Aitia Bio
ew things in medicine seem crueler than giving a placebo to a patient in need. Yet despite ethical concerns, the so-called control arm has been the cornerstone of medical advancement since the advent of modern clinical trials in the 1950s. Without it, the scientific community, patients, and regulatory agencies will lack the data to know whether a new treatment truly outperforms a placebo or existing therapies.
It’s fortuitous then that technology has advanced to virtual representations of patients, or digital twins, that can be used across all aspects of drug development, from discovery of new drug candidates to clinical trial testing. By utilizing newly generated genomic, proteomic, and clinical data, digital twins are being used to improve success rates in translating research results into viable drug targets, reduce the number of people needed in clinical trials, and replace placebo-controlled cohorts with virtual representations of humans.
The Evolution of Digital Twins, from Moonshots to Molecules
Digital twins aren’t new: NASA conceived the idea in the 1960s and used a “living model” to help astronauts on the Apollo 13 mission after its oxygen tank explosion damaged the main engine. And manufacturers have used digital twins to model their own experiments and ideas with physical products. For instance, Unilever used digital twins in its soap and detergent to reduce false alerts that require attention by 90%. In China, Citic Heavy Industries used digital twins and 3D modeling to predict equipment failures and provide operations and maintenance solutions for its cement business, saving more than 30% in costs. KINEXON achieved a 5% increase in their automotive assembly line speed and significantly reduced manual errors and product recalls.
But in healthcare and biomedicine, digital twins have only recently become possible due to the rapid decline of the cost to sequence a person’s DNA—from the $2.7 billion that the Human Genome Project spent by 2003, to Illumina’s $200 price in 2023, combined with the rapid, low-cost generation of high-throughput molecular data measuring the activity of each gene product (multi-omic data), and the exponential rise in computing power. With more genetic and multi-omic data available and generative causal AI running on supercomputers able to rapidly sift through the information, scientists now can experiment in silico (in the computer) before in vivo (in a living animal organism, typically mice) and prior to clinical trials (in humans). And unlike real-world experiments that can take years to determine whether a single experimental drug is safe and effective, with digital twins, researchers can test billions of ideas in hours and days.
Regulatory agencies are working towards adopting new policies and guidelines to allow the use of digital twins in drug development and clinical trials. The FDA released a document detailing current and future uses of the technology, saying they have the potential “to enhance drug development in many ways, including to help bring safe and effective drugs to patients faster; provide broader access to drugs and thereby improve health equity.” In addition, the agency said digital twins “could potentially provide a comprehensive, longitudinal, and computationally generated clinical record that describes what may have happened to that specific participant if they had received a placebo.”
The European Medicines Agency (EMA) has also begun exploring and implementing frameworks that incorporate digital twins, starting with their broader “AI Action Plan,” and a favorable qualification opinion for the application of digital twins in phase 2 and 3 clinical trials. The EMA’s approval marks the first time a regulatory body has endorsed a machine learning-based approach for pivotal trials.
Collaborations between pharmaceutical companies and artificial intelligence (AI) firms are already yielding promising results. Unlearn, an AI company, partnered with Johnson & Johnson to show that digital twins could reduce control arm sizes by up to 33% in phase 3 Alzheimer’s trials. And clinical development data analytics company Phesi demonstrated in June that AI-powered digital twins could replace standard-of-care control arms in trials for chronic graft-versus-host disease.
“Digital twin technology allows sponsors to ‘meet the patients’ before starting a clinical trial; eliminate costly protocol amendments through better alignment between trial design and the target patient population; and improve commercial viability,” wrote Gen Li, PhD, MBA, CEO and founder of Phesi, in a September 2024 article in Applied Clinical Trials.
Pixels over Pipettes: Advancing Pre-Clinical Research
But a bigger revolution in healthcare may come from utilizing digital twins earlier in the drug development process: during the pre-clinical phase that determines which treatments should be advanced to patients. On average, the time from target discovery to approval of a new drug is about 14 years, at a cost of $1 billion or more. And the failure rate is 90%-95%. Given these odds, biopharma companies have been looking for new and better ways to improve their success rate.
In recent years, the biopharma industry has faced a concerning decline in clinical development success rates, with fewer drugs making it through to approval. Factors like complex disease biology and inadequate target selection contribute to high attrition rates, with one recent report pegging the average likelihood of approval for a new phase 1 drug at a depressing 6.7%.
