How Digital Twins Are Reshaping Clinical R&D
By Matt Truppo, Ph.D., Global Head of Computational & AI Strategy, R&D, Sanofi
Clinical drug development remains the longest, most expensive, and most critical phase of biopharmaceutical R&D. On average, bringing a single new molecular entity to market requires more than a decade and more than $2 billion1,2 — a reality largely driven by a staggering 90% failure rate3 in human clinical trials.
Over the past decade, the cost of advancing investigational assets into clinical testing has roughly doubled without a corresponding rise in success rates or a reduction in development times.2
For an industry striving to deliver potentially lifesaving breakthroughs to patients faster, this trajectory is unsustainable. We are now in an era where we must move beyond the traditional trial-and-error paradigm that has characterized clinical research for decades. At Sanofi, we are embedding AI and advanced predictive modeling across our value chain to transform how medicines are discovered, tested, and manufactured.
One of the most promising technologies advancing this shift is digital twinning — the creation of virtual patient populations and digital process models that mirror real-world biological systems and clinical environments. By simulating human physiology, disease progression, and drug behavior in silico, digital twins allow us to evaluate the safety and efficacy of drug candidates with greater precision, streamline trial recruitment, optimize manufacturing, and bring treatments to patients with unprecedented speed.
Digital Twins Are Rooted In Real Human Biology
Digital trials with virtual patients go far beyond simply feeding legacy trial data into standard machine learning algorithms. Instead, they depend on deep patient profiling and a mechanistic understanding of disease pathways.
At the core of this approach is quantitative systems pharmacology (QSP) modeling. QSP integrates all available data on disease biology, pathophysiology, cell types, cytokines, known pharmacology, and clinical trial results into a single computational framework. By combining these biological mechanisms with clinical data from real patients, researchers can construct digital twins that accurately simulate an investigational compound’s mechanism of action (MOA).
Testing new treatments in digital patients before or alongside human trials offers profound operational and scientific advantages. Researchers can achieve faster, highly accurate initial assessments, identify specific patient subgroups most likely to respond, and optimize dose selection. Crucially, digital twins can also signal early when a compound is unviable. Being able to terminate unpromising programs sooner spares human participants unnecessary trial exposure and allows R&D teams to reallocate critical resources toward more promising compounds.
Bypassing Trial Phases Without Compromising Rigor
Virtual clinical trials allow sponsors to evaluate investigational compounds at earlier stages, compress study timelines, and reimagine the conventional progression of human trial phases.
A compelling example of this approach in action involved a Phase 1b proof-of-mechanism study for a biologic being investigated to treat asthma. Before advancing the asset into the next clinical phase, Sanofi researchers needed to determine whether the compound could deliver a meaningful clinical benefit over existing therapies in a crowded treatment landscape.
To evaluate this, researchers deployed QSP-based virtual asthma patients incorporating relevant cell types, cytokines, and physiological endpoints, such as lung function and annual exacerbation rates. The digital twin model was first validated in a blind prediction against the Phase 1b trial data, demonstrating a close match to observed human results.
Confidence in the digital twin's predictive accuracy was high enough that our team bypassed the traditional Phase 2a dose-finding study entirely, proceeding directly to a Phase 2b trial. Bypassing Phase 2a saved nearly one year of development time for the compound while maintaining complete scientific and regulatory rigor.
Furthermore, every virtual trial creates a self-sharpening loop. As ongoing clinical studies generate fresh data, these insights continuously feed back into Sanofi's virtual patient models across therapeutic areas, regardless of the treatment's specific MOA. This continuous learning cycle creates a dynamic scientific memory of the disease, steadily refining model accuracy for every future trial. This virtuous cycle enables us to make more informed program decisions faster.
Easing Patient Burden And Streamlining Clinical Recruitment
Recruiting participants for clinical studies is one of the industry's most persistent bottlenecks. By incorporating RWD into QSP models to enrich patient cohorts and analyze biological heterogeneity, sponsors can optimize Phase 3 study designs with far sharper inclusion and exclusion criteria. This results in smaller, highly differentiated patient cohorts without sacrificing statistical power.
