Authors: IJHSB Editorial Board

Publish Date: 6.11.2025

Abstract

Immune digital twins are emerging computational representations of an individual’s immune system that integrate multi-scale biological data with mechanistic and data-driven models to simulate immune trajectories in silico. By enabling virtual experiments on personalised models, they offer a framework for exploring vaccine responses, treatment effects and the dynamics of immune-mediated diseases before interventions are tested at the bedside. Immune digital twins build on the broader concept of medical digital twins in personalised medicine and draw on advances in systems immunology, high-throughput immunoprofiling and artificial intelligence. This short review outlines the conceptual basis and construction of immune digital twins, summarises current and proposed applications in infectious disease, oncology and autoimmunity, and highlights key scientific, technical and ethical challenges that must be addressed for these tools to become reliable and clinically useful.

Introduction

Digital twin technology was first developed in engineering, where virtual replicas of physical systems are continuously updated with sensor data to monitor performance, predict failures and optimize interventions [1]. In healthcare, this concept has been extended to medical digital twins: computational models of organs or patients that assimilate longitudinal clinical data to forecast disease trajectories and support treatment planning [1,2,12–14,18]. These models are viewed as central components of future personalized medicine, as they provide a formal framework for mechanistically grounded, data-driven prediction and scenario testing [1–3,12,17,18].

Within this broader landscape, attention has turned to the immune system, which plays a pivotal role in infection, cancer, autoimmunity, transplantation and vaccine responses. A series of recent position papers has introduced the notion of immune digital twins, defined as patient-specific computational models that represent key components of an individual’s immune system and their interactions over time [3–5]. These twins aim to reproduce essential features of immune function and dysregulation in specific clinical contexts, and to predict responses to interventions such as vaccines, immunotherapies or changes in immunosuppressive regimens.

The development of immune digital twins is driven by rapid advances in systems immunology, quantitative systems pharmacology, high-throughput immune profiling and computational modeling [2–5,9–11,16]. At the same time, it is constrained by the inherent complexity of the immune system, limitations of available data, and the need for robust validation and governance frameworks [3–7,10,11,15–18]. This manuscript synthesizes current thinking on immune digital twins, emphasizing conceptual foundations, construction workflows, applications and open challenges.

Concept of Immune Digital Twins

Medical digital twins can be defined as computational counterparts of patients that are dynamically linked to real-world data and used to analyze and predict responses to interventions in a specified context of use [1,2]. Immune digital twins apply this concept to the immune system as the central object of modeling [3–5].

_{Figure 1. Conceptual workflow of an immune digital twin. Diverse data sources, including clinical records, laboratory results, immune profiling assays, multi-omics measurements, imaging and digital health data, are first integrated and curated. These harmonized data feed into a mechanistic immune model that represents key cell populations, mediators and tissues, and into data-driven components used for parameter estimation and feature extraction. The mechanistic and data-driven elements are combined to generate a personalized immune digital twin, which can be used to perform in silico experiments and scenario testing, such as simulating vaccine schedules, immunotherapy regimens or changes in immunosuppression. Outputs from these simulations—predicted immune trajectories, treatment response probabilities or risk estimates—inform research and clinical decision-making. New clinical and experimental data are then used to update the twin, forming a feedback loop in which the virtual and physical systems co-evolve.}

An immune digital twin is typically designed around a particular clinical question. Rather than attempting to simulate the entire immune system in all detail, most proposed frameworks advocate purpose-specific twins whose scope is tailored to a disease area or use case, such as sepsis, respiratory viral infection, autoimmune arthritis or tumor–immune interactions in cancer [3,5,9]. The twin represents a reduced but mechanistically meaningful subset of immune cells, cytokines, tissues and molecular pathways, together with their interactions and dynamics.

The degree of personalization in an immune digital twin can vary. In some implementations, the twin represents a virtual patient class that captures variability within a defined subgroup; in others, it is individualized using detailed omics, immunophenotyping and clinical data from a single person [2–4]. Personalization typically involves adjusting parameters that control baseline immune status, antigen exposure, pharmacokinetics and pharmacodynamics of therapeutics, and other host factors that influence immune behavior.

