Drug development remains a high-risk, high-cost endeavour, with a huge proportion of candidates failing to reach approval – often because preclinical models fall short of capturing the complexity of human biology. Companies like Calico Biosystems bridge this gap. We spoke to Marie-Agnès Doucey, Chief Executive Officer, Riccardo Turrini, Chief Scientific Officer, and Lucie Maillet, Chief Business Officer, about bringing together digital prediction and biological reality to drive more precise and efficient drug development.
Limitations of traditional drug development models
Despite decades of progress, the pharmaceutical industry continues to face a sobering reality: more than 90% of cancer drug candidates fail before reaching approval. The deeply inefficient pipeline – which Riccardo dubbed the ‘valley of death’ – highlights a core issue: the models used to predict human responses are still not sufficiently representative of human biology.
Riccardo explained: ‘Cancer drug development has long relied on cell cultures, animal models and, more recently, organoids. While these tools are important, in 2026 they still cannot fully reproduce human-specific patient heterogeneity (between individuals and within tumours) and the tumour microenvironment (where multiple cell types interact dynamically).’
The consequences are significant. Drug development timelines can exceed ten years, with billions of dollars invested, only for most candidates to fail in late-stage clinical trials.
Combining biological data and digital models to optimise pipeline
A new generation of approaches is emerging to address this gap. Among them, digital twins – virtual representations of patients built from biological and clinical data – are gaining traction. Digital twins aim to simulate how a disease evolves and how it may respond to treatment. But, as Marie-Agnès asserted, ‘digital twins are only as reliable as the data they’re built on’.
That’s why, prior to introducing digital models, Calico Biosystems uses native patient tumour specimens to identify promising cancer drug candidates. Testing first on these specimens – taking into account their microenvironment and cellular diversity – generates experimental data that closely reflects clinical outcomes. In turn, this high-fidelity data is fed into an algorithm that produces refined predictions – which improve iteratively over time.
‘We want to unite the best parts of biological and computational approaches,’ Marie-Agnès stated. ‘In fact, our approach is something of a virtuous circle. We can – and do – test multiple drugs on the same patient specimen, which creates a richer dataset on which to train our predictive model, which then generates more accurate predictions, which ultimately provides a more robust foundation for clinical decision-making.’
The result is a shift from hypothesis-driven to hypothesis-confirmed drug development. In Riccardo’s words: ‘We’re not guessing; our predictions are grounded firmly in reality. So far, we’ve conducted nine different validation studies involving over 300 patient tumours and our predictions consistently mirrored the clinical responses observed in patients. This enables the pharma and biotech companies we work with to make better decisions earlier in the pipeline – and ensures only drug candidates that have a high chance of success are actually tested on patients in clinical trials.’
The road to real change: Regulatory and market shifts
The promise of new methods – like Calico’s – only throws into relief the limitations of the current preclinical model. Increasingly, regulators are recognising this. As Lucie put it: ‘The regulatory landscape is clearly moving in our direction, with the Food and Drug Administration (FDA) in the US formally removing longstanding animal testing requirements in favour of New Approach Methodologies (NAMs). The European Medical Agency (EMA) is also advancing in a similar direction.’
After all, the economic pressure is hard to ignore. To be sure, the global oncology drug market reached approximately $256bn in 2025 and is projected to grow to $700bn by 2030. But even more strikingly, more than $100bn is lost each year due to failed drug development, largely due to lack of efficacy.
‘Technologies that enable earlier, more accurate decision-making are not just scientific innovations – they are economic imperatives,’ Lucie asserted. ‘The question, for us, is not whether our methods will be adopted but how quickly they will become the new standard.’
As the Calico team pointed out, the shift is already underway – and they expect it to accelerate in two phases. First, in the next three to five years, data-driven platforms will evolve from complementary tools to primary decision-making systems, guiding early-stage development and clinical strategy. Then, over the next decade, as large-scale datasets accumulate, predictive modelling will underpin a new era of precision oncology, where clinical trials validate well-informed hypotheses – rather than explore uncertain ones – with accelerated timelines.
‘The high failure rates we see today are not inevitable – they result from lacking the tools needed to understand the complex dynamics of biological responses,’ Lucie said. ‘But these tools are now emerging and we believe the industry will reorganise around them, bringing massive benefits to both drug developers and patients.’
Collaboration at the core of better drug development
With all this said, the team emphasised the importance of working with (rather than against) the industry to implement their approach.
‘Our solution is disruptive,’ Marie‑Agnès said, ‘but its adoption won’t be. To have a significant impact, we need to leverage the industry’s existing strengths and integrate our platform seamlessly.’
Crucially, Calico’s platform is designed as a plug-and-play accelerator for both traditional R&D and emerging AI-driven pipelines. ‘Our solution fits perfectly into existing workflows,’ Lucie affirmed, ‘as well as enriching and accelerating data-driven models.’ Riccardo echoed this sentiment: ‘What we want to promote is integration. Our systems can be usefully integrated into all types of drug development pipelines.’
Indeed, digital twins are often described as the future of medicine – some would even argue that they’re increasingly becoming part of its present. But focusing on digital twins at the expense of clinical validation comes with its own risks – which is why an integrated, collaborative approach is so important. ‘Digital twins are going to make a massive difference, that’s for sure,’ Marie-Agnès said. ‘But they must be validated by robust biological and clinical data; otherwise, they will still produce drugs that fail. That’s why we need to synergise – to drive success together.’
It can’t be denied that these new methods have massive potential. In an industry long defined by uncertainty, they mark a decisive shift from conjecture to clarity – and from clinical failure to clinical success.
