Why Drug Discovery Needs Human-Relevant Models

Developing a new medicine is a long, costly, and complex process with significant risk. Despite decades of scientific advances and billions of dollars invested in research and development, around 90% of drugs entering clinical trials ultimately fail. Often, this is because the preclinical models used to predict how a drug will behave in humans are not physiologically relevant enough to accurately predict efficacy and safety.

For years, animal models have played an important role in preclinical drug development, helping researchers understand disease mechanisms and evaluate potential safety risks. Although approaches have advanced considerably, traditional models still cannot fully capture the complexity of human physiology. As a result, researchers are increasingly shifting their focus away from refining animal models and towards developing more predictive human-relevant systems instead.

This shift in thinking has given rise to New Approach Methodologies (NAMs)—a broad and evolving collection of technologies designed to improve how we study disease, predict the safety and efficacy of new medicines, and assess other regulated products.

Rather than relying on a single innovation, NAMs bring together advances in stem cell biology, organoids, organ-on-a-chip technologies, and artificial intelligence (AI), creating a more human-relevant approach to drug discovery.

According to Dr. Joseph C. Wu, director of the Stanford Cardiovascular Institute and co-founder of Greenstone Biosciences, the promise of NAMs relates to their ability to work together.

“The emerging paradigm is not replacement, but integration,” Wu explained. “In silico approaches enable efficient prioritization and design, while experimental NAMs provide mechanistic validation and translational grounding. Together, this complementary framework supports a more predictive, scalable, and human-centric drug discovery pipeline.”

NAMs: A growing toolkit for human-relevant research

The term NAM is used to describe a range of technologies, each designed to answer different scientific questions.

Stem cell-derived models, including those generated from induced pluripotent stem cells (iPSCs), allow researchers to grow specific cell types (e.g., cardiac, brain, or liver cells) in the laboratory. Organoids take this a step further by enabling these cells to self-organize into miniature, three-dimensional tissues that recapitulate key aspects of organ structure and function. Meanwhile, organ-on-a-chip and microphysiological systems can be used to recreate the physical and mechanical environment of living tissues, allowing scientists to study how the cells respond under more physiologically realistic conditions.

More and more, these experimental systems are being paired with AI-driven computational models capable of analyzing huge datasets, identifying patterns, and predicting how potential therapies may perform before they are physically tested in a lab.

Importantly, these technologies are not competing to replace one another. They should be seen as an interconnected toolkit, where the strengths of one approach balance the limitations of another.

For example, AI is transforming the earliest stages of drug discovery. Computational models can rapidly search vast chemical libraries, identify promising drug targets, and predict potential toxicities in a fraction of the time required by traditional laboratory approaches.

“In silico approaches are delivering the greatest value in early and intermediate stages of the drug discovery pipeline, particularly in target identification, molecular design, and safety prediction, where computational scalability accelerates timelines and reduces cost,” said Wu. “By replacing physical throughput with computational exploration, these methods enable faster interrogation of disease biology and chemical space at a human-centric and population scale.”

Wu emphasized that computational approaches also expand access to previously underexplored chemical spaces, for example, by enabling systematic mining of the hidden diversity within natural products.

Beyond generating new experimental data, AI-driven approaches can also help researchers revisit existing clinical and real-world datasets. By analyzing these large datasets retrospectively, computational models may uncover population-specific differences in drug response that were previously overlooked or difficult to detect in clinical trials.

However, computational modeling is only one piece of the puzzle. “Later-stage translation and mechanistic validation remain more dependent on experimental systems,” Wu noted.

He continues: “Model performance is constrained by data quality, representativeness, and standardization, and predictions often lack mechanistic resolution. Human-centric systems, such as iPSC-derived models, organoids, and organ-on-chip platforms, are therefore essential to validate biological effects and assess functional outcomes.”

Capturing the diversity of human biology with NAMs

Building confidence in new approach methodologies

Despite their potential and growing popularity, NAMs are not “flawless”.

Many experimental systems still require improvements in maturity and reproducibility, while AI models remain heavily dependent on the quality and diversity of the data used to train them. At the same time, regulators need robust evidence demonstrating that these approaches can consistently deliver reliable and clinically meaningful results.

According to Wu, collaboration is key to helping overcome these challenges.

“The most pressing bottlenecks to translational adoption fall into three main categories: biological fidelity, data quality, and regulatory confidence,” he explained. “Strengthening trust will depend on demonstrating robustness and transparency across academia, industry, clinicians, the public, regulatory agencies, and patient advocacy groups, without slowing innovation.”

Rather than replacing established approaches overnight, Wu believes the field should focus on developing standardized validation frameworks and integrating complementary technologies that can collectively provide stronger evidence for decision making.

Realizing the potential of NAMs will require closer integration of AI-driven predictions with cellular and organoid models, supported by standardized methods, FAIR data sharing, and clear regulatory pathways. He adds that automation, scalability, and global collaboration will be key to moving NAMs beyond specialist settings and towards broader adoption.

 A new generation of drug discovery

As NAMs continue to advance, their impact is likely to extend far beyond the laboratory. Realizing their full potential will require new approaches to training researchers and closer collaboration between academia, industry, and regulatory agencies.

“Current academic and industry training models have laid a strong foundation, but they are not yet fully aligned with a future where AI, organoids, and regulatory science converge,” said Wu. “Rather than replacing existing models, we probably need to evolve toward a new kind of drug discovery science.”

NAMs help researchers build more biologically relevant models that better reflect the complexity and diversity of human biology, and they also offer a unique opportunity to democratize drug discovery if they are deliberately designed with accessibility, standardization, and global collaboration in mind.

However, Wu cautioned that these technologies “are likely to carry the risk of becoming concentrated in well-resourced institutions, particularly given the technical complexity, infrastructure demands, and data requirements involved.”

“Ultimately, the goal is not just to train more scientists, but to build a multidisciplinary workforce embedded in collaborative ecosystems. By aligning technical innovation with regulatory insight and societal responsibility through public–private partnership-driven frameworks, we can empower a new generation of scientists capable of translating NAMs-driven advances into practical, trustworthy, and globally accessible solutions,” concluded Wu.