IMPROVE Bidders Conference - Overview

Watch the Bidder’s Conference

What is IMPROVE?

IMPROVE is a new National Cancer Institute and Department of Energy (NCI/DOE) Collaboration project focused on improving deep learning models to predict the efficacy of cancer treatments. This project is part of the Joint Design of Advanced Computing Solutions for Cancer (JDACS4C) program.

The overall goal of this project is to engage the broader research community to produce a generalizable and extensible framework for comparing and improving AI models of therapeutic response in a subset of cancer model systems. Additional goals include:

  • Evaluating data in relationship to its cost and patient impact relative to its contribution to model performance
  • Evaluating and determining best existing methods and develop new methods for comparing models, such as methods that consider interpretability, learning capacity, generalizability, stability, etc.
  • Understanding effects of different data, data preprocessing methods, and model architectures on models
  • Generating hypotheses to improve tumor subtyping, define new therapeutic targets, elucidate novel mechanisms of action etc., as well as other biological hypotheses derived from the machine learning results

Why is IMPROVE needed?

Considerable progress has been made in the last decade in the formulation and training of AI/machine learning models to predict cancer drug response. Nonetheless, there is no common set of well-documented and well-characterized approaches to model construction, training, and validation. Differences in the choice of normalization, encoding, filtering, and other factors make it difficult to compare new modeling results in the literature with those from previous studies. In addition, there is a lack of well-curated and standardized training and testing datasets—and a lack of broadly accepted featurization for tumors associated data and for representations of drugs. Therefore, challenges remain in recognizing and understanding new innovations in data driven models of drug response. The IMPROVE project has been created to address these challenges.

Aims

IMPROVE has two inter-related aims:

  • Aim 1: IMPROVE Model Comparison: development of semi-automatic protocols for comparing cancer therapeutic response deep learning models and identifying model attributes that contribute to prediction performance with the goal of IMPROVING future models
  • Aim 2: IMPROVE Model Performance: development of protocols for designing drug and therapeutic screening experiments to generate data explicitly aimed at IMPROVING deep learning model performance.
This RFP focuses on Aim 1—IMPROVE Model Comparison. NOTE: A separate RFI has been issued for Aim 2 and can be found here.

To address these aims, the DOE/Argonne National Laboratory (ANL) in conjunction with NCI/Frederick National Laboratory (FNLCR) will provide open implementation of the following resources to systematically compare models and modeling approaches:

  • Comparison protocols
  • Reference models
  • Test and validation data sets
  • Access to large-scale computing resources
  • Mathematics and statistical expertise
  • Software infrastructure

RFP

Download the PDF of the RFP

Intended outcome

  • Award multiple subcontracts to fund extramural research entities to create the Collaborative Core Modeling Group (CCMG). The CCMG will work collaboratively with the IMPROVE teams at ANL and FNLCR to produce and test state-of-the-art deep learning modeling approaches for multiple cancer use cases.

Offer submission deadline: May 9, 2022

Bidders Conference for this RFP

This Bidders Conference is your opportunity to

  • Learn about the IMPROVE project
  • Meet other potential bidders
  • Obtain information on the subcontracting process
  • Receive answers to questions submitted in advance

Participants are encouraged to submit their questions in advance by emailing IMPROVE-AI@nih.gov.

Background

Since 2016, the National Cancer Institute (NCI) and the U.S. Department of Energy (DOE) have engaged in a strategic, interagency collaboration to accelerate advances in precision oncology and scientific computing. The JDACS4C program is the focal point of this strategic, interagency collaboration and includes multidisciplinary scientific research and development teams.

Leidos Biomedical Research, Inc. provides operational and technical support to the Frederick National Laboratory for Cancer Research, a federally funded research and development center (FFRDC). This support includes the execution of projects sponsored by the National Cancer Institute (NCI)-Department of Energy (DOE) Collaboration. As part of this effort, the successful subcontractors shall provide direct support and work products to the IMPROVE activity with the expectation that all work products shall be made freely available to the public in a timely manner. The subcontractors will be expected to work collaboratively with other subcontractors awarded under this project, with DOE/ANL and NCI/FNLCR (the sponsoring organizations) and with subcontractors who will receive future awards for IMPROVE activities.

Questions? Contact IMPROVE-AI@nih.gov