Digital twins of ex vivo human lungs enable accurate and personalized evaluation of therapeutic efficacy

nature biotechnology Article https://doi.org/10.1038/s41587-026-03121-4 Digital twins of ex vivo human lungs enable accurate and personalized evaluation of therapeutic efficacy Received: 24 September 2024 Accepted: 2 April 2026 Published online: xx xx xxxx Check for updates Xuanzi Zhou1,2,3, Bo Wang4,5,6,7, Yiyang Wei1,2,3, Serena Hacker 1,2, Sumin Kim1,2,4, Thomas Borrillo1,2, Abby McCaig1,2,3, Haaniya Ahmed1,2, Youxue Ren1,2, Olivia Hough1,2, Luca Orsini1,2,8, Bonnie T. Chao1,2, Micheal McInnis9,10, Marcelo Cypel1,2,3,11, Mingyao Liu1,2,3,11, Jonathan C. Yeung 1,2,3,11, Lorenzo Del Sorbo1,2,8, Shaf Keshavjee 1,2,3,11,13 & Andrew T. Sage 1,2,3,11,12,13 Digital twins are an emerging concept in healthcare that envisions integration of molecular, physiological, functional and clinical data to create computational models of biological systems such as cells, organs and individuals. However, the lack of large, multimodal datasets has so far precluded the realization of comprehensive digital twins in medicine. Ex vivo lung perfusion (EVLP) allows the study of human lungs outside the body under physiological conditions and generates multimodal data from imaging, physiologic monitoring and molecular assays. Here we report lung digital twins developed from the largest known clinical EVLP dataset. We show that the digital twin framework accurately models >75 parameters spanning lung physiology, biochemistry, radiography, transcriptomics, metabolomics and proteomics. Furthermore, direct comparison to experimental data on EVLP lungs treated with alteplase demonstrates that digital twins can precisely assess therapeutic efficacy. Together, these results establish human lung digital twins developed using EVLP as a data-rich approach to improve the evaluation of therapeutic effects. Digital twins (DTs) are comprehensive computer-based representations of physical objects. Engineers have recognized the potential of DTs to accelerate research and development by modeling mechanical systems and machines (for example, engines and automobiles)1,2. In healthcare and medical research3, however, high-fidelity multiscale DTs, spanning cellular and clinical data, remain unrealized. Current reports of DTs in medicine are generally extensions of prognostic disease models that do not fully embody the comprehensive nature of DT technology4–6. Ex vivo organ perfusion systems have been developed for human lungs7, livers8, hearts9, pancreases10 and kidneys11 and have been used to assess the condition of donor organs for transplant suitability before surgery. Ex vivo organ perfusion provides an opportunity to develop comprehensive, multimodal DTs of human organs that extend the DT concept beyond engineering to biology. A full list of affiliations appears at the end of the paper. Nature Biotechnology To explore this prospect, we developed a method for creating high-fidelity DTs of human lungs using data generated during ex vivo lung perfusion (EVLP) (Fig. 1a). To date, we have accumulated data and biospecimens from over 1,000 clinical EVLP cases, which have been shown to improve the precision of decisions about organ suitability for transplantation12–14. EVLP is a self-contained, closed circuit that perfuses an isolated human lung at 37 °C (that is, body temperature) while allowing the organ to ‘breathe’ and perform its physiological function of gas exchange using an intensive care unit (ICU)-grade mechanical ventilator7 (Fig. 1b). During EVLP, biological samples (for example, tissue and perfusate) and integrated sensors collect lung-specific data, generating multiscale data spanning molecular to functional measurements. This dataset integrates diverse modalities, including genomic and molecular profiling, physiologic sensor measurements e-mail: ; Article https://doi.org/10.1038/s41587-026-03121-4 a b LA outflow Reservoir ICU ventilator Centrifugal pump Membrane deoxygenator Pressure monitor PA inflow Leukocyte filter Heater/cooler Gas cylinder for deoxygenation (86% N2, 8% CO2, 6% O2) c Imaging • X-ray radiographs • CNN-derived features • Principal image components • Clinical radiological labels Lung biochemistry Electrolytes Acid–base • Calcium • pH • Chloride • Bicarbonate • Sodium • Base excess • Potassium Transcriptomics • Hypoxia • Inflammatory response • Apoptosis • p53 signaling • TNF signaling • Interleukin 2 signaling • PI3K/AKT/mTOR signaling • Interleukin 6 signaling • TGFβ signaling • Oxidative phosphorylation Lung physiology • Airway pressure • Vascular pressure • Partial pressure O2 and CO2 • Lung compliance • Expiratory volume • Edema Metabolomic biomarkers Cellular metabolism • Glucose • Lactate Multimodal human lung data generated during EVLP Protein biomarkers • Interleukin 6 • Interleukin 8 • Interleukin 1β • Interleukin 10 d Human lungs ex vivo (Physical twin) Data Calibration Validation standardization preprocessing Match lung baseline data Historical database Training Model Digital twins Physics-based models Lung mechanics equations Lung function forecasting ′What if′ analysis Inform Data-driven models XGBoost, GRU Python pyTorch Streamlit Tools and Docker deployment google colab Fig. 1 | EVLP generates organ-specific multimodal data for digital lung model construction. a, A representative image of human lungs on the EVLP system (photograph source: UHN). b, Schematic drawing of the individual circuit components for the EVLP system. c, Examples of the multimodal and organspecific data (light-blue boxes) generated during EVLP, which can be analyzed using ML to create DTs of human lungs (center) (photograph source: UHN). d, Summary of workflows and implementation approach for ex vivo human lung DTs (image source: UHN). Panel b adapted from ref. 7 and created in BioRender; Zhou, E. https://biorender.com/02j2eeo (2026). and medical imaging (Fig. 1c). Thus, DTs of human lungs derived from EVLP biospecimens and data capture both comprehensive lung function and the complex interactions of biological processes. An ex vivo human lung DT leverages organ-centric modeling using data from the physical lung to simulate future biological function without altering EVLP system settings. This approach aligns with the term ‘digital shadow’ used in engineering15 and with the DT concept in healthcare literature16–18. In this study, we adopt terminology from the medical literature, referring to the simulation of lung function at future time points as an ex vivo human lung DT or ‘digital lung’. DT development generally follows one of three approaches: physics-based, data-driven or a hybrid approach that combines both. Physics-based models draw on first-principles equations and mechanistic understanding to provide interpretable, grounded representations of system dynamics, while data-driven models can include artificial intelligence (AI) and machine learning (ML) methods. DTs of physical objects (for example, heat transfer in buildings) often rely on physics-base (...truncated)