Innovative Methodologies and New Data for Predictive Oncology Model Evaluation (IMPROVE)
As part of the NCI-DOE Collaboration, the IMPROVE project focuses on challenges facing data-driven modeling for predicting cancer drug response. These challenges include finding consistent ways to compare and evaluate models as well as developing standard methods to improve machine learning models through new data.
The IMPROVE project team seeks to develop a generalizable, open comparison and evaluation framework for deep learning models of cancer drug response. IMPROVE’s team is developing well-curated and standardized training and testing data sets. They’re also developing conventions for defining features of tumor-associated data and representations of molecules and drugs.
IMPROVE is co-led by:
- Rick Stevens, Argonne National Laboratory/University of Chicago
- Jeff Hildesheim, National Cancer Institute
- Ryan Weil, NCI’s Frederick National Laboratory for Cancer Research
Working with the extramural cancer research community, IMPROVE has established a Collaborative Core Modeling Group comprised of experts in cancer therapy and deep learning from several organizations. These include Texas Tech University, Pacific Northwest National Laboratory, and the Mayo Clinic.
Contact the IMPROVE team to learn more about IMPROVE or the Collaborative Core Modeling Group.
Aims of the Project
- Develop protocols for comparing deep learning models for cancer drug response and identify model attributes that contribute to prediction performance with the goal of IMPROVING future models
- Develop approaches for specifying drug screening experiments generating a unique cancer drug response data set explicitly aimed at IMPROVING model performance (training and testing)