# Can AI discover new biology? > Source: > Published: 2026-08-25 12:00:00+00:00 It began with an unexpected result in an automated laboratory in Zhangjiang, Shanghai. In traditional drug discovery, laboratory capacity is tightly budgeted, and evaluating unconventional biological ideas is often viewed as an unjustifiable luxury. But when a bundle of in vivo studies we purchased through a contract research organization had more assay capacity than our programs needed, our neuroscience team decided to run a test. We turned to our [AI](https://www.fastcompany.com/section/artificial-intelligence) target discovery platform, PandaOmics, to prioritize candidate targets from massive multimodal datasets. To our surprise, the AI surfaced Target Z—an innovative biological protein target our company was already actively developing for an entirely different disease area. The platform linked this familiar protein to pain signaling mechanisms for the first time. Our scientists used this unused assay slot to evaluate a newly designed molecule engineered to interact with the Target Z protein—a chance to ask a new question. When the experimental data came back, one molecule accomplished what was previously thought to be highly improbable: It outperformed morphine in preclinical pain models. That breakthrough led to ISM9528, an oral, brain-penetrant, non-opioid candidate. Beyond its immediate clinical potential, this moment revealed a fundamental truth for business leaders and technologists alike: The true power of AI in medicine isn’t just speeding up chemical synthesis—it’s engineering scientific serendipity at scale. Pain management represents one of modern medicine’s most intractable challenges and a global market approaching [$100 billion](https://www.thebusinessresearchcompany.com/report/pain-management-global-market-report). Worldwide, [nearly 27%](https://pubmed.ncbi.nlm.nih.gov/42024903/) of the population suffers from chronic pain with significantly higher prevalence in people over the age of 45 (46.7%). In the United States, chronic pain impacts [24.3% of adults](https://www.cdc.gov/nchs/products/databriefs/db518.htm). For decades, drug hunters have been forced into stark therapeutic compromises. Opioids like morphine and oxycodone are highly effective for acute pain, but carry severe liabilities including high risks of addiction, physical dependence, tolerance, and fatal overdose. NSAIDssuch asibuprofen and naproxen are useful for mild inflammation, but are often ineffective against severe pain involving the nervous system, while causing some toxicity during long-term use. And there are other drug classes with additional issues. The industry’s core bottleneck is a shortage of novel, validated biological targets. Rather than attempting to design another incremental variation on well-worn pain receptors, our team used AI to ask a fundamental biological question: *Can we target an entirely different pathway to achieve pain relief without the side effects and risks of existing treatments?* Uncovering a new mechanism requires a closed loop between biological target discovery, human laboratory validation, and generative chemistry. Crucially, this requires distinguishing between two distinct domains: biology (identifying *where* to intervene) and chemistry (designing *what* molecule to use). This is how we approached it. **1. Target discovery (the biological protein):** Our AI platform synthesized human genetic, omics, and text data, identifying the specific protein pathway we designated as Target Z. **2. Experimental validation (human critical thinking):** Our neuroscience team validated the biology. Using gene-knockout mouse models and specialized pharmacology, the team confirmed that inhibiting the Target Z protein reduced pain in the mice. **3.** **Generative chemistry (the chemical molecule):** With the target validated, our generative AI platform, *Chemistry42*, designed and refined new molecules to meet several requirements at once. It designed novel small molecules capable of crossing the blood-brain barrier and selectively binding to the Target Z protein. This yielded ISM9528, our chosen development candidate molecule. When an AI platform uncovers an unexpected drug candidate, business leaders naturally ask if it is an accidental fluke or a repeatable model for industrial R&D. Historically, breakthrough therapeutics—from penicillin to GLP-1 receptor agonists—were discovered through unplanned scientific serendipity. But relying on pure chance in conventional drug discovery is slow, rigid, and prohibitively expensive. Modern AI changes the equation by turning serendipity into a systematic, scalable capability. We addressed platform efficiency by creating foundational AI infrastructure. In early 2026, we introduced *MMAI Gym*, an open training and benchmarking platform that aggregates thousands of standardized biology and chemistry tasks to fine-tune general large language models into specialized scientific expert models. We found that it improves model performance on drug discovery tasks by up to 10-fold. The biopharmaceutical market is signaling strong interest in this platform model. Our lead, fully AI-discovered asset, rentosertib (ISM001-055) for idiopathic pulmonary fibrosis, entered Phase III clinical trials in mid-2026, becoming the world’s first AI-discovered target and molecule to reach late-stage clinical evaluation. Strong commercial interest provides another measure of the platform’s potential. **Insilico has partnered with 13 of the world’s 20 largest pharmaceutical companies, with many returning for additional collaborations.** When machine intelligence illuminates non-obvious biological hypotheses and human scientists systematically validate them, AI evolves from an exploratory research tool into an enterprise-grade engine for medicine. Advancing ISM9528 toward planned clinical trials in 2027 is a significant milestone, but preclinical success is only the beginning. Animal models do not mirror human pain perfectly. How much drug reaches different parts of the body can vary between animals and humans, and new targets in the nervous system can reveal unexpected safety risks in human trials. Human clinical studies will ultimately decide whether ISM9528 fulfills its promise as a transformative non-opioid medicine. Until then, this program illustrates what can happen when AI expands the biological questions scientists can explore and what human researchers can put to the test. Neither replaces the other: AI can uncover connections that might otherwise be missed, while scientists determine which ideas hold up in the real world. *Alex Zhavoronkov, PhD, is founder and CEO of Insilico Medicine.*