Grant Number:	3R01CA249096-06S1 Interpret this number
Primary Investigator:	Li, Yi
Organization:	University Of Michigan At Ann Arbor
Project Title:	New Statistical Methods for Modelling Cancer Outcomes
Fiscal Year:	2026

PROJECT SUMMARY/ABSTRACT The parent award, New Statistical Methods for Modeling Cancer Outcomes (R01CA249096), focuses on developing cutting-edge high-dimensional statistical and machine learning methodologies to improve the modeling and understanding of complex cancer-related outcomes. These methods integrate multifactorial clinical, biological, and demographic data, while incorporating advanced tools such as ensemble learning and partially linear survival models. A key challenge encountered during the project is the significant computational burden involved in analyzing large-scale, heterogeneous biomedical datasets, particularly those that involve longitudinal structures, recurrent events, and ultra-high-dimensional covariates such as radiomics and genomics. These analyses demand substantial memory, processing power, and reproducible workflows that local computing environments cannot reliably support. As the biomedical research landscape increasingly shifts toward large, cloud-hosted datasets (e.g., All of Us Research Program, Cancer Research Data Commons), the need for scalable and collaborative computational infrastructure has become critical. However, many statistical methods remain constrained by local resources and lack cloud-native implementation, limiting their utility in real-world, high-throughput research settings. To address this pressing barrier, we propose a competitive supplement to build and deploy a flexible, cloud-native pipeline, CLOUD (Computational Learning Over Ultra-high-dimensional Data), specifically tailored to meet the computational needs of the parent award. The activities of the CLOUD supplement are fourfold. First, it will accelerate training and inference for complex, high-dimensional models using elastic computing resources on NIH-supported platforms. Second, it will improve scalability by preparing our workflows for integration with emerging national-scale datasets such as All of Us, positioning our methods for broad applicability. Third, it will strengthen the reproducibility and transparency of our analyses through the adoption of modern cloud development practices, including containerization (e.g., Docker), workflow orchestration (e.g., Nextflow or Cromwell), and infrastructure-as-code (e.g., Terraform). Fourth, the pipeline will be made accessible and portable, with deployment guides and shared repositories that reduce technical barriers for external researchers interested in adapting these tools to their own data. All these directly align with NIH’s strategic emphasis on cloud-enabled, reproducible biomedical data science and address a major computational bottleneck in the parent award.





None. See parent grant details.