Immune digital twins usually combine mechanistic and data-driven elements. Mechanistic components—such as systems of differential equations, network models or agent-based models—encode explicit hypotheses about immune processes, including cell differentiation, migration, activation, tolerance and memory [3–5]. Data-driven components—such as machine-learning models—can be used to infer parameters from data, approximate submodels that are too complex to specify analytically, or map patient features to suitable parameter sets [3–5,9]. The resulting hybrid models aim to preserve biological interpretability while leveraging the flexibility of statistical learning.

Crucially, immune digital twins are not static. They are conceived as models that can be updated as new data are acquired and as biological knowledge evolves [3,4]. This dynamic aspect underlies the “twin” metaphor: the virtual immune system is meant to co-evolve with its real-world counterpart, reflecting changes over time rather than providing a one-off snapshot.

Building Immune Digital Twins

The construction of an immune digital twin can be organized into several interlinked stages, spanning data integration, model development, personalization, calibration and validation, and finally deployment in research or clinical contexts [2–5,9–11,16–18].

The process begins with data integration and curation. Relevant data sources include routine clinical measurements, longitudinal laboratory tests, immune phenotyping by flow or mass cytometry, cytokine profiling, multi-omics data (genomics, transcriptomics, proteomics, metabolomics), imaging and, increasingly, digital health data from wearable devices [2,3,6,8,17]. For immune digital twins, these heterogeneous data types must be harmonized and mapped onto immune-relevant variables such as cell subset abundances, receptor repertoires, antigen loads, tissue states and host risk factors. Standardization and adherence to FAIR data principles are essential to ensure that datasets can be shared and reused across projects [4,7,17].

On this data foundation, a mechanistic immune model is selected or constructed. Existing systems immunology models provide a starting point. For example, the Universal Immune System Simulator (UISS) employs agent-based modeling to represent immune cells, soluble mediators and tissues, and has been applied to infectious diseases and vaccine responses [9,11,16]. Other models use systems of ordinary differential equations to describe population-level dynamics of lymphocytes and cytokines in sepsis or autoimmunity, or hybrid frameworks coupling tissue-scale diffusion with cellular decision rules [3,5,16]. The choice of modeling formalism depends on the clinical question, data availability and computational constraints.

Personalization involves adjusting model structure and parameters to reflect the characteristics of a specific patient or patient subgroup. Personalization strategies include direct parameter estimation from individual-level data, hierarchical modeling that shares information across patients, and selection from libraries of pre-computed virtual patients that approximate observed profiles [2–4,9,16]. Machine-learning approaches can assist in mapping high-dimensional patient features to model parameters or initial conditions.

Following construction and personalization, the immune digital twin must be calibrated and validated. Calibration ensures that model outputs reproduce observed data under the conditions for which data are available—such as vaccine-induced antibody kinetics, tumor burden under a given therapy or flare–remission cycles in autoimmune disease [3,9,11]. Validation evaluates predictive performance outside the calibration domain, for example by comparing model predictions with outcomes from independent cohorts, historical trials or experimental systems [10,11,15,16]. For mechanistic models used in regulated contexts, structured credibility assessment plans and verification–validation–uncertainty quantification workflows have been proposed and applied [11,15].

Once a satisfactory level of credibility has been established for a defined context of use, the immune digital twin can be deployed. Deployment may initially occur in research settings, such as in silico hypothesis generation or trial design, and later in clinical decision-support systems integrated with electronic health records [1–3,13,14,18]. In all cases, careful documentation of model assumptions, limitations and performance characteristics remains critical.

Applications in Personalized Medicine and Vaccine Design

Immune digital twins occupy a central position in emerging visions of personalized medicine, where therapies are tailored to the biological and contextual characteristics of each patient [1,2,12,13]. Applications have been proposed and, in some cases, demonstrated across infectious disease, oncology and autoimmunity.

