MS08 - ONCO-04

Digital twins for clinical oncology and cancer research (Part 2)

Friday, July 18 at 10:20am

SMB2025 Follow

Note: this minisymposia has multiple sessions. The other session is: and Part-1.

Guillermo Lorenzo (University of A Coruna (Spain)), Chengyue Wu (The University of Texas MD Anderson Cancer Center, US), Ernesto A. B. F. Lima (The University of Texas at Austin, US)

Description:

The overall goal of this minisymposium is to present and discuss recent developments of digital twin technologies to (i) address the personalization of clinical management of cancers, and (ii) advance the research on biophysical mechanisms underlying these pathologies from the micro to the macroscale. A digital twin can be defined as a virtual representation of a physical object by means of a computational model that continuously assimilates object-specific data to enable-decision making about the physical object based on its current and future states. Digital twins have a widespread development in multiple areas of engineering and they have also been proven useful in several areas of medicine, such as surgical planning, cardiovascular disease interventions, and glucose monitoring in diabetic patients. Given the increasing success of computational models to predict the development of cancer and its response to treatments, these models could be employed to construct digital twins to support the optimal diagnosis, monitoring, and treatment of individual patients as well as to assist the research of this disease in vitro, in vivo, and in silico. The talks in this minisymposium will present recent efforts in building digital twins for the clinical management of cancers and the (pre)clinical research of these diseases, as well as ongoing work on practical mathematical models and computational methods for the development of these predictive technologies.

Heber L. Rocha

Indiana University

"Agent-Based Modeling of Cancer Drug Response with PKPD Calibration Challenges and Personalized Modeling"

In this talk, I will present an agent-based model developed in PhysiCell that integrates transmembrane diffusion and pharmacokineticpharmacodynamic (PKPD) processes to simulate drug responses in cancer cell cultures. While the model is designed to handle a range of therapeutic scenarios, I will focus on its application to AU565 breast cancer cells, using a previously published dataset of in vitro area cultures. Before calibration, we perform a local sensitivity analysis to identify key parameters influencing cell behavior and to reduce dimensionality in the inference process. We then apply Approximate Bayesian Computation (ABC) to calibrate the model using experimental data, highlighting the challenges posed by limited temporal resolution and spatial variability in the measurements. I will discuss how these limitations lead to parameter non-identifiability, and share insights into how experimental design and model assumptions interact to shape the reliability of inference. This work not only demonstrates the potential of integrating ABMs with data-driven calibration, but also underscores the need for careful consideration of data quality when advancing toward more personalized modeling frameworks.

Marianna Cerasuolo

University of Sussex

"Mathematical and Statistical Insights into Gut Microbiota–Phytocannabinoid–Diet Interplay in Prostate Cancer Progression in Mice"

This study presents a data-driven framework for modelling host-microbiome-tumour interactions in obesity-associated prostate cancer (PCa). We quantitatively investigated PCa progression in the TRAMP mouse model, focusing on the interplay between dietary fat intake, phytocannabinoid therapy, and gut microbiota composition. Mice on regular or high-fat diets (HFD) were treated with enzalutamide, cannabidiol and cannabigerol, alone or in combination. While combination therapy proved most effective across both dietary groups, a high-fat diet alone was associated with tumour acceleration and significant microbiome dysbiosis. To get further insight into the underlying dynamics, we used statistical analyses to characterise microbial community structure and its modulation by treatment. Using Granger causality and generalised linear models, we inferred predictive relationships among microbial taxa over time. These were complemented by support vector regression and network-theoretic measures to characterise microbial ecosystems under perturbation. Statistical modelling revealed that HFD promotes a more clustered, less modular microbiome with altered predictive relationships between bacterial taxa. Under combination therapy, these patterns were partially reversed, including restoration of Bacteroidetes abundance and damping of pro-tumour microbial signals. Building upon these findings, we developed a dynamical framework to simulate the co-evolution of microbial communities and tumour burden under varied interventions. This model is based on established ecological principles to capture the temporal dynamics of microbial interactions within the gut environment. The framework simulates microbiome-mediated tumour progression and therapeutic responses by integrating host, microbial, and therapeutic variables. Our results support using microbiome-informed digital twins in obesity-associated PCa, highlighting the therapeutic value of integrated, data-driven modelling.

