AI-Virtual Trials Spot Hidden Drug Failures As Global Spending Surges
The global biopharmaceutical sector continues to pour roughly $140 billion each year into human clinical testing, yet only a mere 12 % of candidate medicines secure regulatory approval. That stark disparity—between the massive financial outlay and the modest success rate—has long been a source of frustration for researchers and investors alike. Now, an emerging wave of artificial‑intelligence–driven simulations is turning the tide. By constructing millions of virtual patients from the ground up, companies are uncovering drug failures before a single dose is administered to real humans, potentially slashing costs and accelerating the path to market.
The Scale of the Problem
Every year, pharmaceutical giants commit fortunes to discoveries that rarely materialise. The pipeline remains crowded with promising molecules, yet the odds of any given candidate making it through the gauntlet of human trials remain stubbornly low. According to a recent McKinsey report cited by Reuters, investment in AI‑powered drug discovery surged to $8.4 billion in 2025—a figure more than double what was recorded in 2023—and the focus of this funding has sharpened precisely where the technology proves most potent: early‑stage activities such as molecule design and virtual modelling. While significant progress has been made in areas like property prediction and molecular optimisation, the biggest bottleneck—the lengthy, costly demonstration of efficacy in living humans—remains untouched. The consequence for the industry is staggering: a drug that looks promising on paper often fails in the messy reality of human biology, leading to wasted resources and shareholder disappointment. When a late‑stage trial for Novartis’s experimental muscular‑dystrophy treatment, del‑desiran, collapsed in early October, the stock plummeted by $30 billion and the share price dipped 11 %, shattering the CEO’s optimistic projection of $5 billion in annual sales. The episode underscores just how precarious modern drug development really is, and how much uncertainty still lingers even beneath the surface of sophisticated analytics.
How Virtual Simulation Is Changing the Landscape
At the forefront of this transformation stands BioinvestGPT, a Copenhagen‑based startup that has made headlines for delivering remarkably reliable forecasts. During a July simulation involving thousands of virtual patients, the AI platform predicted that del‑desiran would offer only an insignificant clinical benefit—outcome that aligned perfectly with the disappointing real‑world results. Over the past year, the company has logged five correct predictions out of six, including successes for a melanoma vaccine jointly developed by Moderna and Merck and failures for several high‑profile candidates. This track record has attracted interest not just from pharma houses but also from venture capitalists eager to back platforms capable of de‑risking the development pipeline.
The core of BioinvestGPT’s approach lies in its ability to model a human body at the molecular level. By analysing DNA sequences and mapping them onto a digital avatar that mirrors the exact eligibility criteria for a particular trial, the system can then execute intricate virtual trials in a matter of days—or weeks, depending on the scenario. Researchers describe the process as akin to conducting an experiment inside a computer, where variables can be tweaked instantly and hundreds of scenarios explored simultaneously. “We can pinpoint the reason why a drug is effective and safe, and in many cases, why not,” explains Bragi Lovetrue, co‑founder of BioinvestGPT. This granular insight goes beyond simple pass/fail verdicts; it offers mechanistic explanations that can guide medicinal chemists toward better designs before a single synthesis step is taken in the lab.
The speed advantage is perhaps the most compelling selling point. Traditional Phase 1 safety studies typically require months of recruiting, dosing, and monitoring, followed by increasingly elaborate Phase 2 and Phase 3 programmes that demand billions of dollars in expenditure. AI simulations, by contrast, can generate probabilistic outcomes in a fraction of that time—sometimes less than a month. For companies looking to allocate limited capital efficiently, the ability to filter out dead ends early on represents nothing short of a paradigm shift. “We shouldn’t only ask how to run trials faster,” remarks Francisco Beca, chief medical officer at QuantHealth, an Israeli‑based platform that combines real‑world data with artificial intelligence, “We should ask how to run fewer trials that are going to fail.” This philosophy is reshaping how investment decisions are made across the sector.
