Machine Learning Day:
Integrating simulations, HPC, and AI for predictive oncology
Machine Learning Day
TimeWednesday, June 24th4:20pm - 4:40pm
DescriptionCancer is a complex multiscale dynamical systems problem with interactions between the tumor and host at the molecular, cellular, tissue, and organism levels. Predictive oncology aims to predict and steer each patient’s future disease dynamics, leveraging advances in data-rich medical measurements; biological theories of cancer progression, therapeutic response and resistance; multiscale modeling and simulation; artificial intelligence (AI) and deep learning techniques; next-generation high-performance computing; and secure high-speed data infrastructures. In this talk, we will discuss progress towards creating digital twins for predictive cancer care: patient-tailored models that can evaluate thousands of potential therapeutic plans, help clinicians understand and choose the plan that best meets the patient’s objectives, benchmark clinical performance, and continuously integrate new data and knowledge to refine treatment plans.
We will show advances in building detailed models of tumor growth and treatment response that run on single compute nodes with GPU acceleration, and extreme-scale exploration of treatment design spaces on HPC. We show how HPC and AI are being combined to accelerate patient-tailored cancer modeling, including (1) improved model calibration, (2) acceleration of individual model components with surrogate models, (3) automated model coarse-graining for faster simulations at larger scales, and (4) AI-based data analysis. We will close with thoughts on next-generation hybrid AI architectures that tightly integrate multiscale simulations, artificial intelligence, and high performance computing for predictive medicine and accelerated biological discovery. This work reflects a joint initiative of the US National Cancer Institute, the Department of Energy, industry, and academia to build digital twins for predictive oncology.