A catch all category called New Approach Methodologies (NAMs), covering organoids and AI toxicity screens, is reshaping how drug companies decide what reaches human trials.
For decades, a lab rat's response was the closest thing to a vote of confidence an experimental drug could get before it reached a human volunteer. A growing share of that pre-human gatekeeping is now being offloaded to lab-grown tissues and machine-learning models under a category called New Approach Methodologies, or NAMs. The toolkit includes organoids, miniature self-organizing clusters of human cells that mimic organ structure, and AI-enabled transcriptomic profiling, which reads which genes a tissue switches on or off in response to a drug. The pitch from the companies building the tools: more human-relevant data, faster decisions, and progress on the 3Rs, the long-standing principle of replacing, reducing, and refining animal testing wherever the science holds up.
The science holds up unevenly. A virtual webinar scheduled for June 17, 2026 and hosted by Scientist.com, a San Diego company that bills itself as a research-orchestration platform, gives the trend a thin calendar peg (Scientist.com press release on PRNewswire). The featured speakers come from three vendors active in the space: BioIVT, which sells human biological samples; HUB Organoids, which builds organoid disease models; and Cyprotex, a contract research organization focused on in vitro ADME and toxicology. Each has a commercial interest in framing the shift as both inevitable and near-complete.
The broader regulatory backdrop helps the pitch. The FDA Modernization Act 2.0, signed into law in late 2022, ended the long-standing federal mandate that drug developers default to animal tests before human trials, and it explicitly permits validated non-animal alternatives. Adoption, though, has been selective. Organoids are strong on narrow, mechanism-driven questions: how a liver cell handles a candidate compound, or how a tumor responds to a specific pathway inhibitor. They are weaker, and often missing entirely, on the systemic questions that determine whether a drug is safe in a whole organism, including immune response, hormone signaling, and the distribution of a compound across tissues.
AI-enabled transcriptomics tries to close some of that gap by reading the gene-expression signature a tissue produces after exposure, then using a model to predict toxicity, classify a mechanism, or flag a hazard the original screen missed. In practice, the false-positive and false-negative rates vary by tissue and assay, and the field lacks a shared benchmark for organoid maturity or for AI model performance on whole-organism endpoints. The most consequential gap is neurotoxicology and developmental toxicology, exactly the systemic questions organoid and AI screens have the hardest time modeling.
Translational scientists writing in drug-discovery and toxicology literature in recent years have continued to argue that well-characterized animal models remain irreplaceable for systemic endpoints, particularly the immune, endocrine, and pharmacokinetic questions that determine whether a drug actually works in a patient. NAMs are increasingly the first screen, not the last word. The FDA has signaled that it will accept NAMs data in regulatory submissions where the methods are adequately validated, but case-by-case acceptance is not blanket approval, and the agency has not set a deadline for retiring the older paradigms.
What to watch in the rest of 2026: federal guidance on microphysiological systems, which would clarify how organoid and organ-on-chip data can substitute for animal studies in specific contexts, alongside updates to validation roadmaps in the U.S. and Europe. If those guidelines land cleanly, NAMs move from a default alternative to a default tool. If they don't, the marketing momentum outruns the regulatory infrastructure, and the field spends another year in the awkward middle.
The vendors on next week's webinar are selling tools that genuinely work, in narrow contexts, and that genuinely have limits, in broader ones. The buyers walking into that session are not just looking for a product demo. They are looking for an answer to a question the regulators have not yet settled: when is a petri dish, a chip, and a model good enough to retire a rat?