CANcer Distributed Learning Environment (CANDLE)

Extending Deep Learning Capabilities to Accelerate Cancer Research

CANcer Distributed Learning Environment (CANDLE) is an open source, collaboratively developed software platform providing deep learning methodologies. 

CANDLE supports and accelerates the NCI and Department of Energy (DOE) collaborative Joint Design of Advanced Computing Solutions for Cancer (JDACS4C) program, as well as other related areas of cancer research.

Driven by scientific challenges in cancer research that are defined by JDACS4C pilot efforts, CANDLE capabilities enable advances in exascale computing through support from DOE’s Exascale Computing Project (ECP). The project is spearheaded by DOE’s Argonne National Lab with participation across the JDACS4C national laboratories: Oak Ridge National Lab, Lawrence Livermore National Lab, Los Alamos National Lab, and NCI’s Frederick National Lab for Cancer Research. Scientists across the JDACS4C collaboration work in a true team-science manner with combined expertise in cancer and computational, data, and physical sciences in the development of CANDLE.

Features of CANDLE

The CANDLE platform enables the cancer researcher to access key features at the forefront of the data science frontier shaped by advances in deep learning.

  • Hyperparameter optimization efficiently identifies the most effective model implementations.
  • Scalable data parallelism speeds machine learning where very large data are required.

As a scalable system, the CANDLE platform can be installed and run on systems ranging from a laptop to the largest supercomputers available for scientific research.

Open Source

CANDLE leverages the best, emerging open source innovations developed by the deep learning community, innovating, extending, and delivering a scalable platform for deep learning that cancer researchers can access and adapt for use with their own challenges.

CANDLE Benchmarks

The CANDLE benchmarks deliver working examples of large scale deep learning applied to different cancer research challenges. Established to provide a performance baseline from which to measure progress and improvement, the benchmarks implement deep learning architectures that are relevant to scientific challenges in cancer research and opportunities in exascale computing.

Through the deep learning capabilities of CANDLE, demonstrated in the benchmarks, cancer researchers can produce faster results in several areas:

  • Identification of key molecular interactions, based on molecular dynamic simulations of proteins, specifically RAS
  • Predictions of tumor response to drug treatments, based on molecular features of tumor cells and drug descriptors
  • Better characterization of cancer patient trajectories and outcomes using a growing compendium of clinical information

Connecting the Cancer Community

CANDLE helps bring the cancer research community together, with hands-on workshops that offer opportunities for researchers to share insights, and benchmark examples that demonstrate how deep learning and CANDLE can be used broadly to accelerate cancer research.

The Future of CANDLE

The CANDLE project has already delivered software for use by each of the JDACS4C pilots and by other cancer research projects.

Future work is underway for new releases of CANDLE, including:

  • Model optimization joined with parallel computing to increase capabilities for even larger amounts of data, and
  • Development of new areas in which deep learning can accelerate cancer research