Case Studies in Prediction Power
The real‑world stakes of these predictive models become crystal clear when examining the specific outcomes that have been forecasted. The del‑desiran failure stands as the most prominent example, but the platform’s catalogue extends well beyond this single case. Early simulations correctly anticipated the negative result of the melanoma vaccine partnership between Moderna and Merck, delivering a timely warning to stakeholders. Conversely, the same system flagged the weaker-than-expected performance of AstraZeneca and Ionis’s heart medication, Wainua, offering a clearer view of therapeutic gaps. On the opposite side of the spectrum, the model celebrated the success of Vaxcyte’s pneumococcal vaccine, underscoring its capacity to identify bright prospects amidst a crowded pipeline.
Looking ahead, BioinvestGPT has already issued forecasts for several high‑impact trials. Late‑stage investigations involving Novartis’s pelacarsen—a compound targeting the cholesterol carrier lipoprotein(a)—were projected to fail due to mis‑estimated mechanisms regarding genetic variation. Similarly, the platform predicts negative outcomes for Biogen’s litifilimab in lupus patients and for Takeda’s Zasocitinib, a TYK2 inhibitor under study for both lupus and psoriatic arthritis. Each of these predictions carries implications for strategic decision‑making: companies can redirect budget away from paths that history suggests are unlikely to yield returns, whilst competitors might accelerate funding for those deemed more viable. The cumulative effect is a more disciplined allocation of research capital, one that respects the brutal realities of human variability while still allowing for bold innovation.
Yet the technology is far from infallible. Earlier in 2025, the same team had forecast positive results for Pelacarsen, only to learn that Novartis later admitted the drug failed to reduce the risk of major cardiovascular events in patients carrying a specific genetic marker. A subsequent internal review attributed this discrepancy to a fundamental flaw in the underlying mechanism—specifically, the failure to account for genetically determined differences in lipoprotein size. Such setbacks highlight that virtual trials, while powerful, cannot entirely replace empirical validation. They serve instead as a vital complement, providing richer hypotheses than any model alone could generate. “Predictive AI can tell us where a drug might struggle,” observes Bragi Lovetrue, “but the final arbitrator remains the human body.”
Industry Perspectives and Ethical Questions
The broader community is grappling with the philosophical ramifications of delegating so much of clinical decision‑making to algorithms. Diana Gallagher, Biogen’s head of clinical development for multiple sclerosis, immunology and Alzheimer’s, acknowledges that the company utilises every tool at its disposal, including artificial intelligence, yet warns that reliance on such models raises serious ethical concerns. “Is it still ethical in 2026 to expose patients to a trial that most likely will fail?” she ponders. “With the advancement of this technology, coming in 2027, 2028, or even 2029, probably the answer is going to be that it no longer is.” Her sentiment is echoed by Francisco Beca, whose observation captures the shifting mindset within the sector: “We shouldn’t only ask how to run trials faster. We should ask how to run fewer trials that are going to fail.”
Industry analysts are equally divided on the trajectory of regulation. The United States Department of Health and Human Services recently announced initiatives aimed at streamlining clinical development pathways, hoping to incorporate predictive AI into the formal approval process. If these measures gain traction, they could fundamentally alter how safety and efficacy are demonstrated, possibly reducing the number of costly and sometimes unethical trials that plague the current system. Meanwhile, investment in AI drug discovery continues its upward climb, with McKinsey estimating that 2025 saw nearly $8.4 billion poured into the space—more than twice the investment recorded in 2023. This surge signals a maturing field, one where computational insights are becoming integral to the very fabric of pharmaceutical strategy.
Why it Matters
The convergence of astronomical spending, dismal success rates, and cutting‑edge computation presents a pivotal moment for the life sciences. By harnessing virtual human models, AI startups are transforming a historically unpredictable landscape into one where risk can be quantified and managed with unprecedented precision. The potential to cut the average cost of bringing a new drug to market from billions to mere tens of millions could unlock a wave of otherwise untested treatments, especially for rare diseases that have long suffered from dwindling patient pools. Moreover, as the evidence mounts that predictive simulations reliably identify likely failures, the economic and humanitarian benefits of reduced waste become undeniable. Ultimately, the dialogue surrounding AI in drug development forces a reckoning over what constitutes responsible innovation: does efficiency trump the need to protect participants in uncertain experiments? As the technology matures and regulatory frameworks evolve, the balance between accelerated progress and ethical stewardship will define not only the future of pharmaceuticals but also the trust society places in the entire healthcare enterprise.