Guillermo Lorenzo

University of A Coruna

"Patient-specific forecasting of prostate cancer progression to higher-risk disease during active surveillance"

Prostate cancer (PCa) usually exhibits low or intermediate risk at diagnosis, for which active surveillance (AS) is an established clinical option. Patients in AS are monitored via serum Prostate Specific Antigen (PSA), multiparametric magnetic resonance imaging (mpMRI), and biopsies. If these exams indicate tumor progression to higher-risk disease, curative treatment is typically recommended (e.g., surgery, radiotherapy). Hence, AS combats overtreatment of indolent PCa, thereby avoiding unnecessary treatment that can induce side effects reducing quality of life (e.g., incontinence, impotence) but without prolonging longevity. However, monitoring protocols for AS rely on an observational and population-based approach that does not account for the heterogeneous nature of PCa dynamics and cannot provide an early identification of progressing patients. To address these two critical limitations, we propose using personalized predictions of tumor progression based on biomechanistic features that describe the heterogeneous disease dynamics for each patient (e.g., tumor cell density, proliferation activity). To this end, we first calculate these patient-specific features from MRI-informed, organ-scale predictions of a biomechanistic model of prostate cancer growth. Then, a generalized logistic classifier is leveraged to map these features to risk groups. Since our PCa growth predictions are spatiotemporally-resolved, we can calculate the biomechanistic features describing PCa dynamics and the tumor risk over time, thus enabling the calculation of time to progression for each patient. Here, we present a preliminary study of our approach in which we demonstrate the accuracy of our personalized growth and progression predictions in a small patient cohort. Although further improvement and testing in larger cohorts are required, we believe that our predictive technology could be leveraged to inform clinical decision-making and personalize AS protocols for PCa patients.

David A. Hormuth, II

The University of Texas at Austin Texas

"Image-based habitat dynamics in patients with head and neck cancer undergoing radiotherapy"

Intratumoral hypoxia in head and neck cancer plays a crucial role in radiotherapy (RT) response. Accurate prediction of the extent of hypoxia could enable personalized RT planning to target resistant lesions. Multiparameteric MRI (mpMRI) methods have been developed that are sensitive to the underlying tumor biology and microenvironment enabling longitudinal characterization of the dynamics of tumor heterogeneity. Integration of mpMRI methods with biology-based mathematical modeling could enable the prediction of treatment outcomes. We developed a RT response model informed from each patient’s (N = 20) own imaging data to forecast their response at week 4 of RT for both primary and nodal tumors in human papillomavirus-associated oropharyngeal head and neck cancer. Dynamic contrast-enhanced MRI and oxygen-enhanced MRI was collected before and during the delivery of RT. These data were then analyzed to derive physiological parameters describing hypoxia status, perfusion, and cellularity allowing for clustering tumor regions into four distinct habitats. The time course of each habitat’s volume was then tracked to assess tumor dynamics. A four-compartment mathematical model was implemented to describe tumor habitat changes and RT response. Model parameters were optimized using cross-validation and tested on unseen data. Predictions for primary tumor volumes showed strong correlation (Pearson correlation coefficient ; PCC > 0.85) and agreement (concordance correlation coefficient; CCC > 0.78) with measured volumes, with an error of only 5.3% in hypoxic volume at week 4. Predictions for nodal tumors exhibited higher error (16.7%), with moderate correlation (PCC > 0.80) and agreement (CCC > 0.66). By leveraging MRI-derived habitat information, the model provides accurate, patient-specific forecasts of tumor response. These findings support the potential of MRI-based modeling in guiding personalized RT, helping to refine treatment strategies for head and neck cancer.

#SMB2025 Follow

Annual Meeting for the Society for Mathematical Biology, 2025.