The emerging paradigm is integration, not replacement.
For decades, the pharmaceutical industry has sent promising medicines into human trials only to watch nine in ten fail—not from lack of effort, but from a foundational mismatch between animal biology and human complexity. A convergence of stem cell science, organoid engineering, organ-on-chip platforms, and artificial intelligence is now offering a different path: New Approach Methodologies that model human physiology from the inside out. This is not merely a technical upgrade, but a philosophical reckoning with how we have long assumed one creature's body could speak for another's. The question now is whether the institutions of science, industry, and regulation can move together quickly enough to realize what the biology already makes possible.
- Nine out of ten drugs fail in clinical trials, a failure rate so persistent it has forced the field to question its most foundational assumption—that animal models reliably predict human outcomes.
- NAMs are creating urgency across labs, boardrooms, and regulatory agencies, as stem cells, organoids, organ-on-chip systems, and AI are being woven into platforms that can simulate human tissue, diversity, and disease in ways no animal ever could.
- The technology surfaces a deeper disruption: drug-induced heart damage, rare genetic responses, sex- and age-based variation—phenomena that have historically slipped through preclinical testing—can now be systematically interrogated before a single human volunteer takes a dose.
- Three bottlenecks are slowing the transition: experimental systems still lack full biological maturity, AI models are only as good as the data feeding them, and regulators have yet to establish the validation frameworks needed to trust these methods at scale.
- The field is converging on a hybrid future—AI-driven prediction paired with cellular and organoid models, standardized globally, automated for scale—but realizing it demands collaboration across academia, industry, clinics, regulators, and patient communities.
Nine out of ten drugs that reach human trials ultimately fail. The reason is not simply bad science—it is a structural mismatch between the animal models used to predict drug behavior and the far more intricate reality of human physiology. That gap has grown too costly to ignore, and it is driving a quiet but consequential transformation in how medicines are discovered and tested.
The emerging alternative is a toolkit known as New Approach Methodologies, or NAMs. Rather than replacing one method with another, NAMs weave together stem cell-derived models, organoids, organ-on-chip platforms, and artificial intelligence so that each compensates for the limitations of the rest. Computational tools excel early in the process—scanning vast chemical libraries, predicting toxicity, identifying promising targets. But computation alone cannot confirm whether a drug works in living tissue. That is where organoids and organ-on-chip systems take over, allowing cells to self-organize into three-dimensional structures that behave like actual human organs under physiologically realistic conditions.
One of NAMs' most significant advantages is their capacity to reflect human diversity. Animal models cannot account for the genetic variation, sex differences, age effects, and environmental exposures that shape how individual patients respond to medicines. Large collections of induced pluripotent stem cells from diverse donors, analyzed at scale by AI, offer a path toward evaluating drug safety across real populations before any trial begins. Concepts like "organoid villages"—collections of different human cell types studied together—could eventually allow researchers to simulate how entire patient populations respond to a therapy.
The field is not without its challenges. Many experimental systems still need improvements in maturity and reproducibility. AI models depend on the quality of their training data. And regulators require robust, consistent evidence before these approaches can enter mainstream drug development. Overcoming these bottlenecks—biological fidelity, data quality, and regulatory confidence—will demand collaboration across academia, industry, clinics, and patient advocacy groups, along with standardized validation frameworks and transparent methods.
The destination is integration rather than replacement: a more predictive, human-centric pipeline that could meaningfully reduce the failure rate that has defined pharmaceutical development for decades. But arriving there will require time, shared infrastructure, and a willingness to fundamentally rethink how we ask whether a medicine is safe before it ever reaches a patient.
Nine out of ten drugs that make it to human trials eventually fail. The culprit, more often than not, is not bad science or bad luck—it's a fundamental mismatch between what researchers tested in the lab and what actually happens inside a patient's body. For decades, pharmaceutical companies have relied on animal models to predict how a drug will behave in humans. These models have taught us a great deal about disease and safety. But they cannot fully capture the intricate complexity of human physiology, and that gap has become increasingly costly to ignore.
This recognition has sparked a quiet revolution in how drug discovery works. Rather than continuing to refine animal testing, researchers are building something different: human-relevant systems that can predict drug behavior with far greater accuracy. These are called New Approach Methodologies, or NAMs—a toolkit of interconnected technologies that includes stem cell-derived models, organoids, organ-on-chip systems, and artificial intelligence. The idea is not to replace one method with another, but to weave them together so that each compensates for the limitations of the rest.
