Why artificial intelligence and in vivo studies are not rivals, but the tandem that biomedical research needs

According to a report published jointly by BCG and Wellcome Trust, applying AI in the early stages of pharmaceutical R&D can generate time and cost savings of between 25% and 50% up to the preclinical phase. An impressive figure. And yet, the failure rate of drug candidates in clinical trials continues to hover around 90%.

Something does not add up. Or does it?

The enthusiasm for AI is real. And so are its limits.

In recent years, artificial intelligence has burst into biomedical research with a clear promise: accelerate drug discovery, reduce costs, and anticipate failures before they occur. Machine learning platforms that identify therapeutic targets, in silico models that predict toxicity, digital twins that simulate physiological responses. The progress is genuine.

But there is a fundamental difference between predicting and validating.

AI is extraordinarily effective at what it does well: processing large volumes of data, finding patterns in genomic sequences, simulating the molecular behaviour of a candidate molecule. What it cannot do, at least not yet, is reproduce the complexity of a living system.

And that matters. It matters a great deal.

The problem of the translational gap

The 90% failure rate in clinical trials is not a laboratory problem. It is a translation problem: what works in a simulation or in a simplified model does not always behave the same way in a real organism, with its cardiovascular system, its immune response, its hepatic metabolism, its individual variability.

Regulatory science knows this. The FDA itself, in its roadmap published in 2025 to reduce the use of animals in research, explicitly acknowledges that AI-based alternative models must be validated through studies in animal models. Not as a bureaucratic step, but as a scientific requirement.

The question, then, is simpler: which in vivo model brings you closest to the real patient?

This is where animal models change the equation

When AI has completed its work (it has screened compounds, identified targets, prioritized candidates) the moment arrives when no algorithm can substitute a living organism. This is where animal models come in: not as an alternative to AI, but as the necessary next step to validate what AI has predicted.

And within the available animal models, the porcine model occupies a differential position. The reason is its extraordinary similarity to human biology in the systems that matter most for biomedical research.

They are not competitors. AI and the animal model are distinct stages of the same process: one accelerates selection, the other validates reality.

The role of a specialised CRO in this context

Designing a preclinical study with the porcine model is not simply housing animals and executing a protocol. It requires translating the clinical question into a rigorous experimental design, defining the correct endpoints, controlling the variables that may compromise the translation of results, and ensuring that the study complies with the regulatory standards that will make it valid before the FDA or the EMA.

That is precisely the point where the difference between a generalist CRO and a specialised one becomes critical.

At Versa Biomedical we work with the porcine model as our reference standard. Our protocols are designed to maximise the clinical translation of results, and our triple certification framework — GLP, ISO 13485 and ISO 9001 — guarantees consistency, traceability and scientific rigour across every project, in full compliance with EU and FDA requirements.

Every study that leaves Versa is designed so that the data you obtain is exactly what you need to take the next step with confidence.

In conclusion

AI is transforming biomedical research. That transformation is real and welcome. But the biological complexity of a living system cannot be compressed into a mathematical model, however sophisticated.

The future of preclinical research is not about choosing between algorithms and organisms. It is about combining them intelligently: using AI where it accelerates, and using the best available in vivo model where it validates.