Beyond the Animal Model: Computational Modeling and the Future of Human-Relevant Preclinical Science

Animal models have long been central to biomedical research and drug development. From understanding disease mechanisms to evaluating new therapies, researchers have historically relied on mice, rats, dogs, or non-human primates to predict how humans might respond. But as the cost of failed clinical trials continues to rise and the pressure for ethical, efficient, and predictive methods intensifies, the limitations of traditional animal models have become too glaring to ignore.
In recent years, the biomedical field has been forced to confront a fundamental truth: results from animal studies often fail to translate to humans. Despite decades of investment, studies indicate that up to 90% of findings from preclinical animal research fail to replicate in human trials, particularly in areas such as neuroscience, immunology, and oncology. The reasons are multifold: species-specific physiological differences, underpowered experimental designs, protocol variability, and overly artificial test environments all contribute to a widening translational gap. What works in a rodent in a highly controlled lab rarely works in a metabolically and genetically diverse human population.
The implications of this gap are significant. Not only do failed transitions waste billions of dollars and years of research, they also expose patients to interventions that may lack therapeutic value or carry unforeseen risks. Moreover, growing societal and regulatory concerns around the ethics of animal testing have led to increasing demand for alternative approaches.

The Rise of New Approach Methodologies (NAMs)

In response, the scientific community has invested heavily in New Approach Methodologies (NAMs)—a diverse suite of non-animal technologies that aim to replace or complement traditional in vivo testing. These include human-derived in vitro systems, 3D tissue models, organs-on-chips, and human data–driven computational models.
Organs-on-chips (OOC), for instance, replicate key aspects of human physiology—fluid flow, tissue–tissue interactions, mechanical strain—within a controlled microfluidic environment. These models can simulate organ-level responses to drugs and toxins far more accurately than static cell cultures or animal surrogates. Similarly, advanced human cell-based assays and stem cell–derived platforms now allow researchers to study drug metabolism, toxicity, and disease mechanisms in highly human-relevant contexts.
Yet while NAMs provide unprecedented biological realism, their predictive power is still often limited to local, molecular, or cellular scales. They struggle to account for the complex, system-level behaviors—like drug distribution, multiorgan interactions, or whole-body feedback loops—that ultimately determine a therapy’s success or failure in clinical settings. This is where computational modeling becomes indispensable.

Computational Modeling: The Missing Integrator

Mechanistic and multi-scale computational models serve as a bridge between NAMs and whole-body predictions. These models simulate how biological systems respond over time to internal and external inputs (like drugs, mutations, or disease processes) using equations grounded in physiology, pharmacokinetics, and systems biology.
For example, PBPK (physiologically based pharmacokinetic) models can extrapolate concentration data from in vitro assays or organs-on-chips to predict how a drug distributes across tissues in the human body. Quantitative systems pharmacology (QSP) models go further, capturing dynamic interactions between drugs, targets, signaling pathways, and clinical biomarkers across time and patient subpopulations.
In practice, combining NAMs with computational modeling achieves several key benefits:

• Translatability: It enables researchers to translate localized NAM results to systemic effects, supporting decisions on dose selection, safety margins, and biomarker response.
• Scenario testing: Through virtual experimentation, researchers can simulate “what-if” conditions (e.g., genetic variability, comorbidities, or high-risk dosing) that would be impractical or unethical in real-world settings.
• Uncertainty quantification: Modeling frameworks allow for rigorous sensitivity analyses, helping identify which biological parameters most affect outcome predictions.
• Regulatory traction: Regulatory agencies (e.g., FDA, EMA) increasingly recognize model-informed approaches as supportive evidence in lieu of some animal studies—particularly when NAMs are embedded in a quantitative, transparent modeling framework .

These advantages do not merely reduce animal use—they improve predictive accuracy and bring mechanistic clarity to preclinical development. They allow researchers to generate and test hypotheses, uncover failure modes, and design smarter experiments—all before reaching human trials.

From Theory to Standard Practice

Despite the promise, full-scale adoption of NAM-modeling integration faces cultural and infrastructural challenges. Many labs lack the expertise to develop and validate complex models. Academic publishing still favors empirical over computational results. And funding agencies and regulators are just beginning to formalize standards for how these tools should be used in submissions.
But momentum is growing. International coalitions like the US National Academies, the European Union’s PARC initiative, and the OECD’s AOP framework are establishing roadmaps for a more model-driven, human-relevant preclinical ecosystem. Importantly, these initiatives emphasize not just replacing animal tests—but building mechanistic, explainable, and integrative frameworks that link preclinical and clinical worlds

Conclusion

It is no longer enough to critique the limits of animal models. The scientific community must build and scale alternatives that are more human-relevant, ethically sound, and quantitatively robust. By integrating NAMs with computational modeling, we can finally move beyond “look-alike biology” toward predictive science that is grounded in mechanism and designed for people—not proxies. In this paradigm, modeling is not an accessory—it’s the architectural backbone of modern biomedical research.

References

• Pound P, Ritskes-Hoitinga M. Is it possible to overcome issues of external validity in preclinical animal research? Why most animal models are bound to fail. BMJ Open Sci. 2020.
• Leist M et al. Consensus report on NAM-based safety assessment. Arch Toxicol. 2023.
• Robinson S et al. Integrating human-based in vitro models and PBPK modeling for regulatory toxicology. Front Pharmacol. 2021.
• Marx U et al. Biology-inspired microphysiological systems to advance patient benefit and animal welfare. ALTEX. 2022.
• National Academies of Sciences. Reducing the Use of Animals in Research: The Role of Computational Modeling and NAMs. 2023.
• Gkatzis K et al. Mechanistic integration of organ-on-chip data in human systems pharmacology models. CPT Pharmacometrics Syst Pharmacol. 2024.
• U.S. FDA. Advancing Regulatory Science Through NAMs and Modeling. White Paper. 2023.

Originally published at https://www.linkedin.com/pulse/computational-modeling-vs-high-cost-failure-rate-drug-ramyar-azar-wxmhf