MS04 - MFBM-15

Calibrating and Relating agent based models to spatial data

Tuesday, July 15 at 4:00pm in Salon 17/18

SMB2025 Follow

Sydney Ackermann, Ramanarayanan Kizhuttil, Samrat Sohel Mondal (Wodarz lab) (University of California, San Diego)

Description:

Spatial agent-based models are powerful tools for simulating biological systems (such as cancer, bacterial colonies, and many ecological interactions), and for understanding the principles governing their evolution and dynamics. Spatial structure and aspects such as localized interactions between individuals, competition for space and resources, and migration significantly influence evolutionary outcomes. With advancements in technology generating increasing amounts of spatial data through imaging and sequencing, spatial agent-based models are expected to become even more prominent. Yet, computational methodologies to parameterize these models, to quantify their sensitivity to input parameters, and to validate them against complex spatial data sets are much less developed compared to non-spatial modeling approaches. These aspects, however, are critical for successfully and confidently translating agent-based models into clinically relevant settings. This mini-symposium aims to bring together researchers experienced/involved in developing and calibrating agent-based models, to discuss these issues, and to explore the development of new methodologies to analyze them and relate them to data.

Room assignment: Salon 17/18

David A. Hormuth, II

The University of Texas at Austin Texas

"Leveraging longitudinal experimental data to parameterize mathematical models of tumor growth and response"

There is a rich history of developing both phenomenological and mechanistic mathematical models of tumor growth and response. While these models have proven valuable for generating testable hypotheses and exploring a wide range of biological scenarios, their clinical utility has historically been limited by challenges in personalization. A key limitation is the reliance on model parameters that are difficult or impossible to measure directly in individual subjects. Recent advances in non-invasive quantitative imaging—such as in vitro time-resolved microscopy and in vivo magnetic resonance imaging (MRI)—have opened new opportunities to observe tumor growth and treatment response over time. These imaging modalities provide quantitative measures of tumor characteristics, including cell density and vascular architecture, before, during, and after therapy. Such data enable the calibration of mathematical models on a subject-specific basis, thereby enhancing their predictive power and clinical relevance. We hypothesize that integrating longitudinal imaging data into subject-specific models can allow early prediction of treatment response, in silico simulation of tumor-specific treatment regimens, and ultimately, dynamic optimization or adaptation of therapy for individual patients. In this session, we will provide a concise overview of mathematical modeling approaches to tumor growth, outline image processing pipelines required to extract relevant features from longitudinal imaging, and present strategies for calibrating and personalizing model parameters to individual subjects. This personalized modeling framework has the potential to inform real-time clinical decision-making and support the development of predictive, adaptive cancer therapies.

Katarzyna A. Rejniak

Moffitt Cancer Center

"Using tumor histology to analyze cancer immunotherapies with the agent-based micropharamacology model"

Tumor microenvironment is highly heterogeneous in its cellular, physical, and chemical structure. This includes various tissue architectures, diverse extracellular matrix compositions and fibril alignments, and irregular gradients of metabolites or drugs penetrating the tumor tissue. Therefore, spatially-explicit agent-based models are ideal tools for exploring spatial heterogeneities within the tumor and their role in treatment efficacy. In this talk we will discuss the use of tumor tissue histology images, both fluorescent multiplex and single immunohistochemistry staining, to simulate and analyze tumor tissue metabolic landscape, and, in particular, their hypoxic niches. These histology images with calibrated oxygenation maps form a base for simulations of immunotherapies (CAR-T cells and tumor infiltrating lymphocytes (TILs)) in solid tumors.

David Basanta Gutierrez

Moffitt Cancer Center

"Calibrating a Spatial Agent-Based Model of Multiple Myeloma Using In Vivo Data to Predict Immunotherapy Response"

Spatial agent-based models (ABMs) are powerful tools for understanding cancer progression, as localized interactions and competition for space and resources significantly influence evolutionary dynamics. In Multiple Myeloma (MM), a 'vicious cycle' of tumor-stromal interactions creates a complex, spatially explicit system where malignant cells compete for resources within the bone marrow niche. Successfully translating ABMs of such systems into clinically relevant tools requires robust methodologies to parameterize and validate them against complex spatial data. Our work addresses this challenge by presenting a novel hybrid agent-based model (HCA) of MM, with a specific focus on its calibration using time-series data. Our 2D on-lattice HCA explicitly models the spatial dynamics and localized interactions that govern cell birth, death, and migration for key cell types, including myeloma cells, osteoblasts, osteoclasts, and mesenchymal stem cells. The model is hybrid in nature, coupling these discrete agents to continuous reaction-diffusion fields representing cytokines like RANKL. To overcome the critical challenge of parameterization, we calibrated the model by quantitatively matching its outputs to longitudinal in vivo data from murine models. This validation was performed against multiple spatial and population-level metrics derived from imaging, including the change in bone area/total area (BA/TA), multiple myeloma/marrow area (MM/MA), and the shifting densities and locations of bone-remodeling cells over time. This carefully calibrated spatial model now serves as a validated foundation for translation to a clinically relevant problem: predicting and overcoming resistance to T-cell engaging (TCE) immunotherapy. We are extending the model by integrating T-cell dynamics, including spatially dependent recruitment and exhaustion, to simulate the evolution of resistance. The confidence gained from our rigorous calibration process allows us to use this model to explore advanced computational methodologies, such as employing genetic algorithms to optimize TCE dosing schedules. This work demonstrates a framework for relating spatial ABMs to complex biological data, enhancing their potential as tools for designing adaptive therapies in oncology.

#SMB2025 Follow

Annual Meeting for the Society for Mathematical Biology, 2025.