By creating virtual models of biological systems, researchers can better predict which drug candidates are likely to bear fruit in clinical studies, ultimately reducing their dependence on preclinical studies. This isn’t just theoretical; an analysis of AI-discovered drugs found that these therapies are “substantially” more successful than traditionally discovered treatments, with an 80%-90% success rate in phase 1 trials. If these results hold into later phases, it would nearly double the pharmaceutical industry’s existing drug development ability.
In research and development, digital twins can be used to create computational representations of disease that enable scientists to conduct billions of virtual experiments based on human biology. These digital twins are created from reverse-engineering both known and previously unknown genetic and molecular interactions that drive clinical outcomes. The key distinction lies in the use of causal AI: going beyond pattern recognition and correlation as traditionally used in drug discovery to find true causative connections. See the October 2024 issue of Global Forum for a feature on causal AI.
A prime example of this approach can be seen in the use of digital twins in Huntington’s disease. While researchers identified the gene that causes the condition back in 1983, few effective treatments exist. To solve this problem, a digital twin representation of the disease was created using data from some of the largest Huntington’s studies, resulting in a model that includes approximately 23,000 nodes and 5.3 million interactions.
Using this complex model, researchers were able to identify a novel target that affects cognition and motor function in Huntington’s. A small-molecule drug candidate that inhibits this target is expected to slow motor symptom progression without affecting normal cellular metabolism.
This digital twin platform, which combines human multi-omic data and causal AI and pre-clinical and clinical data, is also being applied to a range of oncology and neurological indications, including Alzheimer’s disease, Parkinson’s disease, pancreatic cancer, prostate cancer, and multiple myeloma.
“Right now, the greatest opportunity for use of digital twin methods is around drug discovery—pre-human and manufacturing—as only about 0.1% of drug compounds make it as far as human clinical trials, so our greatest return on investment is increasing this percentage,” said Musaddiq Khan, vice president of digital outcomes and therapeutic areas at Medable, a technology platform provider for clinical trials. “As we are able to collect and investigate more data from different sources—through wearables, health records, genomic data, and historical data sets—we will be even better prepared to simulate specific pathways and systems within a digital twin trial participant.”
Khan added, “It is also imperative that researchers are properly set up to not only collect relevant data but also to make sure that there is an operational advantage using digital twin technology and then ensure that trials are structured appropriately. Then there are ethical considerations, too. How do we validate the prediction? What if two models are behaving differently, driving additional questions around how to standardize across the industry so approvals are appropriate? Digital twins offer huge potential to improve how we do research, but they also require a lot of forethought.”
Bits, Bytes, and Breakthroughs: Charting the Digital Bio-Future
Industry experts see enormous potential for this technology to transform drug development. As the amount and variety of biological data continue to increase exponentially—with advances like single-cell omics providing ever more detailed insights—and as AI and computing capabilities keep pace, expect to see even more accurate and predictive and explanatory digital twins.
Wider use of digital twins can potentially reduce the high attrition rates that have long plagued the pharmaceutical industry. In early-stage drug discovery, they can help identify more promising drug targets and optimized compounds that can inhibit them. In clinical development, digital twins can enrich trial design with patients who will respond to the drug, and help to optimize protocols, predict enrollment rates, and forecast completion timelines.
By simulating patient responses, they may also enable more effective adaptive trial designs, allowing researchers to make data-driven decisions in real time. This could lead to more efficient trials that require fewer patients and resources. They also have the potential to reduce or even eliminate the need for placebo arms in clinical trials.
“Widespread use of digital twins where we will see a transformational impact will require industrywide collaboration and data sharing,” said Khan. “We are moving in the right direction, especially with decentralized trial evidence-generation platforms and AI-powered tools. I anticipate big progress in this area, but the complexity of the human body and drugs being researched means it won’t be easy.
As the field continues to evolve, collaboration between technology companies, drug developers, regulators, and healthcare providers will be crucial. By combining diverse expertise and data sources, the industry can work towards a future where digital twins become an integral part of the drug development process, potentially leading to more effective treatments, reduced development costs, shorter timeframes for bringing a therapy to market, and ultimately, better outcomes for patients.