The search for effective non-opioid pain medicines has produced both breakthroughs and setbacks. While the approval of Vertex Pharmaceuticals’ non-opioid drug Journavx in 2025 offered a new option for acute pain, it didn’t see the immediate commercial success many analysts anticipated, with third-quarter sales falling $3 million short of expectations. This was quickly followed by Vertex then discontinuing a follow-up non-opioid candidate after disappointing clinical results. Meanwhile, Eli Lilly has shelved two non-opioid drug candidates.
Developing non-opioid drugs is difficult because pain is highly heterogeneous and can arise through different biological mechanisms. Patients can experience pain as a result of very different conditions, from diabetic neuropathy to osteoarthritis to acute tissue injury, and the mechanisms driving those experiences can differ accordingly.
Insilico Medicine is now betting that AI can help researchers approach the problem from a different direction. The company recently nominated ISM9528 as a preclinical candidate for pain management, a potential first-in-class, non-opioid therapy.
Using AI to find new biology
To find a new path forward for pain management, the team used PandaOmics, Insilico’s AI engine for identifying and prioritizing therapeutic targets. The platform analyzed omics, textual, and multimodal data to identify a pain-related pattern pointing toward Target Z, a target that had not previously been systematically linked to pain management.
It allows us to essentially look at the biology differently. Is there another way to find a target that we haven't thought about before?
—Carol Satler, Insilico Medicine
Carol Satler, Senior Vice President of Clinical Development, Non-Oncology at Insilico Medicine, told DDN that this ability to approach established disease biology from a different perspective is one of the most exciting aspects of generative AI. “It allows us to essentially look at the biology differently,” Satler said. “Is there another way to find a target that we haven't thought about before?”
The approach is particularly relevant in pain, where researchers have spent decades investigating established mechanisms but continue to struggle to translate promising therapies into the clinic. By identifying biology that has not previously been associated with pain, Insilico is hoping to open up a different therapeutic avenue.
The company then used its Chemistry42 platform to optimize compounds around Target Z, incorporating models designed to improve properties including selectivity, drug metabolism, and pharmacokinetics. The resulting candidate, ISM9528, is orally available and capable of crossing the blood-brain barrier. But can this translate into clinical results?
The challenge of translating pain
Pain drug development has a particularly difficult translational problem. Researchers can measure behaviors associated with pain in animals, but those measurements cannot fully capture the subjective experience of pain in humans. Many established assays rely on reflexive responses, such as withdrawing a limb from an experimentally applied mechanical, thermal, or chemical stimulus.
Satler pointed to several commonly used rodent models, including spinal nerve ligation for chronic neuropathic pain and acetic acid-induced writhing. In these experiments, researchers measure behaviors such as limb withdrawal, twitching, or writhing to determine whether a compound reduced a response to a painful stimulus.
The models provide useful biological signals, but Satler acknowledged that it remains difficult to know how reliably those signals will predict efficacy in patients. “How well does the efficacy that we see in these models ... translate into patients and in humans?” Satler said. “These kind of models may not translate that well to humans, and I think that's another big issue.”
Insilico's preclinical results nevertheless showed encouraging analgesic activity. In a spinal nerve ligation model, ISM9528 dose-dependently reduced the animals’ sensitivity to mechanical stimuli, while in an acetic acid-induced writhing model, the company reported greater analgesic activity than morphine.
In the writhing model, Satler said the team measured the number of pain-related behaviors exhibited by the animals over 30 minutes. Untreated animals showed the most pronounced response, while morphine reduced the number of writhing behaviors by roughly half. Animals treated with ISM9528 showed even fewer pain-related behaviors, with the response described by Satler as “drastically reduced.” However, she cautioned against putting a specific percentage on the difference because the experiments were short and conducted in a small cohort of animals.
“Whether that can translate into the patients would be groundbreaking for treatment and for changing the course of how we treat pain today,” she said.
The candidate also appeared to have a relatively rapid onset and longer duration of action than comparator compounds in preclinical models. These properties could be particularly important for acute pain, which is currently the company’s primary development goal.
The company is currently preparing the toxicology and manufacturing package needed to support an investigational new drug application, with the goal of beginning clinical trials in 2027.
Will AI change the conventional discovery story?
Insilico Medicine has now nominated 32 preclinical candidates since 2021 and says its programs have reached candidate nomination in an average of 12–18 months. Satler said the accumulated knowledge from these programs is itself becoming a resource, allowing researchers to identify and validate new targets increasingly quickly.
Looking more broadly, the familiar narrative of a drug taking 10–12 years and costing around $1 billion to bring to market may be starting to look less inevitable. AI is increasingly being used to compress different stages of the drug discovery process, from revisiting existing drugs and identifying new therapeutic targets to designing novel compounds and simulating clinical trials before they begin.
As Insilico continues to build its pipeline, this process will only become more efficient as it learns from each program. The company has already demonstrated real-world clinical impact with Phase 2a results for rentosertib, the first wholly AI-discovered and AI-designed small molecule drug. The trial showed a favorable safety profile and stabilization or improvement in lung function in patients with idiopathic pulmonary fibrosis. The drug is now in Phase 3, with a planned 320-patient study evaluating once-daily rentosertib over 52 weeks.
Recent setbacks have shown how difficult it is to translate promising pain biology into successful medicines. AI could change that by revealing new biology and mechanisms that conventional approaches have overlooked. For a field that has struggled to develop effective non-opioid treatments, the ability to search beyond established targets could be just as important as the speed at which AI moves a candidate through the discovery pipeline.
