Predictive in vitro models are reshaping how researchers approach disease modelling, drug discovery and preclinical testing.
In a recent thought leadership webinar, Professor Martin Knight, Professor of Mechanobiology at Queen Mary University of London and Co-director of the Centre for Predictive in vitro Models, was joined by fellow Co-director Professor Hazel Screen, Professor of Biomedical Engineering, alongside Helena Rannikmae, Scientific Leader, Translational In Vitro Models at GlaxoSmithKline, and Emily Richardson, Biology Group Leader at CN Bio.
Together, they discussed how these models can move from innovation into routine use, and how collaboration across academia, industry, technology providers and regulators will be critical to reducing animal use and accelerating therapeutic delivery.
Professor Knight explained that predictive in vitro models cover a broad range of technologies, including organ-on-a-chip systems, organoids, spheroids, 3D tissue models and ex vivo tissue models. These platforms can support fundamental research and preclinical testing, offering more human-relevant approaches that may reduce reliance on animal studies while improving confidence in therapeutic development.
A central theme was translation. While organ-on-a-chip and organoid technologies have advanced significantly, their adoption in routine drug efficacy testing remains limited. According to Professor Knight, one major barrier is translational readiness. Many models developed in academic laboratories work well in specific settings but are not yet robust, scalable or accessible enough for wider industry use. Technological challenges, such as inline monitoring of key parameters without sacrificing the model, also remain important.
The panel highlighted why many promising pilots fail to become routine workflows. Models are often not refined to the level required by industry, particularly around reproducibility, robustness and clinical validation. To gain confidence among end users, models must use accessible materials and platforms, generate consistent data, demonstrate relevance to clinical outcomes and be designed around a clear context of use.
Professor Knight stressed that validation must be “fit for purpose.” The validation strategy should depend on the specific question the model is intended to answer. Disease models for therapeutic efficacy testing are often more complex than toxicity models used in safety testing, because disease biology may require multiple mechanistic readouts rather than a single endpoint. The guiding principle, he noted, should be to make models “as complex as needed and as simple as possible.”
From an industry perspective, Rannikmae explained that large pharmaceutical companies are actively using predictive models to support decision-making in drug discovery and development. These models are being applied to improve mechanistic understanding, strengthen safety assessments and support earlier, more confident go/no-go decisions. However, she noted that safety applications are currently progressing faster than disease efficacy modelling because the context of use and endpoints are often clearer.
For disease modelling, Rannikmae emphasized that no single model is likely to answer every question. Instead, pharma often needs a set of carefully defined models, each addressing a narrow biological or mechanistic question. To expand adoption, developers must avoid overstating what models can do and instead provide evidence that a model recapitulates specific human biology for a defined purpose. Robustness, reproducibility, transferability and functional validation using reference compounds are all critical.
The discussion also explored technical barriers, including material absorption, flow-driven responses and pharmacokinetic realism. Rannikmae noted that these limitations do not necessarily prevent model use, provided they are well characterized, controlled and interpretable. Combining in vitro data with in silico modelling, quantitative systems pharmacology and AI-driven approaches may help extend interpretation and manage complexity, although the field is still at an early stage.
Collaboration was identified as essential. Professor Knight noted that successful translation requires input from model developers, end users, technology providers, CROs, regulators, clinicians, bioengineers and disease experts. At Queen Mary’s Centre for Predictive in vitro Models, which he co-directs with Professor Screen, affiliates from more than 100 organizations are working across academia, pharma, SMEs, regulators, technology providers and the third sector to support wider adoption.
Richardson described how technology providers can help bridge the gap between innovation and application. CN Bio develops both organ-on-chip platforms and biological models, while also offering contract research services that allow customers to assess and access the technology without immediately bringing it in-house. She also highlighted the importance of working with CROs, academic groups and pharmaceutical partners to support onboarding, training, standardization and context-specific model development.
A further challenge is skills. Professor Knight pointed to a clear skills gap, with a need for scientists and engineers who can design, build, validate and apply complex in vitro models while understanding industry and regulatory requirements. Queen Mary is helping to address this through a Centre for Doctoral Training, which will train 60 PhD students, and through an MSc programme in organ-on-a-chip technology.
Overall, the discussion made clear that predictive in vitro models have significant potential to reduce.
