“The joint Department of Energy (DOE) and National Cancer Institute (NCI) project—Innovative Methodologies and New Data for Predictive Oncology Model Evaluation (IMPROVE)—aims to address challenges around model comparisons and dataset standardization and availability through the development and provision of a software framework, along with associated computing resources needed, to make it routine practice for the broader community (cancer research and other areas) to compare new machine learning modeling approaches to previous models in a rigorous and comprehensive fashion…”
“This effort will focus on developing a series of ‘experimental campaign proposals’ aimed at elucidating what new experimental data is needed (e.g., types, quality, quantity) to augment existing training data for the purpose of improving machine learning in cancer drug response prediction. Unlike most datasets used by groups building machine learning models for cancer drug response, the goal here is to generate data explicitly targeting improving model performance.”
“In general, data types include but are not limited to DNA and RNA sequencing data including cancer-normal pair data, small nucleotide and copy number variations, proteomics data, surface and secreted markers, imaging data, electronic health records and unstructured data including but not limited to risk factors, previous treatments, stage, grade, and patient outcome.”
“The design of these experimental campaign proposals will involve a group of experts from across the community including machine learning modelers (including the core modeling group), cancer biologists, mathematicians, data scientists, and experimentalists. These groups will be involved through responses to requests-for-information, workshops, and meetings.”
“These experimental campaign proposals are anticipated to underlie a Request for Proposal (RFP) to solicit research involving data generation in furtherance of the IMPROVE project. Possible activities solicited as part of an experimental campaign include:
- Running sensitivity assays in replicate for 1000 of cancer models against 1000 of compounds for which the associated metadata data listed in the introduction can be made available
- The work up and establishment of new cancer model systems based on IMPROVE provided or vendor acquired samples followed by drug sensitivity assays
- The creation of SOPs for and running of novel or investigational assays to obtain additional data types for existing cancer model systems.”
