Research

Our research focuses on developing and applying quantitative modeling techniques to decipher tumor growth and treatment response dynamics. In particular, our research is aimed at quantitative personalized radiation oncology. We develop clinically motivated and experimentally calibrated mathematical and computational models that are informable with patient-specific data for personalized treatment recommendations. In close collaboration with experimentalists and clinicians, mathematical models that are parameterized with experimental and clinical data can help estimate patient-specific disease dynamics and predict treatment success on a per patient basis.

This positions us at the forefront of the advent of digital twins and ‘N=1 virtual trials’ that predict personalized optimized treatment protocols. Our team develops frameworks to calibrate, validate and integrate mathematical modeling into pre-clinical experimentation and clinical practice. We have successfully translated predictions of mathematical models into the first-of-its-kind prospective clinical trial to personalize radiation dose fractionation based on individual pre-treatment tumor growth dynamics.

Cancers are complex adaptive dynamic systems, and the underlying mechanisms of such systems are best understood when subjected to perturbation. Therapies such as radiation, chemotherapy or immunotherapy, among others, introduce a substantial perturbation to a tumor, which offers the opportunity to measure response, identify putative vulnerabilities and dynamically adapt treatment as necessary. Digital twins provide a framework that facilitates adaptive therapies. Digital twins aim to mimic the structure, context, and behavior of the physical twin — the cancer patient in oncology. Digital twins are dynamically updated with data from the patient, have predictive capability and inform clinical decision making. This enables the comparison of different treatment options for an individual patient through in silico trials beyond the average outcomes of historic clinical trials.

We are developing and implementing the first-of-their-kind prospective digital twin clinical trials to compare response and outcomes of digital twin-guided personalized, adaptive therapy against current standard of care therapies. The digital twin research brings together patients, clinicians and data scientists to significantly impact cancer care and improve outcomes. Different efforts of the digital twin research program include:

• Implementation of novel pipelines for data abstraction, data curation, data generation and data integration into institutional data warehouses.
• Development of mathematical and computational models, model calibration, model validation and clinical implementation.
• Adaptive clinical trial design and novel statistical methods for digital twin trial analyses.
• Developing safety guidelines and policies around clinical digital twins.

Over 60% of cancer patients receive radiation therapy (RT) as part of their cancer treatment. Despite the major advances in anatomical and geometric personalization of radiation treatment plans, there is not yet personalization of radiation dose and treatment schedules. Our team focuses on optimizing treatments and outcomes for individual patients. This requires accurate prediction of how each patient will respond to any potential radiation protocol, followed by selection of the protocol that maximizes tumor control while minimizing treatment-associated toxicities.

Furthermore, the current paradigm for selecting RT treatment plans does not allow for any changes in the treatment protocol, in terms of total dose or fractionation, once treatment has started. However, cancer is a complex, adaptive system that will change when perturbed by treatment. We are developing novel algorithms for flexible radiation treatment protocols that can anticipate treatment responses and associated toxicities and adaptively guide treatment changes based on mathematical model predictions (via a digital twin framework).

A major barrier in building predictive models for clinical oncology is missing data in patient data. Generative artificial intelligence (AI) provides a novel set of tools to aid imputation of missing data. Patient time series data (radiology measurements over time such as tumor burden, blood counts, patient reported outcome surveys) are sparse, and general generative AI models are not specifically successful in imputing missing data in such time series. Generative AI, however, has been tremendously successful in generating hyper-realistic images and impute missing information into images. We develop novel approaches to translate time series data into pictorial representation to harness the imputation power of generative AI.

Giving the right treatment at the right time to the right patient is the mantra of precision medicine. In current oncology practice, the “right treatment” at the “right time” is derived from clinical trials that aimed to increase median response, median time to progression, median overall survival or median tolerability. Improving treatment for the average patient leaves half of the patients being overtreated and the other half undertreated. There is a dire need to improve therapy for patients who are not average. The problem in oncology is that we currently do not have sufficient biomarkers to predict how an individual patient will respond to the different treatment options. In the era of data science, we have the opportunity — and responsibility — to re-analyze routinely collected clinical data and bring our integrated, interdisciplinary toolkits to identify novel markers that realize value.

One of the most-commonly collected patient-specific data is a tissue biopsy, which is used for pathology to confirm a cancer diagnosis and provide disease staging. The biopsy tissue, however, contains much more information than a binary disease present or not; the tumor and tumor microenvironment architecture provides an invaluable snapshot in time in the evolutionary dynamics of the complex tumor-patient ecosystem. Being able to derive tumor growth and tissue invasion rates from routinely collected biopsy tissue offers the first opportunity to calibrate mathematical models of cancer therapy and predict how the patient will respond to the different available treatment options. We develop novel approaches that combine machine learning, mathematical oncology and astrophysics to learn patient-specific tumor growth and tissue invasion rates from biopsy tissues, and to use these model parameters to predict treatment response and outcome. Then, we can simulate different treatments for each individual patient prior to therapy to identify the therapy with the highest likelihood of success for each individual patient.

We develop and implement standards for mechanistic mathematical models for translation into clinical decision making. Before a mathematical model can make reliable, testable and translational predictions about novel therapeutic doses, treatment protocols or combination therapies, our research suggest six successive steps must be followed:

1. Identify a putative biomarker
2. Develop mechanistic model
3. Calibrate model with existing data
4. Validate model with untrained data
5. Evaluate predictive performance for known treatments
6. Simulate and predict untested treatments