MS06 - MFBM-17

Immune Digital Twins: Mathematical and Computational Foundations (Part 2)

Thursday, July 17 at 10:20am

SMB2025 Follow

Note: this minisymposia has multiple sessions. The other sessions are: Part-1, and Part-3.

Tomas Helikar (University of Nebraska - Lincoln), Juilee Thakar (Juilee_Thakar@URMC.Rochester.edu) - University of Rochester Medical Center James Glazier (jaglazier@gmail.com) - Indiana University

Description:

Immune Digital Twins (IDTs), virtual models of the human immune system, coupled with periodic real-world data, are a growing focus of precision medicine. Designing control interventions to regulate the immune system requires real-time evaluation of high-dimensional parameter spaces of possible interventions in real-time and the ability to continuously recalibrate models of a changing patient. Because of the sparsity and lack of accuracy of the available experimental data feeds, the lack of first-principles models, the intrinsic stochasticity of the underlying biology, and population heterogeneity building full IDTs presents numerous challenges in model design, parameter identification (in complex agent-based models), uncertainty quantification, forecast deployment and data assimilation for model refinement. Another critical aspect of the development of IDTs is the integration of AI/ML methods with mechanistic models in a variety of roles. Due to the rapidly developing field, we would like to propose and request time for a 5-hour mini-symposium that will focus on modeling and mathematical challenges and achievements related to IDTs. We have secured 10 speakers across disciplines and career stages to cover a wide range of topics, including the use of IDTs in the control of sepsis, respiratory infections, and cancer immunotherapies and approaches to multiscale model construction and parameterization, addressing the aforementioned challenges.

Yi Jiang

Georgia State University

"Immunogenic Cell Death: The Key to Unlocking the Potential for Combined Radiation and Immunotherapy"

Immunogenic cell death (ICD) enhances anti-tumor immunity by releasing tumor-associated antigens and activating the anti-tumor immune system response. Here, we develop a mathe- matical model to quantify the role of ICD in optimizing the efficacy of combined radiotherapy (RT) and macrophage-based immunotherapy. Using preclinical murine data targeting the SIRPα-CD47 checkpoint, we show that RT alone induces minimal ICD, whereas disrupting the SIRPα-CD47 axis significantly enhances both phagocytosis and systemic immune activation. Our model predicts an optimal RT dose (6–8 Gy) for maximizing ICD, a dose-dependent abscopal effect, and a hierarchy of treatment efficacy, with SIRPα-knockout macrophages exhibiting the strongest tumoricidal activity. These findings provide a quantitative framework for designing more effective combination therapies, leveraging ICD to enhance immune checkpoint inhibition and radiotherapy synergy.

Josh Loecker

University of Nebraska-Lincoln

"Adaptive Analysis of Mechanistic Models using Large Language Models"

Large language models (LLMs) hold immense potential for revolutionizing biomedical research and personalized medicine, but their application to mechanistic modeling and immune digital twins (IDTs) remains largely unexplored. This work proposes a novel framework integrating LLMs with mechanistic models to address two critical gaps: (1) translating complex model outputs into actionable insights for patients and clinicians, and (2) automating the analysis and interpretation of large-scale mechanistic models. Our framework leverages a comprehensive library of “Action Intents,” enabling LLMs to interact with and manipulate models, perform complex analyses, and generate human-readable explanations. We will develop novel LLM-driven algorithms for tasks such as parameter sensitivity analysis, critical node identification, and emergent behavior prediction. Furthermore, we will establish robust evaluation metrics to assess LLM performance in this domain, encompassing both quantitative measures of accuracy and qualitative assessments of clinical utility. This framework will empower patients with personalized, understandable insights derived from their Personalized Digital Twin, fostering greater autonomy in healthcare decisions. Simultaneously, it will provide researchers with powerful tools to accelerate the analysis and interpretation of complex biological models, ultimately advancing our understanding of the immune system and accelerating the development of novel therapeutic strategies. This innovative approach promises to bridge the gap between complex biological models and their practical application in personalized medicine, paving the way for more effective and patient-centered healthcare.

Reinhard Laubenbacher

University of Florida

"Immune Digital Twins: Foundational Mathematical Challenges"

The digital twin concept has its origins in industry. One industrial equipment manufacturer advertised its digital twin capabilities to its customers as ”No unplanned downtime” for its products. There is a compelling aspirational analog in healthcare: “No unplanned doctor visits.' Of course, the challenges of building digital twins for human patients are incomparably greater than for machinery. Nonetheless, there are now several instances of what might be called digital twins in medicine, and many more ongoing development projects. Aside from our incomplete understanding of human biology, relative sparseness of data characterizing human patients, and logistical difficulties in implementing computational models in healthcare, there are many mathematical and computational problems that need to be solved. Examples include calibration and validation of multiscale, hybrid, stochastic computational models, forecasting algorithms, and optimal control methods. This talk will describe some of these problems and outline a mathematical research program for the field.

Gary An

University of Vermont

"Curing sepsis with the Critical Illness Digital Twin: An example of the benefit of having a NASEM-compliant Digital Twin"

To date there are no pharmacological agents that can substantively and reliably affect the underlying host pathophysiology of sepsis. The effective control of sepsis requires personalized precision medicine, which requires the capabilities provided by digital twins compliant with industrial standards and consistent with the definition put forth in the National Academies of Science, Engineering and Medicine (NASEM) report entitled 'Foundational Research Gaps and Future Directions for Digital Twins' that provides an operational definition for a digital twin and lists specific challenges moving forward for the development of this technology. NASEM defines a digital twin thusly: 'The key elements that comprise a digital twin include (1) modeling and simulation to create a virtual representation of a physical counterpart, and (2) a bidirectional interaction between the virtual and the physical. This bidirectional interaction forms a feedback loop that comprises dynamic data-driven model updating (e.g., sensor fusion, inversion, data assimilation) and optimal decision-making (e.g., control, sensor steering).' Notably, this definition is not met by the vast majority of currently described biomedical “digital twins,” and this insufficiency limits the applicability of non-NASEM compliant digital twins to provide the true personalized precision medicine required to treat complex immune diseases such as sepsis. We present a prototype Critical Illness Digital Twin developed with a workflow that utilizes mechanistic models with machine learning and artificial intelligence for clinically relevant parameter space identification, trajectory personalization, discovery of novel multimodal/adaptive therapeutic control and guidance for sensor/actuator development. The CIDT is based on a previously validated agent-based model of systemic inflammation, and constructed to conform to a mathematical object terms the Model Rule Matrix (MRM). The MRM employs the Maximal Entropy Principle to account for the latent space of 'what is left out' (e.g. Epistemic Uncertainty) in the rule structure of the CIDT. Operating on the CIDT with a workflow that includes genetic algorithms and active learning we identified non-falsifiable configurations of the MRM with respect to two distinct clinical cytokine time-series datasets, one for burns, one for trauma. We further applied deep reinforcement learning to train an artificial intelligence that can cure sepsis arising from a novel pathogen by modulating host cytokines using only currently FDA-approved biologics. Additional future work must include testing with a sufficiently complex large animal model that can recapitulate the heterogeneity seen in clinical sepsis.

#SMB2025 Follow

Annual Meeting for the Society for Mathematical Biology, 2025.