Why the Old Drug Discovery Model Is Broken

Drug discovery remains one of the most time-consuming and capital-intensive endeavors in modern science. The typical journey from initial discovery to market approval spans 10–15 years, with costs often exceeding $1–3 billion per approved therapy when failures are included. During this period, patent life erodes and competitive advantage diminishes, while overall industry ROI has fallen to low single digits. Despite record R&D spending, outcomes remain poor: more than 90% of drug candidates fail before reaching patients, particularly in high-need areas such as oncology and neurodegenerative diseases like Alzheimer’s. These inefficiencies reflect not just the complexity of human biology, but also the structural limitations of a legacy R&D model that relies on outdated experimental systems, fragmented data, and diminishing returns on innovation.

Structural Failures in the Traditional Model

01 01. Outdated Preclinical Models

A major driver of clinical failure lies in the limited predictive power of conventional preclinical testing—namely 2D cell lines and animal models

2D Cell Cultures: Flat, homogeneous cultures grown on plastic dishes lack the three-dimensional structure, cellular diversity, and microenvironmental context of living tissues. Without gradients, stromal or immune interactions, and genetic heterogeneity, these models frequently overestimate drug efficacy and fail to capture resistance or off-target effects

Animal Models: Despite their ubiquity, mice and other animals differ markedly from humans in genetics, metabolism, and immune function. Many diseases, such as Alzheimer’s, do not naturally occur in standard lab animals. As a result, drugs that shrink tumors or improve cognition in transgenic mice frequently fail to produce benefits in human trials. The well-known saying “mice lie, monkeys exaggerate” reflects decades of overreliance on models that are biologically convenient but clinically misleading.

Because the foundational models are not human-relevant, entire pipelines are often built on weak or distorted signals. Ineffective or toxic compounds advance into costly trials, while potentially promising ones are discarded too early. This predictive gap contributes directly to the industry’s >90% attrition rate

02. Inefficient Clinical Translation and Data Silos

Even when candidates reach the clinic, inefficiencies persist. Many clinical trials recruit narrow, demographically skewed populations in 2020, 75% of participants in approved-drug trials were white, with limited representation of Asian, Black, and Hispanic populations. Such lack of diversity can lead to drugs that perform well in trials but inconsistently in real-world populations.

Equally problematic is the siloed nature of pharmaceutical data. Preclinical and clinical datasets are often proprietary, fragmented, and rarely shared across organizations. Thousands of compounds that fail for one indication are abandoned entirely, even when they might be valuable in another context. This inability to learn from past efforts or repurpose existing assets represents a structural inefficiency: each program starts largely from scratch, repeating the same mistakes and wasting valuable chemical and biological insights.

The combination of weak models, rising complexity, and redundancy across the industry has led to declining productivity. Many “low-hanging fruit” drug targets—such as those in infectious disease or cardiovascular medicine—were successfully addressed in prior decades. The focus has since shifted to complex, multifactorial diseases such as refractory cancers, autoimmune disorders, and neurodegeneration, where biology is poorly understood and clinical endpoints are difficult to measure. At the same time, intense competition within popular target classes leads to overlapping programs and “me-too” compounds, creating attrition even for scientifically valid drugs. The result is an R&D ecosystem where each new generation of medicines is harder, slower, and more expensive to develop than the last.

03. Diminishing Returns on Innovation

Illustrative Case Studies

Oncology: Cures in Mice, Failures in Humans

Oncology exemplifies the translational gap. Roughly 95% of cancer drugs that enter human trials fail to achieve approval, despite strong preclinical efficacy. Many compounds that eradicate tumors in mice fail to replicate these effects in humans due to missing biological complexity—particularly the absence of a functional human immune system in mouse models. Immunotherapies, for instance, cannot be meaningfully evaluated in immunodeficient mice, yet such models remain the norm for testing. As a result, enormous resources are invested in compounds that appear promising in animals but have little chance of success in patients.

Neurodegenerative conditions: The 99% Alzheimer’s Drug Failure Rate

Conclusion

The legacy drug discovery model—long, expensive, and overreliant on reductionist biology—has reached the limits of its efficiency. While traditional cell and animal systems remain useful for early mechanistic insight, they are poorly suited for predicting clinical performance. The outcome is a process that moves too many ineffective drugs forward and abandons promising ones prematurely, creating high attrition, sunk costs, and delayed therapeutic progress.

Modernizing this process requires a shift toward human-relevant, data-integrated, and predictive systems—including 3D organoids, patient-derived models, and AI-driven analytics

Together, these examples highlight a pervasive issue: success in simplified or non-human models rarely predicts human outcomes. The result is a cycle of inflated expectations, expensive failures, and delayed innovation for patients.

////////////////

///////////////////

References:

Wouters, O. J., et al. “Estimated Research and Development Investment Needed to Bring a New Medicine to Market, 2009–2018.” JAMA, 2020

DiMasi, J. A., et al. “Innovation in the Pharmaceutical Industry: New Estimates of R&D Costs.” J. Health Econ., 47 (2016): 20–33.

Paul, S. M., et al. “How to Improve R&D Productivity: The Pharmaceutical Industry’s Grand Challenge.” Nat. Rev. Drug Discov., 9 (2010): 203–214.

Wong, C. H., et al. “Estimation of Clinical Success Rates Across Disease Areas.” Biostatistics, 20 (2) (2019): 273–286.

Osipova, M. “Dive into the Reasons Behind Preclinical Cancer Models Failure.” Oncodesign Services White Paper, 2023.

Weinberg, P. “Mice Don’t Get Alzheimer’s, So Why Test Alzheimer’s Drugs on Them?” Massive Science, 2021.

Hooker, S., et al. “Diversity in Cancer Cell Lines.” Cancer Epidemiology, Biomarkers & Prevention, 2019.

Izumchenko, E., et al. “Patient-Derived Xenografts Effectively Capture Responses to Therapy.” Annals of Oncology, 28 (2017): 2595–2605.

Ooft, S., et al. “Patient-Derived Organoids Can Predict Response to Chemotherapy in Metastatic Colorectal Cancer.” Sci. Transl. Med., 11 (2019): eaay2574.

Jayatunga, M., et al. “How Successful Are AI-Discovered Drugs in Clinical Trials?” Drug Discovery Today, 29 (6) (2024): 104009.