one of the most cited criticisms of the current generation of weight-loss drugs.
How the AI actually designed the molecule
ISM1354 was generated using Chemistry42, Insilico’s proprietary generative chemistry platform, which the company has used across its broader drug pipeline, including a separate program that led to a $2.75 billion licensing deal with Eli Lilly for a different AI-designed candidate. For ISM1354, the R&D team layered multiple prediction modules into the molecule-generation process at once: modules that reinforced binding to the intended GIPR target while penalizing unwanted off-target interactions, a model specifically trained to flag drug-induced liver injury risk for early filtering of unsafe candidates, and free-energy perturbation models to predict binding strength before any molecule was physically synthesized. Co-CEO and Chief Scientific Officer Feng Ren described the approach as proof that “generative AI is demonstrating immense potential in cracking complex drug discovery challenges.” Candidates that passed the computational filters then went through iterative design-make-test-analyze cycles, feeding real lab results back into the model to refine the next round of candidates.
The preclinical numbers behind the hype
In animal studies spanning mice, rats, dogs, and monkeys, Insilico reported that ISM1354 showed oral bioavailability of 75 to 104 percent — a measure of how much of an orally dosed drug actually reaches the bloodstream — and delivered at least 18 times greater plasma exposure than a clinical-stage benchmark compound at equivalent doses. The company also reported an approximately 45-fold safety margin in non-GLP monkey toxicology studies, along with markedly weaker liver-injury signals and transporter-inhibition effects compared with its benchmark. Those are preclinical animal numbers, not human trial results, and obesity drug history is littered with compounds that looked clean in animals before stumbling in humans — a caveat Insilico’s own announcement does not dwell on.
What’s next
Insilico has not disclosed a date for filing an Investigational New Drug application, the step required before ISM1354 can be tested in humans, so the molecule remains, for now, a preclinical bet rather than a treatment anyone can access. The broader significance is less about this one molecule and more about what it represents: an AI drug-discovery company using generative chemistry not just to find more molecules faster, but to pursue a scientifically contested mechanism that a traditional discovery program might have considered too risky to prioritize. Whether GIPR antagonism or agonism eventually wins out in the clinic, ISM1354’s progress will be watched as a test of whether AI-designed obesity drugs can meaningfully improve on the muscle-loss problem that has become the GLP-1 era’s most persistent asterisk.
Photo: PublicDomainPictures / PIXABAY via Pixabay
