Even after years of laboratory research and early clinical testing, most drug candidates fail before reaching patients. Recent advances in AI and computational drug development have yet to change that reality, with the percentage of Phase 2 and 3 trials terminated before completion more than doubling from 11 percent to 23 percent between 2013 and 2023.
While many failures arise from a lack of drug efficacy or issues with safety, fundamental flaws in clinical trial design can also significantly hinder success. Inappropriate patient selection, poorly chosen endpoints, and incorrect dose selection can all undermine a study before it begins. Current modeling approaches are largely retrospective tools, helping researchers understand how a drug behaves in the body or why a clinical study succeeded or failed.
Phase Advance is taking a different approach. Rather than using computational models to interpret existing clinical data, the company is attempting to simulate the clinical trial itself using virtual patients. The company is now applying its platform to LEAF4Life's hypoxia-targeting therapy, KizaVie, across 16 therapeutic programs, including a basket trial spanning seven oncology indications. Using virtual patient populations, the platform aims to prioritize indications while informing dose selection, treatment schedules, biomarker strategies, patient selection, and broader clinical trial design.
A non-linear mathematical model
Drug development already relies on mathematical modeling to understand how therapies behave in the human body. For example, pharmacokinetic/pharmacodynamic (PK/PD) models use observed drug concentration and effect data to describe how a drug behaves over time and how its concentration relates to a physiological response. In contrast, physiologically-based pharmacokinetic (PBPK) models take a more mechanistic, bottom-up approach, using anatomical and physiological parameters to simulate how a drug is absorbed, distributed, metabolized, and excreted throughout the body.
Quantitative systems pharmacology (QSP) models build on both of these approaches by combining pharmacology, systems biology, physiology, and mathematics to model the dynamic interactions between a drug and a biological system.
Now, Phase Advance's approach takes the concept of biological complexity further. "Historically, the goal has been to build the simplest model possible that still explains the biology," Tawanda Gumbo, the cofounder and CEO of Phase Advance, told DDN. "We actually did it the other way around."
Rather than simplifying the body's complexity, Phase Advance attempts to mathematically model biological processes across multiple scales — from subcellular reactions and gene regulation to cells, tissues, organs and ultimately entire patient populations. AI is then used to accelerate the enormous computational workload and identify biological patterns.
This mechanistic approach distinguishes the platform from many AI-based drug discovery tools, which typically require large volumes of experimental or clinical data. Instead, Gumbo said Phase Advance begins with the biological mechanism of a disease and a drug's mechanism of action, allowing the company to simulate how thousands of virtual patients with different genetic backgrounds, lifestyles, and disease histories might respond to treatment.
This approach allows Phase Advance to look into the future, rather than the past. "Whereas most PK/PD and even QSP give you an explanation after you've collected the data — why did this fail, why did this succeed, why did we get these doses — we do it the other way around. We ask, will this work?”
KizaVie as the first major test case
This systems-of-systems-level approach is ideal for a drug that could be used across a remarkably broad range of diseases, including acute respiratory distress syndrome (ARDS), sepsis, traumatic injury, and several solid tumors.
Developed by LEAF4Life, KizaVie is a liposomal formulation of transcrocetin, a natural compound derived from spices such as saffron, designed to enhance the diffusion of oxygen from the bloodstream into oxygen-starved tissues. It does this by promoting the formation of new blood vessels and by improving the diffusion of oxygen into those tissues.
“Many of the most lethal illnesses treated in intensive care begin with very different triggers — sepsis, pneumonia, or trauma, for example — but converge on a shared pathophysiology,” Clet Niyikiza, CEO of LEAF4Life, told DDN. “The initiating injury damages the vascular endothelium and disrupts the microcirculation, impairing the movement of oxygen from the blood into tissues. At that point, hypoxia is no longer simply the endpoint of disease. It becomes an active driver of progression.”
Hypoxia can exacerbate endothelial injury and trigger coagulopathy, creating a deadly feedback loop that can lead to microcirculatory failure, multi-organ dysfunction, and death. This means that, in severe acute illness, survival may ultimately be determined less by the initial cause of disease than by the tissue hypoxia it sets in motion.
