NCI Division of Cancer Biology Junior Investigators Meeting (Abstract): Submission #1

Presenter Information
---------------------
First Name: Heber
Middle Initial: {Empty}
Last Name: L. Rocha
Degree(s): Ph.D. in Computational Modeling
Position/Title/Career Status: Assistant Scientist
Organization: Indiana University
Organization Address:
Bloomington

Email: hlimadar@iu.edu

Abstract Information
--------------------
Abstract Category: Consider my abstract for a Methodology/Technology presentation
Abstract Keywords: agent-based modeling, uncertainty quantification, PhysiCell, model calibration, Bayesian inference, tumor microenvironment, computational oncology, open-source software
Abstract Title: UQ-PhysiCell: Uncertainty Quantification and Calibration for Agent-Based Models of Cancer
Abstract:
My research in computational and mathematical oncology centers on multiscale agent-based models (ABMs) of cancer, in which individual cells are autonomous agents governed by rules for proliferation, migration, death, and signaling, coupled to continuum descriptions of oxygen, nutrients, and cytokines. As one of the developers of PhysiCell, an open-source framework for physics-based multicellular simulation, I have built ABMs of tumor growth, immune-tumor interactions, and treatment response that reproduce experimentally observed behaviors. However, the high-dimensional parameter spaces, stochasticity, and computational cost of ABMs pose major challenges for calibration, uncertainty quantification (UQ), and systematic comparison of competing mechanistic hypotheses, capabilities that are essential for predictive, data-grounded modeling. The next stage of my research addresses this gap. I lead development of UQ-PhysiCell, an extensible open-source Python framework that enables uncertainty quantification, calibration, and model selection for PhysiCell models. UQ-PhysiCell manages simulation inputs and outputs (parameters, initial conditions, cell behavior rules) and orchestrates large simulation ensembles with multiple levels of parallelism, integrating directly with established Python libraries for sensitivity analysis, optimization, Bayesian inference, and surrogate modeling. By decoupling model execution from statistical analysis, it lowers the barrier to rigorous, reproducible uncertainty-aware analysis and moves ABMs beyond single best-fit simulations toward interpretable predictions with quantified confidence. In collaborative cancer research projects, I contribute end-to-end modeling support: formalizing mechanistic hypotheses, building and calibrating tumor microenvironment ABMs against imaging and spatial omics data, and quantifying uncertainty in model-based predictions to guide experimental design.