In infectious disease and vaccinology, in silico models of the immune system have been used to explore vaccine-induced immunity and to support dose selection and regimen design. Agent-based simulations implemented in UISS, for example, have been applied to tuberculosis vaccines, including the RUTI® therapeutic vaccine, to predict patterns of artificial immunity and to interpret trial outcomes [9,11]. Similar approaches have been used to evaluate multi-epitope influenza vaccine candidates, where libraries of digital patients were generated and used to predict the distribution of immune responses across virtual cohorts [9]. Reviews of in silico studies in vaccinology emphasize how such models can complement experimental systems, help prioritize candidates and suggest efficient trial designs [10,16]. In principle, immune digital twins derived from these frameworks could be used to explore vaccines and booster strategies for emerging pathogens across diverse host backgrounds prior to large-scale clinical testing [3,5,9–11,16].

In immuno-oncology, mechanistic and hybrid models of tumor–immune interactions have been developed to study responses to immune checkpoint inhibitors, adoptive cell therapies and cancer vaccines [3,5,9]. These models can represent neoantigen presentation, effector and regulatory lymphocyte dynamics, immunosuppressive microenvironments and treatment effects. When linked to patient-specific genomic, transcriptomic and clinical data, they provide a basis for immune digital twins that simulate alternative immunotherapy regimens and predict potential responders, non-responders and toxicity profiles [3,5,9]. Such models may support rational selection of drug combinations, dosing schedules and sequencing strategies in individual patients.

In autoimmune and inflammatory diseases, immune digital twins have been proposed as tools to integrate genetic risk variants, immune phenotypes, environmental triggers and treatment histories in order to understand heterogeneous trajectories and to guide immunomodulatory therapy [3,5,16]. Mechanistic models of immune dysregulation, calibrated and personalized using longitudinal data, could be used to investigate the effect of tapering immunosuppressive drugs, switching biologic agents or introducing lifestyle interventions on a patient’s risk of flare.

Beyond direct clinical applications, immune digital twins can function as research and methodological tools. They provide controlled environments to explore mechanistic hypotheses, perform virtual experiments and prototype adaptive trial designs [4,9–11,16]. They can also serve as educational resources, offering intuitive visualizations of immune dynamics under varying conditions, and as testbeds for integrating organoid and organ-on-chip data with in silico models [8,10]. In the broader health-systems context, they intersect with digital twin–enabled care pathways and data-driven decision-making frameworks in healthcare [13,14,18].

Challenges and Limitations

Despite their promise, immune digital twins face substantial scientific, technical and ethical challenges that currently limit their widespread adoption.

A primary scientific challenge is the multiscale complexity and context-dependence of the immune system. Immune responses emerge from interactions across molecular, cellular, tissue and organismal scales, influenced by age, genetics, microbiota, comorbidities and environmental exposures. Capturing this complexity in a computational model inevitably requires simplification, and different simplifications can produce similar observable outputs, a phenomenon known as equifinality [3–5,16]. This complicates interpretation of model parameters and reduces confidence in extrapolations beyond observed data.

A second challenge involves data limitations and bias. High-quality, longitudinal immune data suitable for model calibration and validation are still relatively scarce, and often originate from small or unrepresentative cohorts [3,6,17]. Structural biases related to geography, ethnicity, sex and gender, and socioeconomic status can propagate into digital twins and lead to unequal performance or miscalibration across patient groups [6,16,17]. Addressing these issues requires deliberate strategies for inclusive data collection, bias-aware modeling and subgroup-specific evaluation.

Third, there are significant validation and regulatory hurdles. Regulatory agencies have begun to recognize the value of mechanistic in silico models in drug development, but clear guidance for evaluating and approving immune digital twins as components of clinical decision-making remains under development [10,11,15]. Credibility assessment frameworks emphasize the need for transparent documentation of model structure, assumptions and limitations, specification of context of use, and systematic verification, validation and uncertainty quantification [11,15]. Implementing these frameworks in practice is resource-intensive and requires close collaboration between modelers, clinicians and regulators.