Dr. Joseph C. Wu, director of the Stanford Cardiovascular Institute and co-founder of Greenstone Biosciences, describes the shift this way: computational methods excel at the earliest stages of drug discovery, rapidly searching through vast chemical libraries and predicting toxicity in a fraction of the time traditional labs would need. They can identify promising drug targets and expand the search into chemical spaces that were previously too large to explore systematically. But computation alone cannot validate whether a drug actually works in living tissue. That is where the experimental systems come in. Stem cells can be coaxed into becoming specific cell types—cardiac cells, brain cells, liver cells. Organoids take this further, allowing those cells to self-organize into miniature three-dimensional tissues that behave like actual organs. Organ-on-chip platforms recreate the physical and mechanical environment of living tissue, letting researchers watch how cells respond under conditions that more closely resemble the human body.
One of the most powerful advantages of NAMs is their ability to capture human diversity in ways that animal models simply cannot. Traditional preclinical testing struggles to account for the genetic, biological, and environmental differences that determine how individual patients respond to medicines. A variant in a gene involved in drug metabolism can alter someone's risk of adverse effects or treatment failure. Sex-based differences affect treatment response. Age matters. Environmental exposures matter. These variations are not quirks—they are fundamental to understanding whether a drug will work for a given person. Large-scale collections of induced pluripotent stem cells from diverse donors, combined with computational models that can analyze variation at scale, offer a path toward systematically evaluating drug safety and efficacy across different populations before any human ever takes a pill.
Consider drug-induced heart damage, a serious safety concern. A coordinated NAM platform can assess multiple mechanisms of toxicity—arrhythmia, fibrosis, mitochondrial dysfunction—while accounting for individual differences. Emerging concepts like "organoid villages," collections of different human cell types grown and studied together, could eventually allow researchers to simulate how entire patient populations respond to a therapy before clinical trials begin. This could help identify who is most likely to benefit and who is most likely to suffer harm.
But NAMs are not yet flawless. Many experimental systems still need improvements in maturity and reproducibility. AI models depend heavily on the quality and diversity of the data used to train them. Regulators need robust evidence that these approaches deliver reliable, clinically meaningful results consistently. Wu identifies three main bottlenecks: biological fidelity, data quality, and regulatory confidence. Overcoming them will require collaboration across academia, industry, clinics, regulatory agencies, and patient advocacy groups—and it will require standardized validation frameworks, transparent methods, and clear regulatory pathways.
The field is not abandoning established approaches overnight. Instead, it is building toward integration: AI-driven predictions paired with cellular and organoid models, supported by standardized methods and global data sharing. Automation and scalability will be essential to moving NAMs beyond specialist laboratories and into mainstream drug development. The promise is substantial—a more predictive, human-centric drug discovery pipeline that could reduce the staggering failure rate that has plagued the industry for decades. But realizing that promise will take time, collaboration, and a willingness to rethink how we test whether a medicine is safe and effective before it ever reaches a patient.
Notable Quotes
The emerging paradigm is not replacement, but integration. In silico approaches enable efficient prioritization and design, while experimental NAMs provide mechanistic validation and translational grounding.— Dr. Joseph C. Wu, Stanford Cardiovascular Institute
The most pressing bottlenecks to translational adoption fall into three main categories: biological fidelity, data quality, and regulatory confidence.— Dr. Joseph C. Wu
The Hearth Conversation Another angle on the story
Why has animal testing persisted for so long if it's so unreliable at predicting human outcomes?
Animal models taught us a tremendous amount about disease and safety mechanisms. They were the best tool we had. But as drug development became more sophisticated, the gap between what we learned in mice or dogs and what actually happened in human trials became harder to ignore. The cost of that gap—in failed drugs, wasted research dollars, and patients who didn't benefit—finally became too high to accept.
So NAMs aren't about replacing animals with computers. It sounds more like building a whole new system.
Exactly. A computer can search through millions of chemical compounds in hours, but it can't tell you whether a drug will cause heart damage or how it will behave in someone with a specific genetic variant. A human organoid can show you that, but you'd need to test thousands of them to understand population-wide patterns. Together, they're far more powerful than either alone.
The diversity angle seems crucial. How does that actually work in practice?
Instead of testing a drug on cells from one or two donors, you test it on cells from dozens of people with different genetic backgrounds, ages, and sexes. You feed that variation into computational models that can spot patterns humans might miss. You're essentially asking: will this work for everyone, or only for some people? That's a question animal models can't really answer.
What's the biggest obstacle to making this mainstream?
Trust. Regulators need to see consistent, reproducible evidence that these methods work. Researchers need standardized ways of doing things so results from one lab can be compared to another. And the field needs to move beyond small specialist labs to places where drug companies can actually use these tools at scale. That's not a science problem—it's an infrastructure and collaboration problem.
How soon could this actually change how drugs are developed?
It's already starting. But widespread adoption probably takes another five to ten years. The science is there. What's needed now is standardization, regulatory clarity, and the kind of global collaboration that's harder to build than a new technology.