Improving translatability in preclinical research: reducing clinical failure through predictive in vivo models

More than 90% of drug candidates that enter clinical trials never reach patients. According to the FDA, a 10% improvement in predicting preclinical failures would save $100 million per drug in development costs. The root cause, in most cases, is poor translatability: the inability of preclinical models to accurately predict what will happen in humans. 

Root causes of poor translatability in preclinical research and their impact on clinical failure

A 2022 analysis of clinical trial data from 2010 to 2017 by Sun et al. identifies four root causes behind the 90% failure rate of clinical drug development: lack of clinical efficacy (40–50%), unmanageable toxicity (30%), poor drug-like properties (10-15%), and inadequate strategic planning (10%). Critically, this 90% figure applies only to candidates that have already entered Phase I; it does not include the many more that fail in preclinical stages. The actual attrition across the full drug discovery continuum is substantially higher.

Despite decades of investment in target validation, high-throughput screening, structure-activity relationship (SAR) optimization, and increasingly sophisticated toxicology assays, the overall success rate of clinical drug development has remained at 10-15%. Each step of the pipeline has been rigorously improved, yet the system-level outcome has barely changed. What’s holding up the progress?

First, preclinical research tends to overemphasize potency and plasma exposure as selection criteria. Yet what ultimately determines clinical dose, efficacy, and toxicity is the balance of drug exposure in disease-targeted tissues versus healthy organs, which is not usually evaluated systematically. Second, safety and efficacy assessments typically occur independently in conventional workflows, missing interactions only visible in a whole-organism context. A compound may clear a cell-based efficacy screen while carrying organ-level safety liabilities that are not detected until costly rodent studies. 

On top of these issues is the broader problem of reproducibility and model relevance. Many preclinical findings are irreproducible or do not transfer to human biology because the models used are often poorly matched to the human disease being studied. Animal models, despite their utility for understanding disease mechanisms, have limited predictive power when the biological discrepancy between species is not adequately accounted for. Furthermore, data generated across different technological platforms (in vitro assays, in vivo models, omics approaches) is frequently not integrated in a way that would allow a more complete picture of a compound's profile to emerge before the transition to clinical development.

How predictive in vivo models improve translational confidence in drug development

Improving translatability begins with selecting preclinical research models that are biologically relevant to the human condition being studied. Two principles reinforce predictive confidence: generating efficacy and safety data within the same organism, and demonstrating activity consistently across independent model systems.

When a compound performs consistently across independent measures, the probability that this performance will translate to clinical studies is substantially higher than when efficacy and safety data are drawn from separate, non-interacting platforms. There is also a specific translational principle that emerges from this approach: compounds that show activity across species may translate better clinically.

Zebrafish occupy a uniquely positioned place in this framework. As a vertebrate organism with approximately 82% of human disease-related genes conserved, they provide systemic, whole-organism data that in vitro models cannot replicate. Their small size allows experiments in multi-well plate formats, enabling automated, high-throughput compound testing across large cohorts simultaneously. Many studies support the idea that the results obtained in zebrafish are later confirmed in mice, increasing clinical translatability. 

A great example is the large-scale phenotypic screen in a zebrafish model of retinitis pigmentosa of 2934 human-approved drugs (Zhang et al. 2021). 9 of 11 leads identified held up in a subsequent mouse model validation, demonstrating activity across species. This cross-species consistency is precisely the kind of signal that increases confidence in clinical translation.

Rather than running safety and efficacy evaluations as separate studies in different model systems, zebrafish enable both to be assessed within the same organism. At ZeClinics, we develop tailored disease models that allow efficacy and multi-organ preclinical toxicity studies in the same biological context, generating a more integrated and predictive compound profile than disconnected in vitro and in vivo systems can provide.

Integrating translatability-driven strategies into preclinical study design and CRO workflows

Improving translational outcomes is not only a question of which models to use, but it is also a question of when and how they are integrated into the development workflow. The most impactful point to introduce a zebrafish-based in vivo screen is after a primary in vitro or in silico campaign has identified an initial hit list, and before those candidates are advanced into mammalian studies. At this stage, the scientific question is straightforward: which of these compounds is worth the cost of a rodent study?

This kind of decision support is precisely what a preclinical contract research organization (CRO) partnership should deliver. The goal is to provide the scientific basis for better go/no-go decisions, earlier, faster, and with greater confidence. At ZeClinics, we have built our solutions around this principle. Our capabilities span the full early-stage discovery workflow: from CRISPR/Cas9-based disease model generation (licensed under the Broad Institute of MIT/Harvard) to high-throughput phenotypic screening using automated platforms such as the VAST BioImager and DanioVision, and deep-learning-assisted image analysis for quantitative, multi-parametric readouts. This infrastructure means that you will receive decision-ready insights, supported by validated protocols and interpretable results that are designed to inform what happens next in your pipeline.

The combination of a biologically relevant whole-organism model, multi-species orthogonal data collection, and rigorous experimental design can substantially improve the predictive value of early-stage findings. For organizations looking to reduce attrition, compress timelines, and bring better-validated candidates into clinical development, integrating zebrafish into the preclinical workflow is a competitive need. 

Sources

Seyhan AA. Lost in translation: the valley of death across preclinical and clinical divide. Transl Med Commun. 2019;4:18. doi:10.1186/s41231-019-0050-7

Sun D, Gao W, Hu H, Zhou S. Why 90% of clinical drug development fails and how to improve it? Acta Pharm Sin B. 2022 Jul;12(7):3049-3062. doi: 10.1016/j.apsb.2022.02.002.

Zhang L, Chen C, Fu J, Lilley B, Berlinicke C, Hansen B, Ding D, Wang G, Wang T, Shou D, Ye Y, Mulligan T, Emmerich K, Saxena MT, Hall KR, Sharrock AV, Brandon C, Park H, Kam TI, Dawson VL, Dawson TM, Shim JS, Hanes J, Ji H, Liu JO, Qian J, Ackerley DF, Rohrer B, Zack DJ, Mumm JS. Large-scale phenotypic drug screen identifies neuroprotectants in zebrafish and mouse models of retinitis pigmentosa. Elife. 2021 Jun 29;10:e57245. doi: 10.7554/eLife.57245.

Zon LI, Peterson RT. In vivo drug discovery in the zebrafish. Nat Rev Drug Discov. 2005 Jan;4(1):35-44. doi: 10.1038/nrd1606.

By Javier Terriente

Javier is the co-founder of ZeClinics and ZeCardio Therapeutics, two biotech firms specializing in zebrafish-based preclinical drug discovery for cardiovascular, neural, and toxicology applications. He combines scientific leadership with business acumen, having successfully driven fundraising efforts and strategic partnerships.

Currently leading scientific efforts at ZeCardioTx (and formerly CSO at ZeClinics), Javier also serves on the Board of Directors of AseBio, where he advocates for industry collaboration. His academic background includes a PhD in Molecular Biology and a Marie Curie Fellowship. Recognized as an expert in zebrafish models, he has published extensively and has supervised five industrial PhD theses.