Fourth, immune digital twins raise ethical, legal and social questions. Because they rely on detailed personal health data, issues of privacy, consent, data ownership and governance are central [2,6,16,17]. Concerns about transparency and explainability also arise: clinicians and patients must be able to understand, at least at a high level, how model outputs are generated and how uncertainties are quantified [6,16]. There is an ongoing debate about accountability when decisions are informed by complex models and digital decision-support tools. Finally, the deployment of digital twins in healthcare systems risks exacerbating existing inequities if access to high-quality modeling and data infrastructures is limited to well-resourced institutions or regions [6,17,18].

These challenges underline that immune digital twins should be understood as evolving, context-dependent tools whose reliability and fairness must be demonstrated rather than assumed.

Current Initiatives and Future Direction

Several international initiatives and consortia are actively advancing the field of immune digital twins and medical digital twins more broadly. The Research Data Alliance Working Group on Building Immune Digital Twins (BIDT) provides a focal point for community-building, articulating shared goals and developing standards for data and model interoperability [4,7]. This working group coordinates case studies, promotes FAIR principles and engages with stakeholders from academia, industry and clinical practice.

Complementary activities are occurring in large-scale digital health and digital twin programs in Europe and elsewhere, which aim to develop validated digital twin infrastructures across multiple organ systems and diseases [1–4,13,18]. These programs highlight the importance of robust data governance, scalable computational infrastructures and integration with clinical workflows. They also stress the need for epidemiological data and population-level modeling to situate individual twins within broader risk landscapes [17,18].

Methodologically, future immune digital twins are likely to take the form of hybrid, modular systems. Such systems would combine mechanistic multi-scale immune models with machine-learning components for parameter estimation and surrogate modeling, incorporate federated and privacy-preserving analytics to leverage distributed clinical datasets, and interface with experimental platforms such as organoids and organ-on-chip devices for cross-validation [4,8,10,11,16]. Advances in complex systems theory and network science are expected to inform strategies for coupling different model layers while maintaining computational tractability and interpretability [4,16].

At the health system level, immune digital twins are increasingly framed as components of learning health systems, in which data from routine care iteratively update models and, in turn, model outputs inform care decisions and research priorities [2,17,18]. Realizing this vision will require long-term investment in data infrastructure, workforce training and multidisciplinary collaboration [1–4,13,14,17,18].

Overall, the trajectory of immune digital twins will be shaped not only by technical progress, but also by institutional, regulatory and societal decisions about how such models are developed, validated and governed.

Conclusion

Immune digital twins represent an emerging and rapidly evolving approach to modeling the human immune system in a patient-specific manner. By integrating mechanistic systems immunology with rich data streams and contemporary computational methods, these twins offer a framework for exploring immune dynamics, testing interventions and supporting personalized decision-making in infectious disease, oncology and autoimmunity [1–5,9–11,16]. Early applications in in silico vaccine design, immuno-oncology and immune-mediated inflammatory diseases illustrate their potential to complement experimental and clinical approaches [3,5,9–11,16].

At the same time, the development and deployment of immune digital twins is constrained by biological complexity, limitations and biases in available data, the demands of validation and regulatory evaluation, and significant ethical, legal and social considerations [3–7,10,11,15–18]. Addressing these challenges will require sustained interdisciplinary collaboration, transparent and rigorous modeling practices, inclusive data strategies and careful governance.

If these conditions can be met, immune digital twins may become important tools in personalized medicine and immunology, contributing to more precise, anticipatory and equitable care.

