ONCO-01 Contributed Talks

Gustav Lindwall

Max Planck Institute for Evolutionary Biology

"A Mathematical Model for Pseudo-Progression in CAR-T therapy of B-cell Lymphomas"

CAR-T cell therapy, where patient T cells are genetically modified to target CD19-presenting B cells, has transformed the treatment landscape for several types of B-cell lymphomas. However, this therapy often triggers a strong inflammatory response, which can cause the tumor to temporarily swell in the days following CAR-T infusion — a phenomenon known as pseudo-progression. In this work, we present a dynamical model of CAR-T cell therapy that explicitly incorporates the role of pro-inflammatory cytokines in shaping treatment outcomes. Our model reproduces a wide spectrum of clinical trajectories, including complete remission, treatment failure, and transient pseudo-progression. Importantly, the model’s parameters correspond to measurable patient-specific factors, allowing us to explore how individual patient characteristics influence long-term treatment success. The parametrization of the model maps on to measurable patient characteristics, and we discuss how these parameters impact the long term behavior of the model.

---

Rafael Bravo

University of Texas at Austin

"Testing the feasibility of estimating the migration to proliferation rate ratio in glioblastoma from single time-point MRI data"

Introduction: Glioblastoma tumors with a high migration to proliferation ratio (D/k ratio) are more drug resistant, suggesting 1) that estimating D/k ratios can help predict patient responses, and 2) investigating the cellular mechanisms behind D/k ratios can help understand the mechanisms of drug resistance. Here we quantify the identifiability of D/k ratios using synthetic tumor data. Materials and Methods: We used the standard reaction-diffusion model (i.e. logistic growth and diffusion) initialized with a single cell. Using this model, we grew synthetic tumors, saving the density field for calibration once the tumor had filled 10% of the domain. We fixed proliferation (k, 1/day) and used grid search followed by Levenberg-Marquardt optimization to calibrate the diffusion rate (D, mm2/day) and growth time (t, days) that produced density distributions matching the synthetic data as closely as possible. We quantified the ability of the algorithm to accurately and precisely identify D/k ratios in both the presence and absence of Gaussian noise. Results: Our algorithm finds D/k ratios with very high accuracy: 0.95 +/- 0.72% difference from correct D/k with k fixed at 0.01/day. We found that with 5% noise added the ability to accurately recover D/k ratios improved as its magnitude increased: 9.6 +/- 13.38% when D/k = 10-2 mm2 versus 2.7 +/- 3.65% when D/k = 1 mm2 Future Directions: Ongoing work will establish the minimal data requirements to accurately estimate D/k ratios within 10% of the correct values, and then apply the technique to the brain tumor image segmentation (BRATS) dataset. All BRATS patients have a single segmented pre-treatment MRI (N = 103) which we will use to estimate their D/k ratios. A subset of the BRATS patients also has RNA microarray data available (N = 91). We plan to correlate the patients’ D/k ratios with gene set scores derived from their microarray data to identify cellular mechanisms that potentially underly the D/k ratios.

---

Aaron Li

University of Minnesota

"Using a pharmacokinetic ctDNA shedding model to develop a biomarker of tumor response to targeted therapy"

Early prediction of response to therapy or lack thereof is essential for efficient treatment planning. Next-gen sequencing (NGS) has made it possible to non-invasively collect and sequence circulating tumor DNA (ctDNA) from longitudinal plasma samples. While ctDNA data continues to be collected on a wide variety of cancers and treatment types, it is still unclear what ctDNA biomarkers are most indicative of treatment success or failure. We present a pharmacokinetic model of ctDNA shedding under targeted therapy. Using this model to simulate a cohort of virtual patients, we demonstrate the predictive potential of a biomarker based on early ctDNA sampling. We compare the performance of the biomarker to that of a neural network classifier as well as existing ctDNA biomarkers. We show that our biomarker is able to match and exceed the performance of alternatives, both in terms of accuracy and earliness of prediction.

---

Pujan Shrestha

Texas A&M University

"An ODE-SDE Model for Ct-DNA dynamics"

Effective cancer therapies, while continuously improving, are often constrained by lower detection limits of disease. Tumor-immune dynamics in this limit present one of the most pressing knowledge gaps as cancer ultimate escape or elimination are often determined following an intervening period of population equilibrium sustained at low population size. Population dynamics in this small-population limit are affected by intrinsic noise in the tumor-immune interaction, as are estimates of population disease burden by extrinsic noise in acquiring such estimates through associated biomarkers. We present a modeling framework that investigates the interactions between tumor cells and the immune system in the small population regime, focusing on how these interactions influence biomarker levels. The framework combines deterministic elements, which describe tumor growth and immune responses, with stochastic components that capture the inherent variability in biomarker release. We use a system of ordinary differential equations (ODEs) to represent the tumor-immune dynamics between an adaptive immune compartment, immunogenic tumor cells, and evasive tumor cells. The immune system’s role in controlling tumor growth is reflected in the tumor-immune interaction terms. Apoptotic death via tumor-immune interactions and necrotic death via the tumor competition under a shared carrying capacity both contribute to the release of a tumor biomarker. We focus on applying our model to ct-DNA, wherein we frame ct-DNA dynamics using a stochastic differential equation (SDE). This SDE framework accounts for the variability in ct-DNA release due to the dynamic tumor-immune interactions, as well as inherent biological noise, such as DNA degradation and clearance. By coupling the ODE system of equations for tumor-immune dynamics with the SDE for ct-DNA release, we can use the model to study the fluctuations in ct-DNA levels driven by tumor-immune dynamics and exogenous sampling noise.

---