This capability is especially transformative in rare diseases, where trial recruitment is severely constrained by very small patient populations. In a study for an enzyme replacement therapy treating acid sphingomyelinase deficiency (ASMD) — a devastating, extremely rare genetic disorder — QSP digital twins played a decisive role in trial design and regulatory approval.
By calibrating QSP models with adult and pediatric biomarker data, researchers demonstrated mechanistic similarity in disease progression and treatment response across age groups. This computational evidence supported pediatric extrapolation, enabling regulatory approval for both adult and pediatric patients without requiring a separate full-scale pediatric trial.
Overcoming Biological Complexity With Foundation Models
While the benefits of digital twins are clear, the technology faces inherent challenges. Digital twins are most effective in disease areas supported by rich biological and clinical data sets. In highly complex diseases where underlying genetics, biomarkers, and tissue microenvironments are less understood — such as certain cancers — current models face limitations in generalizability and predictive power. Data quality, cross-system integration, and international regulatory harmonization also remain critical hurdles.
To bridge these gaps, Sanofi is investing heavily in next-generation computational modeling and data infrastructure. A key initiative is a research collaboration with the BioMed X Institute in Heidelberg, Germany, focused on developing a comprehensive biology foundation model. Trained in large-scale multimodal data sets spanning genomics, transcriptomics, clinical trial results, and RWE, this deep learning framework aims to decode complex disease mechanisms and predict clinical efficacy even in sparse data environments.
At the same time, regulatory authorities in both the U.S. and Europe are making significant strides in recognizing the validity of computational modeling. Emerging guidance from agencies such as the FDA and EMA increasingly supports the integration of modeling, simulation, and digital twin evidence in formal regulatory submissions.
The AI Co-Pilot: Governance And The Future Of Clinical Trials
As biopharma accelerates its digital transformation, establishing clear governance around AI and modeling is essential. At Sanofi, digital twins and AI systems are designed to function as intelligent co-pilots — augmenting human expertise rather than replacing scientific judgment.
Every strategic decision, from selecting target molecules to defining patient cohorts and filing regulatory submissions, remains firmly under the authority of human experts. By pairing predictive computational "brains" with expert oversight and automated laboratory execution, scientists and machines operate as true partners, with each cycle of experimentation refining the next.
As these technologies mature, scaling digital twins across the global biopharma ecosystem promises to bend the time and cost curve of drug development. By reducing candidate failure rates, streamlining recruitment, and expediting regulatory reviews, digital twinning will play a pivotal role in delivering innovative, high-quality therapies to patients everywhere with unprecedented speed.
References:
- DiMasi JA, Grabowski HG, Hansen RW. Innovation in the pharmaceutical industry: New estimates of R&D costs. J Health Econ. 2016;47:20-33. doi:10.1016/j.jhealeco.2016.01.012
- Terry C, Chapman D. Measuring the return from pharmaceutical innovation. Deloitte Global. March 19, 2025. Accessed August 18, 2026. https://www.deloitte.com/us/en/industries/life-sciences-health-care/research/measuring-the-return-from-pharmaceutical-innovation.html
- Sun D, Gao W, Hu H, Zhou S. Why 90% of clinical drug development fails and how to improve it? Acta Pharm Sin B. 2022;12(7):3049-3062. doi:10.1016/j.apsb.2022.02.002
About the Author:
Matt Truppo, Ph.D., is global head of computational & AI strategy, R&D, at Sanofi. He oversees platform capabilities across all therapeutic areas and modalities, embedding AI and advanced predictive technologies throughout the research and development value chain. With more than 25 years of biopharmaceutical experience, he focuses on leveraging computational approaches to accelerate drug discovery, optimize clinical trials, and bring first-in-class therapies to patients worldwide.