---

References

1. Silva, A. C., & Vale, N. (2025). Digital twins in personalized medicine: Bridging innovation and clinical reality. Journal of Personalized Medicine, 15(11), 503.
2. Laubenbacher, R., Niarakis, A., Helikar, T., An, G., Malik-Sheriff, R. S., & Pappalardo, F. (2022). Building digital twins of the human immune system: Toward a roadmap. NPJ Digital Medicine, 5, 64.
3. Niarakis, A., Laubenbacher, R., An, G., Shapiro, B., Malik-Sheriff, R. S., Sego, T. J., & Pappalardo, F. (2024). Immune digital twins for complex human pathologies: Applications, limitations, and challenges. NPJ Systems Biology and Applications, 10, 141.
4. Laubenbacher, R., Adler, F., An, G., Castiglione, F., Eubank, S., Fonseca, L. L., Helikar, T., Jett-Tilton, M., Kirschner, D., Macklin, P., Mehrad, B., Moore, B., Pasour, V., Shmulevich, I., Smith, A., Voigt, I., & Yankeelov, T. E. (2024). Forum on immune digital twins: A meeting report. NPJ Systems Biology and Applications, 10, 19.
5. Laubenbacher, R., Adler, F., An, G., Castiglione, F., Fonseca, L. L., Helikar, T., & Pappalardo, F. (2024). Toward mechanistic medical digital twins: Some use cases in immunology. Frontiers in Digital Health, 6, 1349595.
6. Weinberger, N., Hery, D., Mahr, D., Degeling, K., Schmid, M., & Kickbusch, I. (2025). Beyond the gender data gap: Co-creating equitable digital patient twins. Frontiers in Digital Health, 7, 1584415.
7. Research Data Alliance (RDA) Working Group on Building Immune Digital Twins (BIDT). (2024). Case statement and working group outputs. Research Data Alliance.
8. Elfatimi, E., Lekbach, Y., Prakash, S., Karan, S., Dorotta, J. C., Garcia, A., Suoth, K., Carmona, E., Thompson, S., & Fatkhutdinova, A. (2025). Artificial intelligence-, organoid-, and organ-on-chip-powered models to improve pre-clinical animal testing of vaccines and immunotherapeutics: Potential, progress, and challenges. Frontiers in Artificial Intelligence, 8, 1681106.
9. Russo, G., Crispino, E., Maleki, A., Pennisi, M., Brusic, V., & Pappalardo, F. (2023). Beyond the state of the art of reverse vaccinology: Predicting vaccine efficacy with the Universal Immune System Simulator for influenza. BMC Bioinformatics, 24, 231.
10. Saldanha, L., Langel, Ü., & Vale, N. (2023). In silico studies to support vaccine development. Pharmaceutics, 15(2), 654.
11. Pennisi, M., Russo, G., Sgroi, G., Gallo, G., Pappalardo, F., & Motta, S. (2019). Predicting the artificial immunity induced by RUTI® vaccine against tuberculosis using Universal Immune System Simulator (UISS). BMC Bioinformatics, 20, 504.
12. Hamburg, M. A., & Collins, F. S. (2010). The path to personalized medicine. New England Journal of Medicine, 363(4), 301–304.
13. Attaran, S., & Attaran, M. (2024). Advancing healthcare through the integration of digital twins technology: Personalized medicine’s next frontier. Future Internet, 16, 477.
14. Oulefki, A., Amira, A., & Foufou, S. (2025). Digital twins and AI transforming healthcare systems through innovation and data-driven decision making. Health and Technology, 15, 299–321.
15. Musuamba, F. T., Emili, L., Pozzi, A., Sunker, R., Van der Graaf, P. H., & Rostami-Hodjegan, A. (2021). Scientific and regulatory evaluation of mechanistic in silico drug and disease models in drug development: Building model credibility. CPT: Pharmacometrics & Systems Pharmacology, 10(8), 804–825.
16. Pappalardo, F., Russo, G., Pennisi, M., & Motta, S. (2024). Simulating to predict: Computational models in infectious diseases and immunology. In Simulating to predict: Computational models in infectious diseases.
17. Vallée, A. (2025). Digital twins for personalized medicine require epidemiological data and mathematical modeling: Viewpoint. Journal of Medical Internet Research, 27, e72411.
18. Saratkar, S. Y., Rai, A., Nisar, A., & colleagues. (2025). Digital twin for personalized medicine development. Frontiers in Digital Health, 5, 1583466.