ARDS is a particularly severe example of this process, characterized by widespread damage to the alveoli that disrupts gas exchange and leads to dangerously low blood oxygen levels. Moderate-to-severe ARDS is associated with in-hospital mortality between 40 and 45 percent, and treatment remains largely supportive, relying on measures such as mechanical ventilation and prone positioning.
While these interventions can improve oxygen delivery to the lungs and blood, they do not directly address the final step of oxygen diffusion from the microcirculation into tissues. KizaVie is designed to target this process directly.
From COVID to cancer
The drug's first major clinical test was in 2020, as COVID-19 swept through hospitals. Severe COVID-19 frequently led to ARDS, leaving patients with dangerously low oxygen levels and, in the most serious cases, multisystem organ failure.
With intensive care units struggling to cope with the number of critically ill patients, LEAF4Life fast-tracked KizaVie into a Phase 1/2 trial. The early results were encouraging. Within 24 hours, 45 percent of patients had experienced a 25 percent or greater improvement in blood oxygenation. By day three, fewer patients needed noradrenaline to support their blood pressure, while average organ failure scores also began to fall. At 28 days, 11 of the 12 patients enrolled were still alive.
The findings were preliminary, but, according to Niyikiza, the data caught the attention of French regulators. “After the first 18 patients, they told us to stop,” he said. “They said, ‘We think there is something important here. Let's go to Phase 3.’”
That Phase 3 trial is now ongoing for patients with ARDS due to any cause. However, this is far from the only clinical trial investigating Kizavie. “You start asking, ‘How many diseases could you prevent?’” Niyikiza said. “Of course, there are acute diseases such as ARDS and cardiovascular disease. But there are also chronic diseases, including diabetes, as well as ageing-associated diseases such as dementia and impaired healing.”
Another huge opportunity is cancer. As tumors grow, their demand for oxygen can outstrip the supply provided by the surrounding blood vessels. This creates pockets of hypoxia in the tumor environment, changing how cancer cells behave, encouraging proliferation, migration and invasion, while also influencing how tumors respond to chemotherapy, radiotherapy, and immunotherapy. Moreover, tumor hypoxia has been shown to correlate with poorer prognosis.
When Kizavie was tested in mouse models of triple-negative breast cancer, a highly aggressive tumor type associated with extensive hypoxia, it significantly reduced tumor hypoxia, with maximal reoxygenation observed 72 hours post-treatment. Combined with radiotherapy, the drug inhibited tumor growth 1.8 times more effectively than radiation alone, and even produced a complete tumor response and 100 percent survival at 50 days.
“Once you give this drug to an animal with a hypoxic tumor, standard treatments like radiation basically blow the tumor away,” said Niyikiza. Kizavie is now being investigated for soft tissue sarcomas and glioblastomas, while Phase Advance is helping to develop a seven-tumor basket trial. The trial will initially focus on cancers selected for their high levels of hypoxia, including pancreatic, triple-negative breast, lung, colorectal and head and neck cancers.
Is this the future of clinical trials?
Although KizaVie represents the first multi-indication application of Phase Advance's platform, the company has previously tested its models by prospectively predicting the outcomes of 10 clinical trials before the trial results became available. Comparing its predictions with the eventual results from more than 29,000 participants, Phase Advance reported that the selected clinical endpoints were within 99.94 percent of the observed human outcomes.
For both companies, however, the broader opportunity lies beyond a single drug. Rather than relying solely on conventional preclinical studies to decide which programs should advance, mechanistic virtual patient models could allow developers to compare multiple indications, optimize dosing regimens, identify predictive biomarkers, and refine patient selection before launching costly clinical trials.
For Gumbo, the long-term goal is not to replace clinical trials but to make them more efficient. "We can model thousands of possibilities, discard those that are not going to work and concentrate on the ones that have the greatest chance of success,” he said.
Whether mechanistic whole-body modeling can consistently deliver on that promise remains to be seen. But as computational power grows and developers search for ways to reduce clinical attrition, the ability to test numerous possibilities in silico before committing to them in the clinic could become an increasingly important part of drug development.
“We have passed a tipping point,” Gumbo said. “We now have tools that allow us to make medicines faster and, as I like to say, get a glimpse into the future without waiting for the future to come to us.”
For an industry that has spent decades learning from failed clinical trials, the next shift may be towards predicting which ones should happen in the first place.
