My research lies at the intersection of mathematical oncology, computational science, and data science. I develop mechanistic, data-informed models to study how tumor heterogeneity, the microenvironment, drug delivery, and treatment response interact across biological scales. My group combines ordinary and partial differential equations, stochastic models, and spatial agent-based models with experimental and clinical data to investigate tumor growth, therapeutic resistance, and the design of more effective treatment strategies.
We use parameter estimation, Bayesian inference, uncertainty quantification, global sensitivity analysis, virtual cohorts, spatial statistics, and ensemble simulations to identify the mechanisms and parameter combinations associated with treatment success or failure. We also develop surrogate-modeling and interpretable machine-learning methods that make computationally intensive models easier to calibrate, analyze, and connect with sparse biological data.
Our goal is to create experimentally grounded computational models that function as virtual laboratories, helping researchers test biological hypotheses, prioritize experiments, predict heterogeneous treatment responses, and identify promising therapeutic strategies.
Please describe one or two of your most interesting projects.
One of my most interesting projects is SMoRe ParS—Surrogate Modeling for Reconstructing Parameter Surfaces—a framework for connecting complex agent-based models with experimental data. Agent-based models can represent spatial organization, cellular heterogeneity, and individual cell behaviors in considerable detail, but their high-dimensional parameter spaces and computational cost make them difficult to calibrate directly.
SMoRe ParS addresses this challenge by using simpler mechanistic surrogate models to capture key outputs of the agent-based system. We reconstruct the relationships between agent-based model inputs, surrogate-model parameters, and observable experimental responses. This allows us to constrain model parameters, quantify uncertainty, and identify the biological mechanisms that most strongly influence tumor behavior.
We have applied SMoRe ParS to models of vascular tumor growth, cancer-cell proliferation, and treatment response. The broader goal is to make detailed multiscale models more computationally tractable, data-driven, and experimentally useful while preserving their mechanistic and spatial insights.
What is the most significant scientific contribution you would like to make?
The most significant contribution I hope to make is to develop trustworthy, data-informed computational models that can become practical decision-support tools for cancer research. I want these models to do more than reproduce observed behavior. They should reveal the mechanisms driving heterogeneous treatment responses, identify which measurements are most informative, and help researchers determine which therapeutic strategies are most promising before costly experiments or clinical studies are undertaken.
A central part of this goal is making complex multiscale and agent-based models easier to calibrate, interpret, and validate. Through methods such as surrogate modeling, uncertainty quantification, sensitivity analysis, and interpretable machine learning, I hope to bridge the gap between detailed mechanistic models and noisy, sparse, multidimensional biological data. Ultimately, I would like this work to contribute to more rational, individualized, and effective cancer treatment strategies.