Drug development is notoriously expensive, time-consuming, and prone to failure. As therapies become more targeted and complex, the need for smarter, more predictive strategies has never been greater. Model-informed approaches are capable of transforming the development process from discovery to clinical trials and precision dosing.
With many promising compounds failing before reaching the market, and companies often spending vast sums on candidates that go nowhere, long development timelines can eat into patent life, leaving limited time for commercialization. Mathematical modeling helps make better use of data, both in-house and from the public domain, to generate predictions, reduce risk, and accelerate progress. For instance, instead of testing every possible dose or treatment permutation in large patient populations, modeling helps focus on the most promising approaches. This means faster, more focused trials with fewer patients and lower costs. Modeling enables more data-driven decisions by pulling deeper insights from existing datasets, whether generated in-house or from the literature, so companies can be more confident as they move from preclinical to clinical phases. It’s not about replacing experiments or trials, but using modeling to prioritize the right experiments, doses, and strategies. It can simulate scenarios that would be impractical, unethical, or too costly to test directly. This is especially valuable in areas such as pediatrics or early oncology trials where patient populations are vulnerable or limited.
It also gives researchers a shared framework to represent what they know. Traditionally, scientific understanding can live in researchers' heads, but modeling forces an explicit representation of that knowledge. That structure challenges assumptions, explores “what if” scenarios, and adapts as new evidence emerges.
There are ethical and regulatory drivers, too. Regulators are increasingly supporting model-informed drug development, particularly when trials involve high-risk or hard-to-reach populations. Modeling helps reduce patient burden, while generating useful evidence.
The difference between mathematical modeling and machine learning
AI typically learns from historical data, which is great for many problems, but drug development often involves doing something that’s never been done before – such as first in human trials. There’s no historical data for that. Mathematical models, however, are built on fundamental biological or pharmacological principles. They’re interpretable, adjustable, and not “black boxes”. You can see the logic, adjust parameters, and understand how an output was reached. That transparency matters for trust. AI and mechanistic models are not mutually exclusive. AI can speed up model development, but it doesn’t always know which is the right tool to use at the right time.
One often-made misconception is expecting the model to be 100 percent right. Models aren’t oracles; they’re tools to make better decisions than could be made unaided. As George Box famously said, “All models are wrong, but some are useful.” Another is that modeling is only for big pharma. That’s not true. Small and mid-sized biotechs can absolutely use modeling with the same preclinical and clinical data they’d already collect for an IND filing, which can be supplemented with data from the literature.
Modeling is now being encouraged, sometimes expected, by regulators – especially in oncology with initiatives such as Project Optimus, which aims to improve dose optimization in clinical trials by bringing model-informed dose optimization into both trial design and bedside dosing.
From an operations perspective, modeling helps design better trials through the simulation of scenarios, exploration of optimal doses, and a reduction in patient numbers – all while still gathering meaningful data. Early on in a drug development trial, you have limited data. It's like seeing a ball coming at you from a distance: you start moving based on its trajectory, but as you get more information, you refine your position. Modeling works the same way, updating predictions as new data emerges.
Clinical delivery
Rigorously validated mechanistic models, intuitive workflows, Bayesian updating, and point-of-care access all combine to derisk trials and inform patient-by-patient dosing decisions in practice. The global demand for this kind of modeling in trials is increasing – and there is strong momentum toward using modeling in drug development, especially for derisking and accelerating programs.
As for clinical use, there’s huge potential here too, and model-informed drug dosing is a growing market (set to reach over $280 million by 2029). Every patient is different – whether in weight, age, metabolism, or genetics – and all these variances affect drug response. For many drugs, giving every patient the same dose just doesn’t work. As therapies get more expensive and more targeted, modelling will be crucial to ensure patients receive the most effective and safest dose.
Oncology is clearly the initial focus, but the adoption of modeling in other therapeutic areas is happening. Oncology provides a rich landscape to start with because of the amount of data and the high need for precision, but the broader benefit is that as models prove successful in oncology, their value becomes clearer across therapeutic areas. Especially with regulatory bodies encouraging the use of modeling in development and submissions, this exposure is likely to accelerate adoption – helping clinicians and the wider drug development community understand the influence of model-derived evidence in areas where they may have traditionally relied solely on clinical experience. In personalized dosing, models become a tool to enhance, not replace, clinical decision-making.
There’s a huge shift happening across life sciences, and regulators are responding by becoming increasingly supportive of modeling throughout the development process. Their encouragement helps normalize modeling as a standard part of evidence generation, which will inevitably lead to wider use. When modeling becomes embedded in regulatory submissions, it encourages all stakeholders, pharma, CROs, and clinicians, to engage earlier and more often. That visibility will widen its impact and foster greater understanding, acceptance, and trust in model-informed decisions.
All of this innovation comes, however, with a caveat – or a learning curve. Part of the learning curve is simply reframing expectations. Modeling isn’t there to replace clinicians or override judgment. It’s there to support better decisions. The core skill set is a willingness to engage with data in a new way, to interpret, rather than be dictated to. These tools help create evidence that fits into the workflow of people already making tough decisions.
If there is to be a shift, it will be more of a mindset than a technical or technological shift. And it will all be about precision, balance, and increasing success rates for patients and the programs themselves. Using modeling from the start can avoid underdosing or overdosing patients and wasting early trial opportunities. This will generate more efficient trials and better outcomes. These tools, however, aren’t meant to give binary “positive” or “negative” answers. They're guides. That nuance is critical, especially for clinicians. Tools like Bayesian dosing software support decision-making by offering a probability-informed recommendation. They don’t remove the clinician from the equation – they enhance the decision-making process.
As drug development evolves, so must the tools we use to navigate it. Mathematical modeling is becoming a necessity. Whether guiding dose selection, streamlining trial design, or personalizing treatment, modeling offers a smarter, more efficient path to better therapies.